Despite the flashy title that's the first "sober" analysis from a CEO I read about the technology. While not even really news, it's also worth mentioning that the energy requirements are impossible to fulfill
Also now using ChatGPT intensely since months for all kinds of tasks and having tried Claude etc. None of this is on par with a human. The code snippets are straight out of Stackoverflow...
Take this "sober" analysis with a big pinch of salt.
IBM have totally missed the AI boat, and a large chunk of their revenue comes from selling expensive consultants to clients who do not have the expertise to do IT work themselves - this business model is at a high risk of being disrupted by those clients just using AI agents instead of paying $2-5000/day for a team of 20 barely-qualified new-grads in some far-off country.
IBM have an incentive to try and pour water on the AI fire to try and sustain their business.
Asking because the biggest IT consulting branch of IBM, Global Technology Services (GTS), was spun off into Kyndryl back in 2021[0]. Same goes for some premier software products (including one I consulted for) back in 2019[1]. Anecdotal evidence suggests the consulting part of IBM was already significantly smaller than in the past.
It's worth noting that IBM may view these AI companies as competitors to it's Watson AI tech[2]. It already existed before the GPU crunch and hyperscaler boom - runs on proprietary IBM hardware.
I know people who still work there and are doing consultancy work for clients.
I am a former IBMer myself but my memory is hazy. IIRC there was 2 arms of the consultants - one was the boring day to day stuff, and the other was "innovation services" or something. Maybe the spun out the drudgery GTS and kept the "innovation" service? No idea.
The part that was spun off was "Infrastructure Services" (from the Wiki article.) Outsourcing and operations, not the business consulting organization that provides high level strategy to coding services.
Missed the boat? Have you been living under a rock? Watson AI advertising has been everywhere for years.
It’s not that they aren't in the AI space, it’s that the CEO has a shockingly sober take on it. Probably because they’ve been doing AI for 30+ years combined with the fact they don’t have endless money with nowhere to invest it like Google.
> Missed the boat? […] Watson AI advertising has been everywhere for years.
They were ahead of the game with their original Watson tech, but pretty slow to join and try get up to speed with the current GenAI families of tech.
The meaning of “AI” has shifted to mean “generative AI like what ChatGPT does” in the eyes of most so you need to account for this. When people talk about AI, even though it is a fairly wide field, they are generally referring to a limited subset of it.
> a high risk of being disrupted by those clients just using AI agents instead of paying $2-5000/day for a team of 20 barely-qualified new-grads in some far-off country
Is there any concrete evidence of that risk being high? That doesn't come from people whose job is to sell AI?
> IBM have an incentive to try and pour water on the AI fire to try and sustain their business.
IBM has faced multiple lawsuits over the years. From age discrimination cases to various tactics allegedly used to push employees out, such as requiring them to relocate to states with more employer friendly laws only to terminate them afterward.
IBM is one of the clearest examples of a company that, if given the opportunity to replace human workers with AI, would not hesitate to do so. Assume therefore, the AI does not work for such a purpose...
IBM was ahead of the boat! They had Watson on Jeopardy years ago! /s
I think you make a fair point about the potential disruption for their consulting business but didn't they try to de-risk a bit with the Kyndryl spinout?
I am a senior engineer, I use cursor a lot in my day to day. I find I can code longer and typically faster than without. Is it on par with human? It’s getting pretty darn close to be honest, I am sure the “10x” engineers of the world would disagree but it definitely has surpassed a junior engineer. We all have our anecdotes but I am inclined to believe on average there is net value.
I think surpassed is not the right word because it doesn't create/ideate. However it is incredibly resourceful. Maybe like having a jr engineer to do your bidding without thinking or growing.
Surpassed is probably the wrong word but the intent is more that it can comprehend quite complicated algorithms and patterns and apply them to your problem space. So yea it’s not a human but I don’t think saying subpar to a human is the right comparison either. In many ways it’s much better, I can run N parallel revisions and have the best implementation picked for review. This all happens in seconds.
Yes, this. Creating multiple iterations in parallel allows much more meaningful exploration of the solution space. Create a branch for each framework and try them all, compare them directly in praxis not just in theory. My brother is doing this to great effect as a solopreneur, and having the time of his life.
Largely agree. Anything that is just a multi-file edit, like an interface change, it can do. Maybe not immediately, but you can have it iterate, and it doesn't eat up your attention.
It is without a doubt worth more than the 200 bucks a month I spend on it.
I will go as far as to say it has decent ideas. Vanilla ideas, but it has them. I've actually gotten it to come up with algorithms that I thought were industry secrets. Minor secrets, sure. But things that you don't just come across. I'm in the trading business, so you don't really expect a lot of public information to be in the dataset.
Senior engineer here as well. I would say Opus 4.5 is easily a mid-level engineer. It's a substantial improvement over Sonnet 4.5, which required a lot more hand-holding and interventions.
I'm not sure what you mean with the "code snippets are straight out of Stackoverflow".
That is factually incorrect just by how LLM works. By now there has been so much code ingested from all kinds of sources, including Stackoverflow LLM is able to help generate quite good code in many occasions.
My point being it is extremly useful for super popular languages and many languages where resources are more scarce for developer but because they got the code from who knows where, it can definitely give you many useful ideas.
It's not human, which I'm not sure what is supposed to actually mean. Humans make mistakes, humans make good code. AI does also both. What it definitely needs is a good programmer still on top to know what he is getting and how to improve it.
I find AI (LLM) very useful as a very good code completion and light coder where you know exactly what to do because you did it a thousand times but it's wasteful to be typing it again. Especially a lot of boilerplate code or tests.
It's also useful for agentic use cases because some things you just couldn't do before because there was nothing to understand a human voice/text input and translate that to an actual command.
But that is all far from some AGI and it all costs a lot today an average company to say that this actually provided return on the money but it definitely speeds things up.
An issue with the doom forecasts is most of the hypothetical $8tn hasn't happened yet. Current big tech capex is about $315bn this year, $250bn last against a pre AI level ~$100bn so ~$400bn has been spent so far on AI boom data centers. https://sherwood.news/business/amazon-plans-100-billion-spen...
The future spend is optional - AGI takeoff, you spend loads, not happening not so much.
Say it levels of at $800bn. The world's population is ~8bn so $100 a head so you'd need to be making $10 or $20 per head per year. Quite possibly doable.
Yesterday I was talking to coworkers about AI I mentioned that a friend of mine used ChatGPT to help him move. So a coworker said I have to test this and asked ChatGPT if he could fit a set of the largest Magnepan speakers (the wide folding older room divider style) in his Infinity QX80. The results were hilarious. It had some of the dimensions right but it then decided the QX80 is as wide as a box truck (~8-8.5 feet/2.5 m) and to align the nearly 7 foot long speakers sideways between the wheel wells. It also posted hilariously incomprehensible ASCII diagrams.
Here's what Gemini 3 says (don't know if correct but seems reasonable):
"Can my friend fit a set of the largest Magnepan speakers (the wide folding older room divider style) in his Infinity QX80?"
Yes, your friend can very likely fit them, but they will need to prepare the vehicle specifically for a tight fit.
Based on the "folding room divider" description, you are looking at the vintage Magnepan Tympani series (likely the Tympani I, III, or IV). These are famous for being massive, hinged panels that measure 72 inches (6 feet) tall.
Here is the breakdown of the fit and the strategy for moving them safely:
1. The Math: It fits, but use the "Captain's" space
The Speaker Size: When folded, a Tympani speaker stack is roughly 72" Long x 18" Wide x 4" Thick. You have two of these stacks (one for left, one for right).
The QX80 Cargo Space: With the 2nd and 3rd rows folded flat, the Infinity QX80 offers about 86–87 inches of length from the liftgate to the back of the front seats.
The Verdict: You have roughly 14 inches of clearance lengthwise. They will fit lying flat.
2. Critical "Must-Dos" Before Loading
Even though they fit on paper, these speakers are fragile and awkward. Do not skip these steps:
Remove the Feet: The speakers have metal feet (T-shaped or U-shaped) screwed into the bottom to make them stand up. You must bring a screwdriver/Allen wrench to remove these. If you leave them on, the speakers will not stack flat, they will slide around, and the feet could tear the QX80's leather or carpet.
Bring Moving Blankets (Lots of them): Magnepans are essentially mylar stretched over magnets. If something pokes the grille cloth, it can ruin the diaphragm. Lay a thick blanket on the cargo floor, place the first speaker stack, add a layer of blankets, and then the second stack.
Slide the Front Seats Forward: Just to be safe, move the driver and passenger seats forward a couple of inches before loading. This ensures you don't accidentally crush the top of the speaker frame when closing the liftgate.
3. A Warning on "Folding"
Check the Hinges: These speakers are 30-40 years old. The fabric hinges that connect the panels often rot. When you go to fold them, they might separate completely. This is actually good for moving (smaller pieces are easier to handle), but be prepared for them to come apart.
Do NOT stack anything on top of them: These are heavy (approx. 90-100 lbs per speaker). Do not put boxes or amps on top of the planar surfaces during the drive.
Summary for your friend: Tell him to clear the car completely, bring tools to take the feet off, and treating them like panes of glass. They will slide right in.
I've found Claude's usefulness is highly variable, though somewhat predictable. It can write `jq` filters flawlessly every time, whereas I would normally spend 30 minutes scanning docs because nobody memorizes `jq` syntax. And it can comb through server logs in every pod of my k8s clusters extremely fast. But it often struggles making quality code changes in a large codebase, or writing good documentation that isn't just an English translation of the code it's documenting.
The problem I had that the larger your project gets, the more mistakes Claude makes. I (not a parent commenter) started with a basic CRUD web app and was blown away by how detailed it was, new CSS, good error handling, good selection and use of libraries, it could even write the terminal commands for package management and building. As the project grew to something larger Claude started forgetting that some code already existed in the project and started repeating itself, and worse still when I asked for new features it would pick a copy at random leaving them out of sync with eachother. Moving forward I've been alternating between writing stuff with AI, then rewriting it myself.
> The problem I had that the larger your project gets, the more mistakes Claude makes
I think the reason for this is because these systems get all their coding and design expertise from training, and while there is lots of training data available for small scale software (individual functions, small projects), there is much less for large projects (mostly commercial and private, aside from a few large open source projects).
Designing large software systems, both to meet initial requirements, and to be maintainable and extensible over time, is a different skill than writing small software projects, which is why design of these systems is done by senior developers and systems architects. It's perhaps a bit like the difference between designing a city and designing a single building - there are different considerations and decisions being made. A city is not just a big building, or a collection of buildings, and large software system is not just a large function or collection of functions.
Here's mine fully deployed, https://hackernewsanalyzer.com/. I use it daily and have some users. ~99.7% LLM code. About 1 hour to first working prototype then another 40 hours to get it polished and complete to current state.
41 hours total of prompting, looking at code diffs, reverting, reprompting, and occasional direct code commits. I do review the full code changes nearly every step of the way and often iterate numerous times until I'm satisfied with the resulting code approach.
If you haunt the pull requests of projects you use I bet you'll find there's a new species of PR:
> I'm not an expert in this language or this project but I used AI to add a feature and I think its pretty good. Do you want to use it?
I find myself writing these and bumping into others doing the same thing. It's exciting, projects that were stagnant are getting new attention.
I understand that a maintainer may not want to take responsibility for new features of this sort, but its easier than ever to fork the project and merge them yourself.
I noticed this most recently in https://github.com/andyk/ht/pulls which has two open (one draft) PRs of that sort, plus several closed ones.
Issues that have been stale for years are getting traction, and if you look at the commit messages, it's AI tooling doing the work.
People feel more capable to attempt contributions which they'd otherwise have to wait for a specialist for. We do need to be careful not to overwhelm the specialists with such things, as some of them are of low quality, but on the whole it's a really good thing.
If you're not noticing it, I suggests hanging out in places where people actually share code, rather than here where we often instead brag about unshared code.
> People feel more capable to attempt contributions
That does not mean that they are more capable, and that's the problem.
> We do need to be careful not to overwhelm the specialists with such things, as some of them are of low quality, but on the whole it's a really good thing.
That's not what the specialists who have to deal with this slop say. There have been articles about this discussed here already.
At this point my prior is that all these 300/ns projects are some kind of internal tools, with very narrow scope and many just for a one-off use.
Which is also fine and great and very useful and I am also making those, but it probably does not generalize to projects that require higher quality standards and actual maintenance.
There's a massive incentive not to share them. If I wrote a project using AI I'd be reluctant to publish it at all because of the backlash I've seen people get for it.
People are and always were reluctant to share their own code just the same. There is nothing to be gained, the chances of getting positive reviews from fellow engineers are slim to none. We are a critical and somewhat hypocritical bunch on average.
This is a "scratch an itch" project I initially started to write manually in the past, but never finishing. I then used claude to do it basically on the side while watching the world series http://nixpkgs-pr-explorer.s3-website-us-west-2.amazonaws.co...
Maybe the most depressing part of all this is if people start thinking they would not have been able to do things without the LLM. Of course they would have, it's not like LLMs can do anything that you cannot. Maybe it would have taken more time at least the first time and you would have learned a few things in the process.
I understand the point, and to some degree agree. For myself, I really couldn't (not to say it wouldn't have been possible). I tried many many times over so many years and just didn't have the mental stamina for it, it would never "click" like infra/networking/hardware does etc and I would always end up frustrated.
I have learnt so much in this process, nowhere near as much as someone that wrote every line (which is why I think being a good developer will be a hot commodity) but I have had so much fun and enjoyment, alongside actually seeing tangible stuff get created, at the end of the day, that's what it's all about.
I have a finite amount of time to do things, I already want to do more than I can fit into that time, LLMs help me achieve some of them.
Sure, I can write all of it. But I simply won’t. I have Claude generated Avalonia C# applications and there is no way I would have written the thousands of lines of xaml they needed for the layouts. I would just have done it as a console app with flags.
But reducing friction, eliminating the barrier to entry, is of fundamental importance. It's human psychology; putting running socks next to your bed at night makes it like 95% more likely you'll actually go for a run in the morning.
Claude has taught me so much about how to use jq better. And really, way more efficient ways of using the command line in general. It's great. Ironically, the more I learn the less I want to ask it to do things.
The former tasks are directly from the training material, directly embedded into the model. For the latter task, it needs a context window and intelligence.
It'll be a common paradigm. Some agents support the coding agent discover relevant context for a plan, others will help the agent stay on track and ensure no rules break.
They really should have been supplying at least a week worth of readymade "projects" to every freelance AI promoter out there to demonstrate x9000 AI productivity gains for the skeptics.
Because vibing the air about those gains without any evidence looks too shilly.
Pointing out the where the burden of proof lies is not an ad hominem. Calling it such is in fact a good example of poisoning the well. all the fan girls have to do is post links to code they have vibe coded. some people have even done that in this thread. it's not an unreasonable standard.
It’s not just good for small code bases. In the last six months I’ve built a collaborative word processor with its own editor engine and canvas renderer using Claude, mostly Opus. It’s practically a mini Google Docs, but with better document history and an AI agent built in. I could never have built this in 6 months by myself without Claude Code.
I think if you stick with a project for a while, keep code organized well, and most importantly prioritize having an excellent test suite, you can go very far with these tools. I am still developing this at a high pace every single day using these tools. It’s night and day to me, and I say that as someone who solo founded and was acquired once before, 10 years ago.
I'm just as much of an avid llm code generator fan as you may be but I do wonder about the practicality of spending time making projects anymore.
Why build them if other can just generate them too, where is the value of making so many projects?
If the value is in who can sell it the best to people who can't generate it, isn't it just a matter of time before someone else will generate one and they may become better than you at selling it?
> Why build them if other can just generate them too, where is the value of making so many projects?
No offence to anyone but these generated projects are nothing ground-breaking. As soon as you venture outside the usual CRUD apps where novelty and serious engineering is necessary, the value proposition of LLMs drops considerably.
For example, I'm exploring a novel design for a microkernel, and I have no need for machine generated boilerplate, as most of the hard work is not implementing yet another JSON API boilerplate, but it's thinking very hard with pen and paper about something few have thought before, and even fewer LLMs have been trained on, and have no intelligence to ponder upon the material.
To be fair, even for the most dumb side-projects, like the notes app I wrote for myself, there is still a joy in doing things by hand, because I do not care about shipping early and getting VC money.
The value is that we need a lot more software and now, because building software has gotten so much less time consuming, you can sell software to people that could/would not have paid for it previously at a different price point.
We don’t need more software, we need the right software implemented better. That’s not something LLMs can possibly give us because they’re fucking pachinko machines.
Here’s a hint: Nobody should ever write a CRUD app, because nobody should ever have to write a CRUD app; that’s something that can be generated fully and deterministically (i.e. by a set of locally-executable heuristics, not a goddamn ocean-boiling LLM) from a sufficiently detailed model of the data involved.
In the 1970s you could wire up an OS-level forms library to your database schema and then serve literally thousands of users from a system less powerful than the CPU in modern peripheral or storage controller. And in less RAM too.
People need to take a look at what was done before in order to truly have a proper degree of shame about how things are being done now.
Most CRUD software development is not really about the CRUD part. And for most framework, you can find packages that generate the UI and the glue code that ties it to the database.
When you're doing CRUD, you're spending most of the time with the extra constraints designed by product. It's dealing with the CRUD events, the IAM system, the Notification system,...
> That’s not something LLMs can possibly give us because they’re fucking pachinko machines.
I mostly agree, but I do find them useful for fuzzing out tests and finding issues with implementations. I have moved away from larger architectural sketches using LLMs because over larger time scales I no longer find they actually save time, but I do think they're useful for finding ways to improve correctness and safety in code.
It isn't the exciting and magical thing AI platforms want people to think it is, and it isn't indispensable, but I like having it handy sometimes.
The key is that it still requires an operator who knows something is missing, or that there are still improvements to be made, and how to suss them out. This is far less likely to occur in the hands of people who don't know, in which case I agree that it's essentially a pachinko machine.
I’m with you. Anyone writing in anything higher level than assembly, with anything less than the optimization work done by the demo scene, should feel great same.
Down with force-multiplying abstractions! Down with intermediate languages and CPU agnostic binaries! Down with libraries!
I agree. re: energy and other resource use: the analogy I like is with driving cars: we use cars for transportation knowing the environmental costs so we don’t usually just go on two hour drives for the fun of it, rather we drive to get to work, go shopping. I use Gemini 3 but only in specific high value use cases. When I use commercial models I think a little about the societal costs.
In the USA we have lost the thread here: we don’t maximize the use of small tuned models throughout society and industry, instead we use the pursuit of advanced AI as a distraction to the reality that our economy and competitiveness are failing.
You could have your morning shower 1°C less hot and save enough energy for about 200 prompts (assuming 50 litres per shower). (Or skip the shower altogether and save thousands of prompts.)
I'd rather phrase it as "code is straight out of GitHub, but tailored to match your data structures"
That's at least how I use it. If I know there's a library that can solve the issue, I know an LLM can implement the same thing for me. Often much faster than integrating the library. And hey, now it's my code. Ethical? Probably not. Useful? Sometimes.
If I know there isn't a library available, and I'm not doing the most trivial UI or data processing, well, then it can be very tough to get anything usable out of an LLM.
> While not even really news, it's also worth mentioning that the energy requirements are impossible to fulfill
If you believe this, you must also believe that global warming is unstoppable. OpenAI's energy costs are large compared to the current electricity market, but not so large compared to the current energy market. Environmentalists usually suggest that electrification - converting non-electrical energy to electrical energy - and then making that electrical energy clean - is the solution to global warming. OpenAI's energy needs are something like 10% of the current worldwide electricity market but less than 1% of the current worldwide energy market.
Google recently announced to double AI data center capacity every 6 month. While both unfortunately deal with exponential growth, we are talking about 1% growth CO2 which is bad enough vs 300% effectively per year according to Google
Constraints breed innovation. Humans will continue to innovate and demand for resources will grow. it is fairly well baked into most of civilization. Will that change in the future? Perhaps but it’s not changing now.
Imagine how big pile of trash as the current generation of graphics cards used for LLM training will get outdated. It will crash the hardware market (which is a good news for gamers)
> Krishna also referenced the depreciation of the AI chips inside data centers as another factor: "You've got to use it all in five years because at that point, you've got to throw it away and refill it," he said
This doesn't seem correct to me, or at least is built on several shaky assumptions. One would have to 'refill' your hardware if:
- AI accelerator cards all start dying around the 5 year mark, which is possible given the heat density/cooling needs, but doesn't seem all that likely.
- Technology advances such that only the absolute newest cards can be used to run _any_ model profitably, which only seems likely if we see some pretty radical advances in efficiency. Otherwise, it seems like assuming your hardware is stable after 5 years of burn in, you could continue to run older models on that hardware at only the cost of the floorspace/power. Maybe you need new cards for new models for some reason (maybe a new fp format that only new cards support? some magic amount of ram? etc), but it seems like there may be room for revenue via older/less capable models at a discounted rate.
Isn’t that what Michael Burry is complaining about? That five years is actually too generous when it comes to depreciation of these assets and that companies are being too relaxed with that estimate. The real depreciation is more like 2-3 years for these GPUs that cost tens of thousands of dollars a piece.
That's exactly the thing. It's only about bookkeeping.
The big AI corps keep pushing depreciation for GPUs into the future, no matter how long the hardware is actually useful. Some of them are now at 6 years. But GPUs are advancing fast, and new hardware brings more flops per watt, so there's a strong incentive to switch to the latest chips. Also, they run 24/7 at 100% capacity, so after only 1.5 years, a fair share of the chips is already toast. How much hardware do they have in their books that's actually not useful anymore? Noone knows! Slower depreciation means more profit right now (for those companies that actually make profit, like MS or Meta), but it's just kicking the can down the road. Eventually, all these investments have to get out of the books, and that's where it will eat their profits. In 2024, the big AI corps invested about $1 trillion in AI hardware, next year is expected to be $2 trillion. Only the interest payments for that are crazy. And all of this comes on top of the fact that none of the these companies actually make any profit at all with AI. (Except Nvidia of course) There's just no way this will pan out.
There are three distinct but related topics here, it's not "just about bookkeeping" (though Michael Burry may be specifically pointing to the bookkeeping being misquoted):
1. Financial depreciation - accounting principals typically follow the useful life of the capital asset (simply put, if an airplane typically gets used for 30 years, they'll split the cost of purchasing an airplane across 30 years equally on their books). Getting this right has more to do with how future purchases get financed due to how the bookkeepers show profitability, balance sheets, etc.. Cashflow is ultimately what might create an insolvent company.
2. Useful life - per number 1 above - this is the estimated and actual life of the asset. So if the airplane actually is used over 35 years, not 30, it's actual useful life is 35 years. This is to your point of "some of them are 6 years old". Here is where this is going to get super tricky with GPUs. We (a) don't actually know what the useful life is or is going to be (hence Michael Burry's question) for these GPUs (b) the cost of this is going to get complicated fast. Let's say (I'm making these up) GPU X2000 is 2x the performance of GPU X1000 and your whole data center is full of GPU X1000. Do you replace all of those GPUs to increase throughput?
3. Support & maintenance - this is what actually gets supported by the vendor. There doesn't seem to be any public info about the Nvidia GPUs but typically these are 3-5 years (usually tied to the useful life) and often can be extended. Again, this is going to get super complicated to financially because we don't know what future advancements might happen to performance improvements to GPUs (and therefore would necessitate replacing old ones and therefore creating renewed maintenance contracts).
Typical load management that’s existed for 70 years: when interactive workloads are off-peak, you do batch processing. For OpenAI that’s anything from LLM evaluation of the days’ conversations to user profile updates.
Flops per watt is relevant for a new data center build-out where you're bottlenecked on electricity, but I'm not sure it matters so much for existing data centers. Electricity is such as small percentage of total cost of ownership. The marginal cost of running a 5 year old GPU for 2 more years is small. The husk of a data center is cheap. It's the cooling, power delivery equipment, networking, GPUs etc that costs money, and when you retrofit data centers for the latest and greatest GPUs you have to throw away a lot of good equipment. Makes more sense to build new data centers as long as inference demand doesn't level off.
How different is this from rental car companies changing over their fleets? I don't know, this is a genuine question. The cars cost 3-4x as much and last about 2x as far as I know, and the secondary market is still alive.
> How different is this from rental car companies changing over their fleets?
New generations of GPUs leapfrog in efficiency (performance per watt) and vehicles don't? Cars don't get exponentially better every 2–3 years, meaning the second-hand market is alive and well. Some of us are quite happy driving older cars (two parked outside our home right now, both well over 100,000km driven).
If you have a datacentre with older hardware, and your competitor has the latest hardware, you face the same physical space constraints, same cooling and power bills as they do? Except they are "doing more" than you are...
The traditional framing would be cost per flop. At some point your total costs per flop over the next 5 years will be lower if you throw out the old hardware and replace it with newer more efficient models. With traditional servers that's typically after 3-5 years, with GPUs 2-3 years sounds about right
The major reason companies keep their old GPUs around much longer with now are the supply constraints
The used market is going to be absolutely flooded with millions of old cards. I imagine shipping being the most expensive cost for them. The supply side will be insane.
Think 100 cards but only 1 buyer as a ratio. Profit for ebay sellers will be on "handling", or inflated shipping costs.
I assume NVIDIA and co. already protects themselves in some way, either by the fact of these cards not being very useful after resale, or requiring them to go to the grinder after they expire.
Cards don't "expire". There are alternate strategies to selling cards, but if they don't sell the cards, then there is no transfer of ownership, and therefore NVIDIA is entering some form of leasing model.
If NVIDIA is leasing, then you can't get use those cards as collateral. You can't also write off depreciation. Part of what we're discussing is that terms of credit are being extended too generously, with depreciation in the mix.
The could require some form of contractual arrangement, perhaps volume discounts for cards, if they agree to destroy them at a fixed time. That's very weird though, and I've never heard of such a thing for datacenter gear.
They may protect themselves on the driver side, but someone could still write OSS.
I think it's a bit different because a rental car generates direct revenue that covers its cost. These GPU data centers are being used to train models (which themselves quickly become obsolete) and provide inference at a loss. Nothing in the current chain is profitable except selling the GPUs.
You say this like it's some sort of established fact. My understanding is the exact opposite and that inference is plenty profitable - the reason the companies are perpetually in the red is that they're always heavily investing in the next, larger generation.
I'm not Anthropic's CFO so i can't really prove who's right one way or the other, but I will note that your version relies on everyone involved being really, really stupid.
The current generation of today was the next generation of yesterday.
So, unless the services sold on inference can cover the cost of inference + training AND gain money, they are still operating at loss.
“like it's some sort of established fact” -> “My understanding”?! a.k.a pure speculation. Some of you AI fans really need to read your posts out loud before posting them.
The GPUs going into data centers aren't the kind that can just be reused by putting them into a consumer PC and playing some video games, most don't even have video output ports and put out FPS similar to cheap integrated GPUs.
And the big ones don't even have typical PCIe sockets, they are useless outside of behemoth rackmount servers requiring massive power and cooling capacity that even well-equipped homelabs would have trouble providing!
At that point it isn't a $10k card anymore, it's a $5k card. And possibly not a $5k card for very long in the scenario that the market has been flooded with them.
Ah well yes to a degree that’s possible but at least at the moment you’d still be better off buying a $5k Mac Studio if it’s just inference you’re doing
Why would you do that when you can pay someone else to run the model for you on newer more efficient and more profitable hardware? What makes it profitable for you and not for them?
There’s no marginal tax impact of discarding it or not after 5 years - if it was still net useful to keep it powered, they would keep it. Depreciation doesn’t demand you dispose of or sell the item to see the tax benefit.
No but it tips the scales. If the new hardware is a little more efficient, but perhaps not so much so that you would necessarily replace it, the ability to appreciate the new stuff, but not the old stuff might tip your decision
I thought the same until I calculated that newer hardware consumes a few times less energy and for something running 24x7 that adds up quite a bit (I live in Europe, energy is quite expensive).
So my homelab equipment is just 5 years old and it will get replaced in 2-3 years with something even more power efficient.
Asking coz I just did a quick comparison and it seems to depend but for comparison I have a really old AMD Athlon "e" processor (like literally September 2009 is when it came out according to some quick Google search, tho I probably bought it a few months later than that but still ...) that runs at ~45W TDP. In idle conditions, it typically consumes around 10 to 15 watts (internet wisdom, not kill-a-watt-wisdom).
Some napkin math says it would cost me about 40 years worth of amortization to replace this at my current power rates for this system. So why would I replace it? And even with some EU countries' power rates we seem to be at 5-10 years amortization upon replacement. I've been running this motherboard, CPU + RAM combo for ~15 years now it seems, replacing only the hard drives every ~3 years. And the tower it's in is about 25 years old.
Oh I forgot, I think I had to buy two new CR2032 batteries during those years (CMOS battery).
Now granted, this processor can basically do "nothing" in comparison to a current system I might buy. But I also don't need more for what it does.
Spinning rust and fans are the outliers when it comes to longevity in compute hardware. I’ve had to replace a disk or two in my rack at home, but at the end of the day the CPUs, RAM, NICs, etc. all continue to tick along just fine.
When it comes to enterprise deployments, the lifecycle always revolves around price/performance. Why pay for old gear that sucks up power and runs 30% slower than the new hotness, after all!
But, here we are, hitting limits of transistor density. There’s a reason I still can’t get 13th or 14th gen poweredge boxes for the price I paid for my 12th gen ones years ago.
I think its illustrative to consider the previous computation cycle ala Cryptomining. Which passed through a similar lifecycle with energy and GPU accelerators.
The need for cheap wattage forced the operations to arbitrage the where location for the cheapest/reliable existing supply - there rarely was new buildout as the cost was to be reimbursed by the coins the miningpool recovered.
For the chip situation caused the same apprecaition in GPU cards with periodic offloading of cards to the secondary market (after wear and tear) as newer/faster/more efficient cards came out until custom ASICs took over the heavy lifting, causing the GPU card market to pivot.
Similarly in the short to moedium term the uptick of custo ASICs like with Google TPU will definately make a dent in bot cpex/opex and potentially also lead to a market with used GPUs as ASICs dominate.
So for GPUs i can certainly see the 5 year horizon making a impact in investment decisions as ASICs proliferate.
> They still work fine but power costs make them uneconomical compared to latest tech.
That's not necessarily the driving financial decision, in fact I'd argue company's with data center hardware purchases barely look at this number. It's more simple than that - their support runs out and its cheaper to buy a new piece of hardware (that IS more efficient) because the hardware vendors make extended support inordinately expensive.
Put yourselves in the shoes of a sales person at Dell selling enterprise server hardware and you'll see why this model makes sense.
It’s far more extreme: old servers are still okay on I/O, and memory latency, etc. won’t change that dramatically so you can still find productive uses for them. AI workloads are hyper-focused on a single type of work and, unlike most regular servers, a limiting factor in direct competition with other companies.
I mean you could use training GPUs for inference right? That would be use case number 1 for a 8 * a100 box in a couple of years. It can also be used for non IO limited things like folding proteins or other 'scientific' use cases. Push comes to shove im sure an old A100 will run crysis.
All those use cases would probably use up 1% of the current AI infrastructure, let alone ahat they're planning to build.
Yeah, just like gas, possible uses will expand if AI crashes out, but:
* will these uses cover, say, 60% of all this infra?
* will these uses scale up to use that 60% within the next 5-7 years, while that hardware is still relevant and fully functional?
Also, we still have railroad tracks from the 1800s rail mania that were never truly used to capacity and dot com boom dark fiber that's also never been used fully, even with the internet growing 100x since. And tracks and fiber don't degrade as quickly as server hardware and especially GPUs.
LambdaLabs is still making money off their Tesla V100s, A100s, and A6000s. The older ones are cheap enough to run some models and very cheap, so if that's all you need, that's what you'll pick.
The V100 was released in 2017, A6000 in 2020, A100 in 2021.
Even if the power is free you still need a grid connection to move it to where you need it, and, guess what, the US grid is bursting at the seams. This is not just due to data center demand; it was struggling to cope with the transition away from coal well before that point.
You also can’t buy a gas turbine for love nor money at the moment, and they’re not ever going to be free.
If you plonked massive amounts of solar panels and batteries in the Nevada desert, that’s becoming cheap but it ain’t free, particularly as you’ll still need gas backup for a string of cloudy days.
If you think SMRs are going to be cheap I have a bridge to sell you, you’re also not going to build them right next to your data centre because the NRC won’t let you.
So that leaves fusion or geothermal. Geothermal is not presently “very cheap” and fusion power has not been demonstrated to work at any price.
Power consumption is only part of the equation. More efficient chips => less heat => lower cooling costs and/or higher compute density in the same space.
In-house hyperscaler stuff gets shredded, after every single piece of flash storage gets first drilled through and every hard drive gets bent by a hydraulic press. Then it goes into the usual e-waste recycling stream (ie. gets sent to poor countries where precious metals get extracted by people with a halved life expectancy).
Off-the-shelf enterprise gear has a chance to get a second life through remarketing channels, but much of it also gets shredded due to dumb corporate policies. There are stories of some companies refusing to offload a massive decom onto the second hand market as it would actually cause a crash. :)
Similar to corporate laptops where due to stupid policies, for most BigCos you can't really buy or otherwise get a used laptop, even as the former corporate used of said laptop.
I used (relatively) ancient servers (5-10 years in age) because their performance is completely adequate; they just use slightly more power. As a plus it's easy to buy spare parts, and they run on DDR3, so I'm not paying the current "RAM tax". I generally get such a server, max out its RAM, max out its CPUs and put it to work.
Same, the bang for buck on a 5yo server is insane. I got an old Dell a year ago (to replace our 15yo one that finally died) and it was $1200 AUD for a maxed out recently-retired server with 72TB of hard drives and something like 292GB of RAM.
The idle wattage per module has shrunk from 2.5-3W down to 1-1.2 between DDR3 & DDR5. Assuming a 1.3W difference (so 10.4W for 8760 hours), a DDR3 machine with 8 sticks would increase your yearly power consumption by almost 1% (assuming avg 10,500kWh/yr household)
That's only a couple dollars in most cases but the gap is only larger in every other instance. When I upgraded from Zen 2 to Zen 3 it was able to complete the same workload just as fast with half as many cores while pulling over 100W less. Sustained 100% utilization barely even heats a room effectively anymore!
A similar one I just ran into: my Framework Desktop was idling @ 5W more than other reported numbers. Issue turned out to be the 10 year old ATX PSU I was using.
The one thing to be careful with Zen 2 onwards is that if your server is going to be idling most of the time then the majority of your power usage comes from the IO die. Quite a few times you'd be better off with the "less efficient" Intel chips because they save 10-20 Watts when doing nothing.
Maybe? The price difference on newer hardware can buy a lot of electricity, and if you aren't running stuff at 100% all the time the calculation changes again. Idle power draw on a brand new server isn't significantly different from one that's 5 years old.
To be clear, this server is very lightly loaded, it's just running our internal network services (file server, VPN/DNS, various web apps, SVN etc.) so it's not like we're flogging a room full of GeForce 1080Ti cards instead of buying a new 4090Ti or whatever. Also it's at work so it doesn't impact the home power bill. :D
Manipulating this for creative accounting seems to be the root of Michael Burry’s argument, although I’m not fluent enough in his figures to map here. But, commenting that it interesting to see IBM argue a similar case (somewhat), or comments ITT hitting the same known facts, in light of Nvidia’s counterpoints to him.
with Michael Lewis, about 30 mins long. Highlights - he thinks we are near the top, his puts are for two years time. If you go long he suggests healthcare stocks. He's been long gold some years, thinks bitcoin is dumb. Thinks this is dotcom bubble #2 except instead of pro investors it's mostly index funds this time. Most recent headlines about him have been bad reporting.
Eh, not exactly. If you don't run CPU at 70%+ the rest of the machine isn't that much more inefficient that model generation or two behind.
It used to be that new server could use half power of the old one at idle but vendors figured out that servers also need proper power management a while ago and it is much better.
Last few gens increase could be summed up to "low % increase in efficiency, with TDP, memory channels and core count increase".
So for loads not CPU bound the savings on newer gen aren't nearly worth it to replace it, and for bulk storage the CPU power usage is even smaller part
Definitely single thread performance and storage are the main reasons not to use an old server. A 6 year old server didn't have nvme drives, so SATA SSD at best. That's a major slow down if disk is important.
Aside from that there's no reason to not use a dual socket server from 5 years ago instead of a single socket server of today. Power and reliability maybe not as good.
NVMe is just a different form factor for what's essentially a PCIe connection, and adapters are widely available to bridge these formats. Surely old servers will still support PCIe?
But if your competitor is running newer chips that consume less power per operation, aren't you forced to upgrade as well and dispose of the old hardware?
Sure, assuming the power cost reduction or capability increase justifies the expenditure. It's not clear that that will be the case. That's one of the shaky assumptions I'm referring to. It may be that the 2030 nvidia accelerators will save you $2000 in electricity per month per rack, and you can upgrade the whole rack for the low, low price of $800,000! That may not be worth it at all. If it saves you $200k/per rack or unlocks some additional capability that a 2025 accelerator is incapable of and customers are willing to pay for, then that's a different story. There are a ton of assumptions in these scenarios, and his logic doesn't seem to justify the confidence level.
> Sure, assuming the power cost reduction or capability increase justifies the expenditure. It's not clear that that will be the case.
Share price is a bigger consideration than any +/- differences[1] between expenditure vs productivity delta. GAAP allows some flexibility in how servers are depreciated, so depending on what the company wants to signal to shareholders (investing in infra for futur returns vs curtailing costs), it may make sense to shorten or lengthen depreciation time regardless of the actual TCOO keep/refresh cost comparisons.
1. Hypothetical scenario: a hardware refresh costs $80B, actual performance increase is only worth $8B, but the share price increases the value of org's holding of its own shares by $150B. As a CEO/CFO, which action would you recommend- without even considering your own bonus that's implicitly or explicitly tied to share price performance.
Illustration numbers: AI demand premium = $150 hardware with $50 electricity. Normal demand = $50 hardware with $50 electricity. This is Nvidia margins @75% instead of 40%. CAPEX/OPEX is 70%/20% hardware/power instead of customary 50%/40%.
If bubble crashes, i.e. AI demand premium evaporates, we're back at $50 hardware and $50 electricity. Likely $50 hardware and $25 electricity if hardware improves. Nvdia back to 30-40% margins, operators on old hardware stuck with stranded assets.
The key thing to understand is current racks are sold at grossly inflated premiums right now, scarcity pricing/tax. If the current AI economic model doesn't work then fundmentally that premium goes away and subsequent build outs are going to be costplus/commodity pricing = capex discounted by non trivial amounts. Any breakthroughs in hardware, i.e. TPU compute efficiency would stack opex (power) savings. Maybe by year 8, first gen of data centers are still depreciated to $80 hardware + $50 power vs new center @ $50 hardware + $25 power. That old data center is a massive write-down because it will generate less revenue than it costs to amoritize.
A typical data centre is $2,500 per year per kW load (including overhead, hvac and so on).
If it costs $800,000 to replace the whole rack, then that would pay off in a year if it reduces 320 kW of consumption. Back when we ran servers, we wouldn't assume 100% utilisation but AI workloads do do that; normal server loads would be 10kW per rack and AI is closer to 100. So yeah, it's not hard to imagine power savings of 3.2 racks being worth it.
Thanks for the numbers! Isn't it more likely that the amount of power/heat generated per rack will stay constant over each upgrade cycle, and the upgrade simply unlocks a higher amount of service revenue per rack?
Not in the last few years. CPUs went from ~200W TDP to 500W.
And they went from zero to multiple GPUs per server. Tho we might hit "the chips can't be bigger and the cooling can't get much better" point there.
The usage would be similar if it was say a rack filled with servers full of bulk storage (hard drives generally keep the power usage similar while growing storage).
But CPU/GPU wise, it's just bigger chips/more chiplets, more power.
I'd imagine any flattening might be purely because "we have DC now, re-building cooling for next gen doesn't make sense so we will just build servers with similar power usage as previously", but given how fast AI pushed the development it might not happen for a while.
> Isn't it more likely that the amount of power/heat generated per rack will stay constant over each upgrade cycle,
Power density seems to grow each cycle. But eventually your DC hits power capacity limits, and you have to leave racks empty because there's no power budget.
Or they could charge the same as you and make more money per customer. If they already have as many customers as they can handle doing that may be better than buying hardware to support a larger number of customers.
Given power and price constraints, it's not that you cannot run them in 5 years time it's that you don't want to run them in 5 years time and neither will anyone else that doesn't have free power.
It’s not about assumptions on the hardware. It’s about the current demands for computation and expected growth of business needs. Since we have a couple years to measure against it should be extremely straightforward to predict. As such I have no reason to doubt the stated projections.
Networking gear was famously overbought. Enterprise hardware is tricky as there isn’t much of a resale market for this gear once all is said and done.
The only valid use case for all of this compute which could reasonably replace ai is btc mining. I’m uncertain if the increased mining capacity would harm the market or not.
BTC mining on GPUs haven't been profitable for a long time, it's mostly ASICs, GPUs can be used for some other altcoins which makes the potential market for used previous generation GPUs even smaller.
That assumes you can add compute in a vacuum. If your altcoin receives 10x compute then it becomes 10x more expensive to mine.
That only scales if the coin goes up in value due to the extra "interest". Which isn't impossible but there's a limit, and it's more often to happen to smaller coins.
Failure rates also go up. For AI inference it’s probably not too bad in most cases, just take the node offline and re-schedule the jobs to other nodes.
There is the opportunity cost of using a whole datacenter to house ancient chips, even if they're still running. You're thinking like a personal use chip which you can run as long as it is non-defective. But for datacenters it doesn't make sense to use the same chips for more than a few years and I think 5 years is already stretching their real shelf life.
Do not forget that we're talking about supercomputers. Their interconnect makes machines not easily fungible, so even a low reduction in availability can have dramatic effects.
Also, after the end of the product life, replacement parts may no longer be available.
You need to get pretty creative with repair & refurbishment processes to counter these risks.
5 years is maybe referring to the accounting schedule for depreciation on computer hardware, not the actual useful lifetime of the hardware.
It's a little weird to phrase it like that though because you're right it doesn't mean you have to throw it out. Idk if this is some reflection of how IBM handles finance stuff or what. Certainly not all companies throw out hardware the minute they can't claim depreciation on it. But I don't know the numbers.
Anyways, 5 years is an infection point on numbers. Before 5 years you get depreciation to offset some cost of running. After 5 years, you do not, so the math does change.
that is how the investments are costed though, so makes sense when we're talking return on investment, so you can compare with alternatives under the same evaluation criteria.
Historically, GPUs have improved in efficiency fast enough that people retired their hardware in way less than 5 years.
Also, historically the top of the line fabs were focused on CPUs, not GPUs. That has not been true for a generation, so it's not really clear if the depreciation speed will be maintained.
> Historically, GPUs have improved in efficiency fast enough that people retired their hardware in way less than 5 years.
This was a time when chip transistor cost was decreasing rapidly. A few years earlier even RAM cost was decreasing quickly. But these times are over now. For example, the PlayStation 5 (where the GPU is the main cost), which launched five years ago, even increased in price! This is historically unprecedented.
Most price/performance progress is nowadays made via better GPU architecture instead, but these architectures are already pretty mature, so the room for improvement is limited.
Given that the price per transistor (which TSMC & Co are charging) has decreased ever more slowly in recent years, I assume it will eventually come almost to a halt.
By the way, this is strictly speaking compatible with Moore's law, as it is only about transistors per chip area, not price. Of course the price per chip area was historically approximately constant, which meant exponentially increasing transistor density implied exponentially decreasing transistor price.
It's worse than that in reality, AI chips are on a two year cadence for backwards compatibility (NVIDIA can basically guarantee it, and you probably won't be able to pay real AI devs enough to stick around to make hardware work arounds). So their accounting is optimistic.
5 years is normal-ish depreciation time frame. I know they are gaming GPUs, but the RTX 3090 came out ~ 4.5 years before the RTX 5090. The 5090 has double the performance and 1/3 more memory. The 3090 is still a useful card even after 5 years.
When you operate big data centers it makes sense to refresh your hardware every 5 years or so because that’s the point at which the refreshed hardware is enough better to be worth the effort and expense.
You don’t HAVE to, but its more cost effective if you do.
(Source, used to operate big data centers)
> In an October letter to the White House's Office of Science and Technology Policy, OpenAI CEO Sam Altman recommended that the US add 100 gigawatts in energy capacity every year.
> Krishna also referenced the depreciation of the AI chips inside data centers as another factor: "You've got to use it all in five years because at that point, you've got to throw it away and refill it," he said.
And people think the climate concerns of AI are overblown. Currently US has ~1300 GW of energy capacity. That's a huge increase each year.
The largest plant in the world is the Three Gorges Dam in China at 22GW and it’s off the scales huge. We’re not building the equivalent of four of those every year.
Unless the plan is to power it off Sam Altman’s hot air. That could work. :)
I am a huge proponent of renewables, but you cannot compare their capacity in GW with other energy sources because their output is variable and not always maximal. To realistically get 100GW in solar you would need at least 500GW of panels.
On the other hand, I think we will not actually need 100GW of new installations because capacity can be acquired by reducing current usage by making it more efficient. The term negawatt comes to mind. A lot of people are still in the stone age when it comes to this even though it was demonstrated quite effectively by reduced gas use in the US after the oil crisis in the 70s. Which basically recovered to the pre crisis levels only recently.
High gas prices caused people to use less and favor efficiency. The same thing will happen with electricity and we'll get more capacity. Let the market work.
> As of 2025, The Medog Dam, currently under construction on the Yarlung Tsangpo river in Mêdog County, China, expected to be completed by 2033, is planned to have a capacity of 60 GW, three times that of the Three Gorges Dam.[3]
Authoritarianism has its draw backs obviously but one of its more efficient points is it can get things done if the will is at the top. Since China doesnt have a large domestic oil supply like the US it is a state security issue to get off oil as fast as possible.
LOL, maybe Sam Altman can fund those power plants. Let me guess: He'd rather the public pay for it, and for him to benefit/profit from the increased capacity.
Big tech is going to have to fund the plants and probably transmission. Because the energy utilities have a decades long planning horizon for investments.
Good discussion about this in recent Odd Lots podcast.
I've read a bunch of the opposite: a lot of secret deals between tech companies and utilities, where when details come out, we find that regular ratepayers are going to be paying a decent chunk of the cost.
Scam Altman wants the US to build a lot of energy plants so that the country will pay the costs and OpenAI will have the profits of using this cheap energy.
I would add an addendum to this -- there is no way the announced spending on AI data centers will all come to fruition. I have no doubt that there will be a massive build-out of infrastructure, but it can't reach the levels that have been announced. The power requirements alone will stop that from happening.
The power requirement is only an issue in western countries, where utilities build at most a double digit buffer, and are used to overall energy use leveling due to efficiency improvements. Now look at China where they routinely maintain a 100% buffer. Demand can double and they can supply that without new generation capacity.
I think you're spot on. OpenAI alone has committed to spending $1.4T on various hardware/DCs. They have nowhere near that amount of money and when pushed Altman gets defensive.
The incentive for CEOs is announcing the plan to do something, they have no idea if they will actually be able to do it, and it probably won't matter.
This happened in the dotcom too btw. Companies built out fibre networks, it wasn't possible to actually build all the physical infra that companies wanted to build so many announced plans that never happened and then, towards the end, companies began aggressively acquiring stakes in companies who were building stuff to get financial exposure (an example was BT, which turned itself briefly into a hedge fund with a telephone network attached...before it imploded).
CEOs do not operate on the timescale of waiting and building. Their timescale is this year's bonus/share options package. Nothing else matters: announce plans to do X or Y, doesn't matter, they know they will be gone long before it happens.
I think this was definitely true of CEOs in the past but Google and Meta have managed spectacular infrastructure buildouts over many decades with proper capacity planning.
There were a lot of lessons learned in the dotcom boom (mainly related to the great telecom heist if you ask me). If you look back on why there was dotcom bubble it's broadly, imho, related to the terrible network quality in the US compared to other first world countries.
Our cellular services for example lag behind basically every asian market by at least 1 maybe 2 generations now.
It took about 10 months to get 200k GPUs going in xAI data centers. Given the stuff has been being announced for the past 2 years, we shall see how they’re going within the next year.
I believe there is a difference between what people say publicly and what they are actually committed to doing on the ground. When all is said and done, I'll be more interested to know what was actually spent.
For example, XYZ AI company may say they are going to spend $1T for AI data centers over the next 5 years.
In actuality, I suspect it is likely that they have committed to something like $5-10B in shovel-ready projects with stretch goals for the rest. And the remaining spend would be heavily conditioned -- is power available? are chips available? is the public support present? financing? etc...
Not to mention, it's a much bigger moat if you can claim you're going to spend $1T. Who else will want to compete with you when you're spending $1T. After the dust has settled and you've managed to be one of the 2-3 dominant AI players, who is going to care that you "only" spent $100B instead of $1T. Look -- you were very capital efficient!
So, do I see it as possible that XYZ AI company could spend $1T, sure. Is it likely? No.
Hmm... "CEOs and teams" don't necessary do what's makes sense mathematically. Many, if not most of them, do whatever that sounds good to shareholders in their quarterly earnings call and ignore the reality or long term development.
If "CEOs and teams" are smart enough, they would not have overhired during 2021-2022 and then do layoffs. Who would be dumb enough to do that?
me very smart. me go college get fancy paper. me move money out of ai stocks. flat rocks not think so well, use too much fire. me trade for shiny yellow rock instead.
If AI is a highlander market, then the survivor will be able to eventually aquire all those assets on the cheap from the failing competitors that flush their debt in bankruptcy.
Meanwhile, highlander hopefuls are spending other peoples money to compete. Some of them with dreams of not just building a tech empire, but to truly own the machine that will rule the world in every aspect.
Investors are keen on backing the winner. They just do not know yet who it will be.
Until China sees it valuable to fund open weights SOTA-ish models, even the winner might struggle. There is very little capture - protocols are mostly standard so models are mostly interchangeable and if you are trying to raise prices enough to break even on the whole operation, somebody else can probably profitably run inference cheaper.
"oh look, with open weights now everyone can run this quadrillion param 'superintelligent' model. but what's that? only we have the power capacity and cheap $/W to actually serve it? funny thing...!"
I interpreted it as "there can only be one" which I believe is a quote from the Highlander movie; it's a "winner takes all" and that winner gets the title of "highlander."
In this situation then everyone who _isn't_ the winner will go broke -> sell off all their stuff on the cheap because they're desperate -> the winner gets all their hardware for a great deal and becomes even more powerful.
He's right to question the economics. The AI infrastructure buildout resembles the dot-com era's excess fiber deployment - valuable long-term, but many individual bets will fail spectacularly. Utilization rates and actual revenue models matter more than GPU count.
100% of technical innovations have had the same pattern. The same thing happens every time because this is the only way the system can work: excess is required because there is some uncertainty, lots of companies are designing strategies to fill this gap, and if this gap didn't exist then there would be no investment (as happens in Europe).
Also, demand wasn't over-estimated in the 2000s. This is all ex-post reasoning you use data from 2002 to say...well, this ended up being wrong. Companies were perfectly aware that no-one was using this stuff...do you think that telecoms companies in all these countries just had no idea who was using their products? This is the kind of thing you see journalists write after the event to attribute some kind of rationality and meaning, it isn't that complicated.
There was uncertainty about how things would shake out, if companies ended up not participating then CEOs would lose their job and someone else would do it. Telecoms companies who missed out on the boom bought shares in other telecom's companies because there was no other way to stay ahead of the news and announce that they were doing things.
This financial cycle also worked in reverse twenty years later too: in some countries, telecoms companies were so scarred that they refused to participate in building out fibre networks so lost share and then ended up doing more irrational things. Again, there was uncertainty here: incumbents couldn't raise from shareholders who they bankrupted in fiber 15 years ago, they were 100% aware that demand was outstripping supply, and this created opportunities for competitors. Rationality and logic run up against the hard constraints of needing to maintain a dividend yield and the exec's share options packages.
Humans do not change, markets do not change, it is the same every time. What people are really interested in is the timing but no-one knows that either (again, that is why the massive cycle of irrationality happens)...but that won't change the outcome. There is no calculation you can make to know more, particularly as in the short-term companies are able to control their financial results. It will end the same way it ended every time before, who knows when but it always ends the same way...humans are still human.
> Also, demand wasn't over-estimated in the 2000s. This is all ex-post reasoning you use data from 2002 to say...well, this ended up being wrong.
Well, the estimate was higher than the reality, by definition it was over-estimated. They built out as if the tech boom was going to go on forever, and of course it didn't. You can argue that they made the best estimates they could with the information available, but ultimately it's still true that their estimates were wrong.
Your blog article stopped at token generation... you need to continue to revenue per token. Then go even further... The revenue for AI company is a cost for the AI customer. Where is the AI customer going to get incremental profits from the cost of AI.
For short searches, the revenue per token is zero. The next step is $20 per month. For coding it's $100 per month. With the competition between Gemini, Grok, ChatGPT... it's not going higher. Maybe it goes lower since it's part of Google's playbook to give away things for free.
They discuss it in the podcast. Laid fiber is different because you can charge rent for it essentially forever. It seems some people swooped in when it crashed and now own a perpetual money machine.
IBM might not have a data strategy or AI plan but he isn’t wrong on the inability to generate a profit.
A bit of napkin math:
NVIDIA claims 0.4J per token for their latest generation
1GW plant with 80% utilisation can therefore produce 6.29 10^16 tokens a year.
There are ~10^14 tokens on the internet. ~10^19 tokens have been spoken by humans… so far.
Don't know what the source is, but it feels missing a few orders of magnitude. Surely it only counts text? I can't imagine there are only so few data on the internet if you count images and videos.
Tokens are, roughly speaking, how you pay for AI. So you can approximate revenue by multiplying tokens per year by the revenue for a token.
(6.29 10^16 tokens a year) * ($10 per 10^6 tokens)
= $6.29 10^11
= $629,000,000,000 per year in revenue
Per the article
> "It's my view that there's no way you're going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest," he said.
$629 billion is less than $800 billion. And we are talking raw revenue (not profit). So we are already in the red.
But it gets worse, that $10 per million tokens costs is for GPT-5.1, which is one of the most expensive models. And the costs don't account for input tokens, which are usually a tenth of the costs of output tokens. And using bulk API instead of the regular one halves costs again.
Realistic revenue projections for a data center are closer to sub $1 per million tokens, $70-150 billion per year. And this is revenue only.
To make profits at current prices, the chips need to increase in performance by some factor, and power costs need to fall by another factor. The combination of these factors need to be, at minimum, like 5x, but realistically need to be 50x.
The math here is mixing categories. The token calculation for a single 1-GW datacenter is fine, but then it gets compared to the entire industry’s projected $8T capex, which makes the conclusion meaningless. It’s like taking the annual revenue of one factory and using it to argue that an entire global build-out can’t be profitable. On top of that, the revenue estimate uses retail GPT-5.1 pricing, which is the absolute highest-priced model on the market, not what a hyperscaler actually charges for bulk workloads. IBM’s number refers to many datacenters built over many years, each with different models, utilization patterns, and economics. So this particular comparison doesn’t show that AI can’t be profitable—it’s just comparing one plant’s token output to everyone’s debt at once. The real challenges (throughput per watt, falling token prices, capital efficiency) are valid, but this napkin math isn’t proving what it claims to prove.
> but then it gets compared to the entire industry’s projected $8T capex, which makes the conclusion meaningless.
Aren't they comparing annual revenue to the annual interest you might have to pay on $8T? Which the original article estimates at $800B. That seems consistent.
Broad estimates I'm seeing on the cost of a 1GW AI datacenter are $30-60B. So by your own revenue projection, you could see why people are thinking it looks like a pretty good investment.
Note that if we're including GPU prices in the top-line capex, the margin on that $70-150B is very healthy. From above, at 0.4J/T, I'm getting 9MT/kWh, or about $0.01/MT in electricity cost at $0.1/kWh. So if you can sell those MT for $1-5, you're printing money.
> So if you can sell those MT for $1-5, you're printing money.
The IF is doing a lot of heavy lifting there.
I understood the OP in the context of "human history has not produced sufficiently many tokens to be sent into the machines to make the return of investment possible mathematically".
Maybe the "token production" accelerates, and the need for so much compute realizes, who knows.
I see a lot of people attacking the messenger but very few addressing the basic logic that you need 800B+ in profit just to pay the interest on some of these investments.
Pointing out IBM's mixed history would be valid if they were making some complex, intricate, hard to verify case for why AI won't be profitable. But the case being made seems like really simple math. A lot of the counterarguments to these economic problems have the form "this time it's different" - something you hear every bubble from .com to 2008.
The interesting macro view on what's happening is to compare a mature data center operation (specifically a commoditized one) with the utility business. The margins here, and in similar industries with big infra build-out costs (ex: rail) are quite small. Historically the businesses have not done well; I can't really imagine what happens when tech companies who've only ever known huge, juicy margins experience low single digit returns on billions of investment.
Worse, is that a lot of these people are acting like Moore's law isn't still in effect. People conflate clock speeds on beefy hardware with moore's law, and act like it's dead, when transistor density rises, and cost per transistor continue to fall at rates similar to what they always have. That means the people racing to build out infrastructure today might just be better off parking that money in a low interest account, and waiting 6 months. That was a valid strategy for animation studios in the late 90s (it was not only cheaper to wait, but also the finished renders happened sooner), and I'd be surprised if it's not a valid strategy today for LLMs. The amount of silicon that is going to be produced that is specialized for this type of processing is going to be mind boggling.
Cost per transistor is increasing. or flat, if you stay on a legacy node. They pretty much squeezed all the cost out of 28nm that can be had, and it’s the cheapest per transistor.
“based on the graph presented by Milind Shah from Google at the industry tradeshow IEDM, the cost of 100 million transistors normalized to 28nm is actually flat or even increasing.”
Yep. Moore's law ended at or shortly before the 28nm era.
That's the main reason people stopped upgrading their PCs. And it's probably one of the main reasons everybody is hyped about Risc-V and the pi 2040. If Moore's law was still in effect, none of that would be happening.
That may also be a large cause of the failure of Intel.
> Moore's law ended at or shortly before the 28nm era.
Moore's law isn't about cost or clock speed, it's about transistor density. While the pace of transistor density increases has slowed, it's still pretty impressive. If we want to be really strict, and say densities absolutely have to double every 2 years, Moore's Law hasn't actually been true since 1983 or so. But it's been close, so 2x/2yr a decent rubric.
The fall-off from the 2x/2yr line started getting decently pronounced in the mid 90s. At the present time, over the past 5-6 years, we're probably at a doubling in density every 4-ish years. Which yes, is half the rate Moore observed, but is still pretty impressive given how mature the technology is at this point.
If you want to be pedantic, the original (and revised) law are definitely about cost. The original formulation was that the number of features (i.e. transistors) on an integrated circuit doubled every two years for the best-priced chips (smallest cost per feature).
> The complexity for minimum component costs has increased at a rate of roughly a factor of two per year (see graph on next page).
And it is formulated in a section aptly titled "Costs and curves". This law has always been an economic law first, some kind of roadmap for fans to follow. But that roadmap drove almost-exponential investment costs as well.
I concede that density still rises, especially if you count " advanced packaging". But the densest and most recent is not the cheapest anymore.
The cloud mega scalers have done very well for themselves. As with all products the question is differentiation. If models can differentiate and lock in users they can have decent margins. If models get commoditized the current cloud providers will eat the AI labs lunch.
I question depreciation. those gpu's will be obsolete in 5 years, but will the newer be enough better as to be worth replacing them is an open question. cpu's stopped getting exponetially faster 20 years ago, (they are faster but not the jumps the 1990s got)
I recently compared performance per dollar for CPUs and GPUs on benchmarks for GPUs today vs 10 years ago, and suprisingly, CPUs had much bigger gains. Until I saw that for myself, I thought exactly the same thing as you.
It seems shocking given that all the hype is around GPUs.
This probably wouldn't be true for AI specific workloads because one of the other things that happened there in the last 10 years was optimising specifically for math with lower size floats.
It's coz of use cases. Consumer-wise, if you're gamer, CPU just needs to be at "not the bottleneck" level for majority of games as GPU does most of the work when you start increasing resolution and details.
And many pro-level tools (especially in media space) offload to GPU just because of so much higher raw compute power.
So, basically, for many users the gain in performance won't be as visible in their use cases
MTBF for data center hardware is short; DCs breeze through GPUs compared to even the hardest of hardcore gamers.
And there is the whole FOMO effect to business purchases; decision makers will worry their models won't be as fast.
Obsolete doesn't mean the reductive notion you have in mind, where theoretically it can still push pixels. Physics will burn them up, and "line go up" will drive demand to replace them.
My bellybutton fluff, uninformed opinion is that heat cycling and effective cooling are probably a much more limiting factor.
If you are running a gpu at 60C for months at a time, but never idling it (crypto use case), I would actually hazard a guess that it is better than cycling it with intermittent workloads due to thermal expansion.
That of course presupposes effective, consistent cooling.
There is the question - will they be worth the upgrade? Either because they are that much faster, or that much more energy efficient. (and also assuming you can get them, unobtainium is worth that what you have).
Also a nod to the other reply that suggests they will wear out in 5 years. I cannot comment on if that is correct but it is a valid worry.
It's not that hard to see the old GPUs being used e.g. for inference on cheaper models, or sub-agents, or mid-scale research runs. I bet Karpathy's $100 / $1000 nanochat models will be <$10 / <$100 to train by 2031
I think real issue is current costs / demand = Nvidia gouging GPU price that costs for hardware:power consumption is 70:20 instead of 50:40 (10 for rest of datacenter). Reality is gpus are serendipidous path dependent locked from gaming -> mining. TPUs are more power efficient, if bubble pops and demand for compute goes down, Nvidia + TMSC will still be around, but nexgen AI first bespoke hardware premium will revert towards mean and we're looking at 50% less expensive hardware (no AI race scarcity tax, i.e. 75% Nvidia margins) that use 20% less power / opex. All of a sudden existing data centers becomes not profitable stranded assets even if they can be stretched past 5 years.
"It is 1958. IBM passes up the chance to buy a young, fledgling company that has invented a new technology called xerography. Two years later, Xerox is born, and IBM has been kicking themselves ever since. It is ten years later, the late '60s. Digital Equipment DEC and others invent the minicomputer. IBM dismisses the minicomputer as too small to do serious computing and, therefore, unimportant to their business. DEC grows to become a multi-hundred-million dollar corporation before IBM finally enters the minicomputer market. It is now ten years later, the late '70s. In 1977, Apple, a young fledgling company on the West Coast, invents the Apple II, the first personal computer as we know it today. IBM dismisses the personal computer as too small to do serious computing and unimportant to their business." - Steve Jobs [1][2][3]
Now, "IBM CEO says there is 'no way' spending on AI data centers will pay off". IBM has not exactly had a stellar record at identifying the future.
> IBM has not exactly had a stellar record at identifying the future.
IBM invented/developed/introduced magnetic stripe cards, UPC Barcodes, the modern ATM, Hard drives, floppies, DRAM, SQL, the 360 Family of Mainframes, the PC, Apollo guidance computers, Deep Blue. IBM created a far share of the future we're living in.
I'm no fan of much of what IBM is doing at the moment but it could be argued that its consultancy/service orientation gives it a good view of how business is and is planning to use AI.
They also either fairly accurately predicted the death of HDDs by selling off their research division before the market collapsed, or they caused the end of the HDD era by selling off their research division. They did a lot of research.
I think the retail market is maybe dead but datacenters are still a fairly large customer I’d think. HDDs really shine at scale where they can be fronted by flash and DRAM cache layers.
They are still cheaper than flash for cold data, but that’s not going to hold for long. Flash is so much denser the acquisition cost difference for a multi-petabyte store becomes small next to the datacenter space and power needed by HDDs. HDDs require research for increasing density while flash can rely on silicon manufacturing advances for that - not that it doesn’t require specific research, but being able to apply the IP across a vast space makes better economical sense.
The other way to look at it is that the entire consulting industry is teetering on catastrophe. And IBM, being largely a consulting company now, is not being spared.
IBM isn't failing, though. They're a profitable company with healthy margins, and enterprises continue to hire them for all sorts of things, in large numbers.
> The other way to look at it is that the entire consulting industry is teetering on catastrophe
Oh? Where'd you get that information?
If you mean because of AI, it doesn't seem to apply much to IBM. They are probably not great at what they do like most such companies, but they are respectable and can take the blame if something goes wrong. AI doesn't have these properties.
If anything there’s likely plenty of work for body shops like IBM in reviewing and correcting AI-generated work product that has been thrown into production recently.
> A failing company may still be right in identifying other companies failure modes.
Agreed if this is what they are doing, but what if theyre spewing claims to try and discredit an industry in order to quell their shareholder concerns?
They are not the only ones looking at the money spent in AI datacentres and concluding most of the investment will not be recovered anytime soon.
A lot of the silicon being deployed is great for training, but inefficient for inference and the training to inference ratio for usage shows a clear tendency to go the inference way. Furthermore, that silicon, with the workloads it runs, doesn’t last long and needs replacement.
The first ones to go online might recover the investment, but the followers better have a plan to pivot to other uses.
I've heard some second hand stories about IBM's way of using "AI" and it is pretty much business oriented and not much of the glamour and galore promises the other companies make (of course you still have shiny new things in business terms). It's actually good entertainment hearing all the internal struggles of business vs fancy during the holidays.
IBM is always very conscious of what their clients need (and the large consultancy business provides a very comprehensive view). It just turns out their clients don’t need IBM to invest in large frontier models.
> IBM invented/developed/introduced magnetic stripe cards, UPC Barcodes, the modern ATM, Hard drives, floppies, DRAM, SQL, the 360 Family of Mainframes, the PC, Apollo guidance computers, Deep Blue. IBM created a far share of the future we're living in.
Well put. “IBM was wrong about computers being a big deal” is a bizarre take. It’s like saying that Colonel Sanders was wrong about chicken because he, uh… invented the pressure fryer.
This is the exact kind of thinking that got us into this mess in the first place, and I'm not blaming you for it, it seems to be something all of us do to an extent. We don't look to Meta, who only a few years ago thought that the Metaverse would be the "next big thing" as an example of failure to identify the future, we look to IBM who made that mistake almost 30 years ago. Underestimating a technology seems to stick much harder than overestimating one.
If you want to be seen as relevant in this industry, or as a kind of "thought leader", the easy trick seems to be to hype up everything. If you do that and you're wrong, people will quickly forget. If you don't and you're wrong, that will stain your reputation for decades.
Good point. That kind of thinking is an absurdity. Saying IBM dropped the ball 70 years ago without acknowledging that lessons were learned, leadership has changed hands a lot since then, and most importantly, the tech landscape back then was very different from today unless you grossly oversimplify everything amounts to nothing more than a fallacious opinion.
Not even much of an IBM fan, myself, but I respect their considerable contribution to the industry. Sure, they missed a shot back then, but I think this latest statement is reliably accurate based on the information we currently have.
The amount of hate I've received here for similar statements is astonishing. What is even more astonishing is that it takes 3-rd grade math skills to work out that the current AI(even ignoring the fact that there is nothing intelligent about the current AI) costs are astronomical and they do not deliver on the promises and everyone is operating at wild loses. At the moment we are at "if you owe 100k to your bank, you have a problem but if you owe 100M to your bank, your bank has a problem". It's the exact same bullshitter economy that people like musk have been exploiting for decades: promise a ton, never deliver, make a secondary promise for "next year", rinse and repeat -> infinite profit. Especially when you rope in fanatical followers.
I don't want to defend musk in any way but I think you are making a mistake there using him as an example because what boosted him quite a lot is that he actually delivered what he claimed. Always late but still earlier than anybody was guesstimating. And now he is completely spiraling but its a lot harder to lose a billion than to gain one so he persists and even gets richer. Plus his "fanatical" followers are poor. It just doesn't match the situation.
Sounds a lot like "I'm not racist but". There's a website dedicated to all of his bs https://elonmusk.today
He is the definition of a cult. Collects money from fanatical followers who will praise every word he says, never delivers, "oh next year guys, for sure, wanna buy a not a flamethrower, while you are at it?". Not to mention what once were laughable conspiracy theories about him turned out to be true(such that even I laughed when I heard them). Torvalds is right with his statement about musk: "incompetent" and "too stupid to work at a tech company".
> We don't look to Meta, who only a few years ago thought that the Metaverse would be the "next big thing" as an example of failure to identify the future, we look to IBM who made that mistake almost 30 years ago.
The grandparent points to a pattern of failures whereas you point to Meta’s big miss. What you miss about Meta, and I am no fan, is that Facebook purchased Whatsapp and Instagram.
In other words, two out of three ain’t bad; IBM is zero for three.
While that’s not the thrust of your argument, which is about jumping on the problem of jumping on every hype train, the post to which you reply is not on about hype cycle. Rather, that post calls out IBM for a failure to understand the future of technology and does so by pointing to a history of failures.
> Many others in this thread have pointed out IBM's achievements but regardless, IBM is far from "zero for three".
I was specifically commenting in the context of this thread.* I was not trying to characterize either IBM or Meta except with reference to the arguments offered by this thread’s ancestors.
I understood (and understand) that such scorekeeping of a company as storied as IBM is at best reductive and at worst misrepresentative.
* Your reference to “this thread” actually addresses sibling comments to OP (ggggp), not this thread which was started by gggp.
He is pointing out that the current costs to create the data centres means you will never be able to make a profit to cover those costs. $800 Billion just to cover the interest.
OpenAI is already haemorrhaging money and the space data centres has already been debunked. There is even a recent paper that points out that LLMs will never become AGI.
The article also finishes out with some other experts giving the same results.
Is this AI paper written by a reputable subject matter expert? It seems to be written by a physicist and also be the only academic work by this author in English
How many articles on this topic do we imagine there are? Thousands? Hundreds of thousands? It is hopeless to read every one by any author, no matter how unrelated to the domain, and judge them individually on their merits. Being a subject domain expert is not a perfect measure of paper quality but it's the only feasible way to make a first pass at filtering.
Even if I did read it, I have no hope of understanding if it has made a fundamental mistake because I don't have the subject matter expertise either.
(I imagine it has made a fundamental mistake anyway: for LLMs to be useful progress toward AGI they don't have to be a feasible way to create AGI by themselves. Innovation very often involves stepping through technologies that end up only being a component of the final solution, or inspiration for the final solution. This was always going to be an issue with trying to prove a negative.)
The core piece as quoted from the abstract: "AGI predictions fail not from
insufficient compute, but from fundamental misunderstanding of what intelligence demands structurally."
Then goes in detail as to what that is and why LLMs don't fit that. There are plenty other similar papers out there.
took me a while but i read it. thought it was actually a pretty good and well researched paper that does a good job rationalizing its thesis. thanks for sharing
Sry to say but the fact that you argue with LLMs never become AGI, you are not up-to-date.
People don't assume LLM will be AGI, people assume that World Models will lead us to AGI.
I personally never asumed LLM will become AGI, i always assumed that LLM broke the dam for investment and research into massivce scale compute ML learning and LLMs are very very good in showing were the future goes because they are already so crazy good that people can now imagine a future were AGI exists.
And that was very clear already when / as soon as GPT-3 came out.
The next big thing will probably be either a LOT more RL or self propelling ai architecture discovery. Both need massive compute to work well but then will potentially provide even faster progress as soon as humans are out of the loop.
AlphaGo Zero doesn't need much human intervention at all.
Regarding world models: All the big ones. LeCun, Demis Hassabis, Fei-Fei Li too. And they are all working on it.
LLMs will definitly play some type of role in AGI. After all you can ask an LLM already a lot of basic things like 'what are common tasks to make a tea'. A type of guide, long term fact memory or whatever this can be called.
Are OpenAI or Anthropic et al seriously building towards “world models”? I haven’t seen any real evidence of that. It seems more like they are all in on milking LLMs for all they are worth.
> In 1977, Apple, a young fledgling company on the West Coast, invents the Apple II, the first personal computer as we know it today. IBM dismisses the personal computer as too small to do serious computing and unimportant to their business.
IBM released the 5100 in September 1975 [0] which was essentially a personal computer in feature set. The biggest problem with it was the price tag - the entry model cost US$8975, compared to US$1298 for the entry Apple II released in June 1977 (close to two years later). The IBM PC was released in August 1981 for US$1565 for the most basic system (which almost no one bought, so in practice they cost more). And the original IBM PC had model number 5150, officially positioning it as a successor to the 5100.
IBM’s big problem wasn’t that they were disinterested in the category - it was they initially insisted on using expensive IBM-proprietary parts (often shared technology with their mainframe/midrange/minicomputer systems and peripherals), which resulted in a price that made the machine unaffordable for everyone except large businesses, governments, universities (and even those customers often balked at the price tag). The secret of the IBM PC’s success is they told the design team to use commercial off-the-shelf chips from vendors such as Intel and Motorola instead of IBM’s own silicon.
Got anything vis-a-vis the message as opposed to the messenger?
I'm not sure these examples are even the gotchas you're positing them as. Xerox is a dinosaur that was last relevant at the turn of the century, and IBM is a $300bn company. And if it wasn't obvious, the Apple II never made a dent in the corporate market, while IBM and later Windows PCs did.
In any case, these examples are almost half a century old and don't relate to capex ROI, which was the topic of dicussion.
If it's not obvious, Steve's quote is ENTIRELY about capex ROI, and I feel his quote is more relevant to what is happening today than anything Arvind Krishna is imagining. The quote is posted in my comment not to grandstand Apple in any sense, but to grandstand just how consistently wrong IBM has been about so many opportunities that they have failed to read correctly - reprography, mini computers and microcomputers being just three.
Yes it is about ROI: "IBM enters the personal computer market in November ’81 with the IBM PC. 1983 Apple and IBM emerged as the industry’s strongest competitors each selling approximately one billion dollars worth of personal computers in 1983, each will invest greater than fifty million dollars for R&D and another fifty million dollars for television advertising in 1984 totaling almost one quarter of a billion dollars combined, the shakeout is in full swing. The first major firm goes bankrupt with others teetering on the brink, total industry losses for 83 out shadow even the combined profits of Apple and IBM for personal computers."
I don’t think this is really a fair assessment. IBM is in fact a huge company today and it is possible that they are because they took the conservative approach in some of their acquisition strategy.
It is a bit like watching someone play poker and fold and then it turns out they had the high hand after all. In hindsight you could of course know that the risk would have been worth it but at the moment perhaps it did not seem like it given the money the first player would be risking.
> I don’t think this is really a fair assessment. IBM is in fact a huge company today and it is possible that they are because they took the conservative approach in some of their acquisition strategy.
I can also imagine IBM was being approached by hundreds, if not thousands, propositions. That they missed three that turned out to be big is a statistical probability.
A big difference is that in the past things like the potential of the PC were somewhat widely underestimated. And then the internet was again as well.
But in modern times it's rather the opposite scenario. The average entity is diving head first into AI simply expecting a revolutionary jump in capability that a more 'informed', for lack of any less snooty term, perspective would suggest is quite unlikely to occur anytime in the foreseeable future. Basically we have a modern day gold rush where companies and taking out unbelievably massive loans to invest in shovels.
The only way this doesn't catastrophically blow up is if AI companies manage to convince the government they're too big to fail, and get the Boeing, Banks, et al treatment. And I expect that's exactly the current strategy, but that's rather a high risk, low reward, type strategy.
>things like the potential of the PC were somewhat widely underestimated.
The potential of the AI that comes within reach at maximum expenditure levels may just be more widely overestimated.
The potential to make "that much money" even more challenging.
A very opposite scenario.
I think so many corporations are looking at how expensive actual humans always have been, and can be sure will always be, so much so that it's a major cost item that can not be ignored. AI opens up the possibility of a whole new level of automation or outright replacement for the routine simple-minded tasks, to a degree that never existed before. More jobs could possibly be eliminated than previous waves of mechanical and digital automation.
When you do the business math, the savings could be enormous.
But you can only realistically save as much as you are actually wasting, otherwise if you go too far you shoot yourself in the foot.
Even with all that money to work with, if you're in practice hunkering down for savings because you can't afford real people any more, you surely can't say the sky's the limit. Not like selling PC's or anything that's capable of more unbridled growth.
When PC's arrived they flew off the shelf even at their high initial retail prices.
People in droves (but not the silent majority) are shunning free AI and the movement is growing with backlash in proportion to the foisting.
I have no special knowledge about IBM Vs Apple historically, but: a quarter billion in CAPEX when you've earned a billion in revenue in a single year is extremely different to what we're seeing now. These companies are spending all of their free cash flow, then taking on debt, to the tune of percentage points of world GDP, and multiples of any revenue they've seen so far. That kind of oversupply is a sure fire way to kill any ROI.
> Got anything vis-a-vis the message as opposed to the messenger?
Sure: People disagree. It's not like there is anything particularly clever that IBM CEO provided here. The guy not investing in something saying it won't work is about as good as the people who do saying it will. It's simply different assumptions about the future.
The message is plain to see with very little advanced math.
The only news is that it is the CEO of IBM saying it out loud.
IMHO he has some of the most credible opinions at this scale that many people have seen.
It's "highly unlikely" that all this money will be paid back to everyone that invested at this point. The losers probably will outnumber the winners, and nobody knows whether it will end up becoming a winner-take-all situation yet. A number of wealthy players remain at the table, raising stakes with each passing round.
It's so much money that it's already too late to do anything about it, and the full amount hasn't even changed hands yet.
And the momentum from something so huge can mean that almost the entire amount will have to change hands a second time before a stable baseline can be determined relative to pre-existing assets.
This can take longer than anyone gives credit for just because of massiveness, in the mean time, established real near-term growth opportunities may languish or even fade as the skew in rationality/solvency balance awaits the rolling dice to come to rest.
IBM is an interesting beast when it comes to business decisions. While I can't give exact details, their business intelligence and ability to predict monetary things is uncannily spot-on at times.
So, when their CEO says that this investment will not pay off, I tend to believe them, because they most probably have the knowledge, insight and data to back that claim, and they have ran the numbers.
Oh, also, please let's not forget that they dabbled in "big AI" before everyone else. Anyone remembers Deep Blue and Watson, the original chatbot backed by big data?
We can cherry-pick blunders made by any big company to make a point. Maybe it would be more honest to also list companies IBM passed on that turned out to be rubbish? And all the technologies that IBM did invest in that made them a ton of money and became industry standards?[0]
Today, Xerox has less total revenue than IBM has profit. DEC went out of business 27 years ago. Apple is an in astoundingly great place right now, but Jobs got kicked out of his own company, and then returned when it was about to fail, having to take investment from Microsoft(!) in order to stay afloat.
Meanwhile, IBM is still here, making money hand over fist. We might not have a ton of respect for them, being mostly a consulting services company these days, but they're doing just fine.
No, but the comment above and variations of it are mentioned in every thread about IBM, so it’s probably just a reflex at this point without much thought behind it.
That's completely beyond the point, though?
Kodak invented the digital camera, did not think anything about it and others then ate their lunch.
Those others are also not crushing it in 2025.
The point is IBM is not the go-to to listen about AI.
Also not saying they are not right, even a broken clock is right 2 times a day.
> The point is IBM is not the go-to to listen about AI.
Why not, though? For better or worse, they're a consulting services company these days, and they work with an eye-wateringly large number of companies. I would expect them to have a very good view as to what companies use AI for, and plan/want to use AI for in the future. They may not be experts in the tech itself, but I think they're decently well-positioned to read the tea leaves.
"The amount being spent on AI data centres not paying off" is a different statement to "AI is not worth investing in". They're effectively saying the portions people are investing are disproportionately large to what the returns will end up being.
It's a difficult thing to predict, but I think there's almost certainly some wasteful competition here. And some competitors are probably going to lose hard. If models end up being easy to switch between and the better model is significantly better than its competitors, than anything invested in weaker models will effectively be for nothing.
But there's also a lot to gain from investing in the right model, even so it's possible those who invested in the winner may have to wait a long time to see a return on their investment and could still possibly over allocate their capital at the expense of other investment opportunities.
> Even if all the people are gone but the culture hasn’t changed
Can you expand on this? What was the culture then versus now?
For example back then it was the culture to have suit inspectors ensure you had the right clothes on and even measure your socks. (PBS Triumph of the Nerds)
I mean, okay, but you're taking the current leadership's words and claiming they are incorrect because IBM management was not great at identifying trends decades ago. Historical trend is not an indicator of the future and it's not engaging in good faith on the conversation if overspending on AI can be backed by revenue in the future. You're attacking the messenger instead of the message.
I’m saying given IBMs track record of completely failing at innovation repeatedly and failing at investing on the correct technologies, why do you assume today’s CEO is better at it and has bucked the milquetoast culture that has pervaded IBM? It’s a company that has largely divested its ability to provide solutions and technology and turned more into a large shop consultancy (+ milking their legacy mainframe contracts and whatnot).
Except that, over the same time period, IBM was also very successful at innovation and investing in technologies. Yes, they made some very high-profile misses that the top poster lists, but they were still a powerhouse doing other stuff during that time (as a commenter replying to that points out).
Except all the technologies listed were basically in the heydey of the 70s and 80s. Gerstner redirected IBM into an enterprise services business in the early 90s and by the late 90s that's where all their money was. The 2002 acquisition of PwC consulting cemented this shift. In the mid 2010s under Ginni Rometty IBM software became secondary and it abandoned selling it as standalone software instead bundling it into its consultancy.
So IBM hasn't been doing hardware R&D for about three decades and abandoned software R&D well over a decade ago. R&D hasn't been in their DNA for a long time, their previous contributions notwithstanding.
I don't understand how calling something "a ship of Theseus in corporate form" or "culture hasn't changed" etc, is not dismissive of the actual comment by the CEO on AI overspending. They dismissed the content of the message by saying IBM's culture sucks, is how i read it. Also things can be funny and dimissive at the same time, they often are.
A revolution means radical changes executed over a short period of time. Well with 4 years in, this has got to be one of the smallest "revolutions" we have ever witnessed in human history. Maybe it's revolutionary for people who get excited about crappy pictures they can insert into their slides to impress the management.
They put these stories out just to make the general public (who might not understand that this is just bs) but makes AI seem scary so people get a lopsided view of AI and capacities that are straight out of science fiction.
Millions is an understatement on how much AI marketing spend is
IBM sees the funding bubble bursting and the next wave of AI innovation as about to begin.
IBM was too early with "Watson" to really participate in the 2018-2025 rapid scaling growth phase, but they want to be present for the next round of more sensible investment.
IBM's CEO is attempting to poison the well for funding, startups, and other ventures so IBM can collect itself and take advantage of any opportunities to insert itself back into the AI game. They're hoping timing and preparation pay off this time.
It's not like IBM totally slept on AI. They had Kubernetes clusters with GPUs. They had models and notebooks. But their offerings were the absolute worst. They weren't in a position to service real customers or build real products.
Have you seen their cloud offerings? Ugh.
They're hoping this time they'll be better prepared. And they want to dunk on AI to cool the playing field as much as they can. Maybe pick up an acquisition or two on the cheap.
How exactly are they poisoning the well..? OpenAI committed to 1.4 trillion investements...with a revenue of ~13B - how is IBM CEO contributing to that absolutely already poisoned situation? Steve Jobs did not care about naysayers when he introduced iPhone - because his product was so innovative for the time. According to AI boosters, we now have a segment of supposedly incredibly powerful and at the same time "dangerous" AI products. Why are they not sweeping the floor off with the "negators", "luddites", "laggards" etc... After so many hundreds of billions of dollars and supposedly so many "smart" AI researchers...Where are the groundbreaking results man? Where are the billion-dollar startups launched by single persons (heck, I'd settle even for a small team)...Where are the ultimate applications..etc?
Steve Jobs, the guy that got booted out of his own company and that required a lifeline from his arch nemesis to survive?
This is all true, but it was only true in hindsight and as such does not carry much value.
It's possible that you are right and AI is 'the future' but with the present day AI offering I'm skeptical as well. It isn't at a level where you don't have to be constantly on guard against bs and in that sense it's very different from computing so far, where reproducibility and accuracy of the results were important, not the language that they are cast in.
AI has killed the NLP field and it probably will kill quite a few others, but for the moment I don't see it as the replacement of general computing that the proponents say that it is. Some qualitative change is still required before I'm willing to check off that box.
In other news: Kodak declares digital cameras a fad, and Microsoft saw the potential of the mp3 format and created a killer device called the M-Pod.
How about check out how many companies exist today vs existed in 1958? If you look at it that way then just surviving is an achievement in itself and then you might interpret their actions as extremely astute business acumen.
But how many companies did IBM pass on that did crash and burn ? And how many did it not pass on and did decently ? They're still around after more than 3 generations worth of tech industry. They're doing something right.
DEC went down the drain, Xerox is 1/1000 of IBM's market cap. IBM made its own, superior by its relative openness, personal computer that ended up running the world, mostly maintaining direct binary compatibility for 40+ years, even without IBM really paying attention.
This isn't even a great argument at a literal level. Nowadays nobody cares about Xerox and their business is selling printers, DEC was bought by Compaq which was bought by HP. Apple is important today because of phones, and itself was struggling selling personal computers and needed a (antitrust-motivated) bailout from Microsoft to survive during the transition.
Cool story, but it’s more than just the opinion of this CEO. It’s logic.
Hardware is not like building railroads, the hardware is already out of date once deployed and the clock has started ticking on writing off the expense or turning a profit on it.
There are fundamental discoveries needed to make the current tech financially viable and an entire next generation of discoveries needed to deliver on the over inflated promises already made.
You, or your existence, probably triggers multiple transactions per day through a POWER mainframe without you even knowing it. Their mainframes handle the critical infrastructure that can't go down.It's so reliable we don't even think about it. I shudder to think about Microsoft or Apple handling that.
For some strange reason a lot of people were attracted by a comment that speaks about everything else BUT the actual topic and its the top comment now. Sigh.
If you think that carefully chosen anecdotes out of many many more are relevant, there needs to be at least an attempt of reasoning. There is nothing here. It's really just barebones mentioning of stuff intentionally selected to support the preconceived point.
I think we can, and should, do better in HN discussions, no? This is "vibe commenting".
I’m sorry, but this is stupid, you understand that you have several logical errors in your post? I was sure Clinton is going to win 2016. Does that mean that when I say 800 is bigger than 8 is not to be trusted?
Do people actually think that running a business is some magical realism where you can manifest yourself to become a billionaire if you just believe hard enough?
The idea that a company DNA somehow lives over 100 years and maintains the same track record is far fetched.
that the OpenAI tech bro are investing in AI using a grown up ROI is similarly far fetched, they are burning money to pull ahead of the reset and assume the world will be in the palm of the winner and there is only 1 winner.
Will the investment pay off if there are 3 neck and neck companies ?
IBM is still alive and kicking well, and definitively more relevant than Xerox or DEC. You are completely misconstruing Jobs’ point to justify the current AI datacenter tulip fever.
A decade ago, IBM was spending enormous amounts of money to tell me stuff like "cognitive finance is here" in big screen-hogging ads on nytimes.com. They were advertising Watson, vaporware which no one talks about today. Are they bitter that someone else has actually made the AI hype take off?
I don't know that I'd trust IBM when they are pitching their own stuff. But if anybody has experience with the difficulty of making money off of cutting-edge technology, it's IBM. They were early to AI, early to cloud computing, etc. And yet they failed to capture market share and grow revenues sufficiently in those areas. Cool tech demos (like the Watson Jeopardy) mimic some AI demos today (6-second videos). Yeah, it's cool tech, but what's the product that people will actually pay money for?
I attended a presentation in the early 2000s where an IBM executive was trying to explain to us how big software-as-a-service was going to be and how IBM was investing hundreds of millions into it. IBM was right, but it just wasn't IBM's software that people ended up buying.
Xerox was also famously early with a lot of things but failed to create proper products out of it.
Google falls somewhere in the middle. They have great R&D but just can’t make products. It took OpenAI to show them how to do it, and the managed to catch up fast.
"They have great R&D but just can’t make products"
Is this just something you repeat without thinking? It seems to be a popular sentiment here on Hacker News, but really makes no sense if you think about it.
Products: Search, Gmail, Chrome, Android, Maps, Youtube, Workspace (Drive, Docs, Sheets, Calendar, Meet), Photos, Play Store, Chromebook, Pixel ... not to mention Cloud, Waymo, and Gemini ...
So many widely adopted products. How many other companies can say the same?
I don't think Google is bad at building products. They definitely are excellent at scaling products.
But I reckon part of the sentiment stems from many of the more famous Google products being acquisitions orignally (Android, YouTube, Maps, Docs, Sheets, DeepMind) or originally built by individual contributors internally (Gmail).
Then here were also several times where Google came out with multiple different products with similar names replacing each other. Like when they had I don't know how many variants of chat and meeting apps replacing each other in a short period of time. And now the same thing with all the different confusing Gemini offerings. Which leads to the impression that they don't know what they are doing product wise.
Starting with an acquisition is a cheap way of accelerating once your company reaches a certain size.
Look at Microsoft - Powerpoint was an acquisition. They bought most of the team that designed and built Windows NT from DEC. Frontpage was an acquisition, Azure came after AWS and was led by a series of people brought in in acquisitions (Ray Ozzie, Mark Russinovich, etc.). It's how things happen when you're that big.
Because those were "free time" projects. It wasn't directed to do by the company, somebody at the company with their flex time - just thought it was a good idea and did it. Googlers don't get this benefit any more for some reason.
Those are all free products, some of them are pretty good. But free is the best business strategy to get a product to the top of the market. Are others better, are you willing to spend money to find out? Clearly, most people are not interested. The fact that they can destroy the market for many different types of software by giving it away and still stay profitable is amazing. But that's all they are doing. If they started charging for everything there would be better competition and innovation. You could move a whole lot of okay-but-not-great cars, top every market segment you want, if you gave them away for free. Only enthusiasts would remain to pay for slightly more interesting and specific features. Literally no business model can survive when their primary product is competing with good-enough free products.
> Products: Search, Gmail, Chrome, Android, Maps, Youtube, Workspace (Drive, Docs, Sheets, Calendar, Meet), Photos, Play Store, Chromebook, Pixel ... not to mention Cloud, Waymo, and Gemini ...
Many of those are acquisitions. In-house developed ones tend to be the most marginal on that list, and many of their most visibly high-effort in-house products have been dramatic failures (e.g. Google+, Glass, Fiber).
I was extremely surprised that Google+ didn't catch on. The week before Google+ launched, me and all my friends agreed that Facebook is toast, Google will do the same thing but better, and everyone has a Gmail account so there will be basically zero barrier to entry. Obviously, we were wrong; Google+ managed to snatch defeat out of the jaws of victory, Google+ never got significant traction, and Facebook managed to keep growing and now they're yet another Big Evil Tech Corporation.
Honestly, I still don't really know how Google managed to mess that up.
I got early access to Google+ because of where I worked at the time. The invite-only thing had worked great for GMail but unfortunately a social network is useless if no-one else is on it. Then the real names thing and the resulting drumbeat of horror stories like "Google doxxed me to my violent ex-husband" killed what little momentum they had stone dead. I still don't know why they went so hard on that, honestly.
They come up with tons and tons of products like Google Glass and Google+ and so on and immediately abandon them. It is easy to see that there is no real vision. They make money off AdSense and their cloud services. That's about it.
Google does abandon a lot of stuff, but their core technologies usually make their way into other, more profitable things (collaborative editing from Wave into Docs; loads of stuff from Google+; tagging and categorizing in Photos from Picasa (I'm guessing); etc)
It annoyed me recently that they dropped support for some Nest/Google Home thermostats. Of course, they politely offered to let me buy a replacement for $150.
I think the sentiment is usually paired with discussion about those products as long-lasting, revenue-generating things. Many of those ended up feeding back into Search and Ads. As an exercise, out of the list you described, how many of those are meaningfully-revenue-generating, without ads?
A phrasing I've heard is "Google regularly kills billion-dollar businesses because that doesn't move the needle compared to an extra 1% of revenue on ads."
And, to be super pedantic about it, Android and YouTube were not products that Google built but acquired.
They bought YouTube but you have to give Google a hell of a lot of credit for turning it into what it is today. Taking ownership of YouTube at the time was seen by many as taking ownership of an endless string of copyright lawsuits, suing them into oblivion.
Youtube maintains an independent campus from the google/alphabet mothership, I'm curious how much direction they get, as (outwardly, at least) appear to run semi-autonomously.
Before Google touched Android it was a cool concept but not what we think of today. Apparently it didn't even run on Linux. That concept came after the acquisition.
Notably all other than Gemini are from a decade or more ago. They used to know how to make products, but then they apparently took an arrow in the knee.
Search was the only mostly original product. With the exception of YouTube which was a purchase, Android and ChromeOS all the other products were initially clones.
Google had less incentive. Their incentive was to keep API bottled up and in brewing as long as possible so their existing moats in search, YouTube can extend in other areas. With openai they are forced to compete or perish.
Even with gemini in lead, its only till they extinguish or make chatgpt unviable for openai as business. OpenAI may loose the talent war and cease to be leader in this domain against google (or Facebook) , but in longer term their incentive to break fresh aligns with average user requirements . With Chinese AI just behind, may be google/microsoft have no choice either
Google was especially well positioned to catch up because they have a lot of the hardware and expertise and they have a captive audience in gsuite and at google.com.
The original statistical machine translation models of the 90's, which were still used well into the 2010's, were famously called the "IBM models" https://en.wikipedia.org/wiki/IBM_alignment_models These were not just cool tech demos, they were the state of the art for decades. (They just didn't make IBM any money.)
Neither cloud computing nor AI are good long term businesses. Yes, there's money to be made in the short term but only because there's more demand than there is supply for high-end chips and bleeding edge AI models. Once supply chains catch up and the open models get good enough to do everything we need them for, everyone will be able to afford to compute on prem. It could be well over a decade before that happens but it won't be forever.
This is my thinking too. Local is going to be huge when it happens.
Once we have sufficient VRAM and speed, we're going to fly - not run - to a whole new class of applications. Things that just don't work in the cloud for one reason or another.
- The true power of a "World Model" like Genie 2 will never happen with latency. That will have to run locally. We want local AI game engines [1] we can step into like holodecks.
- Nobody is going to want to call OpenAI or Grok with personal matters. People want a local AI "girlfriend" or whatever. That shit needs to stay private for people.
- Image and video gen is a never ending cycle of "Our Content Filters Have Detected Harmful Prompts". You can't make totally safe for work images or videos of kids, men in atypical roles (men with their children = abuse!), women in atypical roles (woman in danger = abuse!), LGBT relationships, world leaders, celebs, popular IPs, etc. Everyone I interact with constantly brings these issues up.
- Robots will have to be local. You can't solve 6+DOF, dance
routines, cutting food, etc. with 500ms latency.
- The RIAA is going door to door taking down each major music AI service. Suno just recently had two Billboard chart-topping songs? Congrats - now the RIAA lawyers have sued them and reached a settlement. Suno now won't let you download the music you create. They're going to remove the existing models and replace them with "officially licensed" musicians like Katy Perry® and Travis Scott™. You won't retain rights to anything you mix. This totally sucks and music models need to be 100% local and outside of their reach.
It is very misleading or outright perverse to write "they were selling software as a service in the IBM 360 days" when there was no public network that could be used to the deliver the service. (There were wide-area networks, but each one was used by a single organization and possibly a few of its most important customers and suppliers, hence the qualifier "public" above.)
But anyways, my question to you is, was there any software that IBM charged money for as opposed to providing the software at no additional cost with the purchase or rental of a computer?
I do know that no one sold software software (i.e., commercial off-the-shelf software) in the 1960s: the legal framework that allowed software owners to bring lawsuits for copyright violations appeared in the early 1980s.
There was an organization named SHARE composed of customers of IBM whereby one customer could obtain software written by other other customers (much like the open-source ecosystem) but I don't recall money ever changing hands for any of this software except a very minimal fee (orders of magnitude lower than the rental or purchase price of a System/360, which started at about $660,000 in 2025 dollars).
Also, IIUC most owners or renters of a System/360 had to employ programmers to adapt the software IBM provided. There is software with that quality these days, too (.e.g, ERP software for large enterprises) but no one calls that a software as a service.
> but it just wasn't IBM's software that people ended up buying.
Well, I mean, WebSphere was pretty big at the time; and IBM VisualAge became Eclipse.
And I know there were a bunch of LoB applications built on AS/400 (now called "System i") that had "real" web-frontends (though in practice, they were only suitable for LAN and VPN access, not public web; and were absolutely horrible on the inside, e.g. Progress OpenEdge).
...had IBM kept up the pretense of investment, and offered a real migration path to Java instead of a rewrite, then perhaps today might be slightly different?
I still have PTSD from how much Watson was being pushed by external consultants to C levels despite it being absolutely useless and incredibly expensive. A/B testing? Watson. Search engine? Watson. Analytics? Watson. No code? Watson.
I spent days, weeks arguing against it and ended up having to dedicate resources to build a PoC just to show it didn’t work, which could have been used elsewhere.
If anything, the fact they built such tooling might be why they're so sure it won't work. Don't get me wrong, I am incredibly not a fan of their entire product portfolio or business model (only Oracle really beats them out for "most hated enterprise technology company" for me), but these guys have tentacles just as deep into enterprises as Oracle and are coming up dry on the AI front. Their perspective shouldn't be ignored, though it should be considered in the wider context of their position in the marketplace.
Apples and Oranges from an enterprise perspective, with the additional wrinkle that consumer tech is generally ad-supported (ugh) while Enterprise stuff is super-high margin and paid for in actual currency.
If you assume the napkin math is correct on the $800bn yearly needed to service interest rates on these CAPEX loans, then you’d need the collective revenue of the major players (OpenAI, Google, Anthropic, etc) to pull in as much revenue in a year as Apple, Alphabet, and Samsung combined.
Let’s assume OpenAI is responsible for much of this bill, say, $400bn. They’d need a very generous conversion rate of 24% for their monthly users (700m) to the Pro plan for an entire year to cover that bill, for one year. That’s a conversion rate better than anyone else in the XaaS world who markets to consumers and enterprises alike, and paints a picture of just how huge the spend from enterprises would need to be to subsidize consumer free usage.
And all of this is just for existing infrastructure. As a number of CEBros have pointed out recently (and us detractors have screamed about from the beginning), the current CAPEX on hardware is really only good for three to five years before it has to be replaced with newer kit at a larger cost. Nevermind the realities of shifting datacenter designs to capitalize on better power and cooling technologies to increase density that would require substantial facility refurbishment to support them in a potential future.
The math just doesn’t make sense if you’re the least bit skeptical.
It makes it suspect when combined with the obvious incentive to make the fact that IBM is basically non-existent in the AI space look like an intentional, sagacious choice to investors. It very may well be, but CEOs are fantastically unreliable narrators.
No, I don’t trust a word Sundar or Satya say about AI either. CEOs should be hyping anything they’re invested in, it’s literally their job. But convincing investors that every thing they don’t invest in heavily is worthless garbage is effectively part of their job too.
What is more convincing is when someone invests heavily (and is involved heavily) and then decides to stop sending good money after bad (in their estimation). Not that they’re automatically right, but is at least pay attention to their rationales. You learn very little about the real world by listening to the most motivated reasoner’s nearly fact-free bloviation.
> What is more convincing is when someone invests heavily (and is involved heavily) and then decides to stop sending good money after bad (in their estimation).
Yeah I was going to say the same thing ha. I get what they’re (the commenter) saying, but one could also argue IBM is putting their money where their mouth is by not investing.
I suspect the reality is that they missed the boat, as they have missed tens of other boats since the mainframe market dried up. I guess you could argue they came to the boat too early with their pants on backwards (i.e. Watson), and then left before it showed up. But it’s hard to tell from the outside.
Maybe that will turn out to be a good decision and Microsoft/Google/etc. will be crushed under the weight of hundreds of billions of dollars in write-offs in a few years. But that doesn’t mean they did it intentionally, or for the right reasons.
I see Watson stuff at work. It’s not a direct to consumer product, like ChatGPT, but I see it being used in the enterprise, at least where I’m at. IBM gave up on consumer products a long time ago.
Just did some brief Wikipedia browsing and I'm assuming it's WatsonX and not Watson? It seems Watson has been pretty much discontinued and WatsonX is LLM based. If it is the old Watson, I'm curious what your impressions of it is. It was pretty cool and ahead of its time, but what it could actually do was way over promised and overhyped.
I’m not close enough to it to make any meaningful comments. I just see the name pop up fairly regularly. It is possible that some of it is WatsonX and everyone just says Watson for brevity.
One big ones used heavily is Watson AIOps. I think we started moving to it before the big LLM boom. My usage is very tangential, to the point where I don’t even know what the AI features are.
IBM has been "quietly" churning out their Granite models, with the latest of which performing quite well against LLaMa and DeepSeek. So not Anthropic-level hype but not sitting it out completely either. They also provide IP indemnification for their models, which is interesting (Google Cloud does the same).
It's good we are building all this excess capacity which will be used for applications in other fields or research or open up new fields.
I think the dilemma I see with building so much data centers so fast is exactly like whether I should buy latest iPhone now or should wait few years when the specs or form factor improves later on. The thing is we have proven tech with current AI models so waiting for better tech to develop on small scale before scaling up is a bad strategy.
IBM did a lot of pretty fragmented and often PR-adjacent work. And getting into some industry-specific (e.g. healthcare) things that didn't really work out. But my understanding is that it's better standardized and embedded in products these days.
Not to be rude, but that didn't answer my question.
Taking a look at IBM's Watson page, https://www.ibm.com/watson, it appears to me that they basically started over with "watsonx" in 2023 (after ChatGPT was released) and what's there now is basically just a hat tip to their previous branding.
Yep. When a brand has tarnished itself enough, it makes sense for the brand to step back. Nowadays, we interact with their more popular properties, such as Redhat.
My limited understanding (please take with a big grain of salt) is that they 1.) sell mainframes, 2.) sell mainframe compute time, 3.) sell mainframe support contracts, 4.) sell Red hat and Redhat support contracts, and 5.) buy out a lot of smaller software and hardware companies in a manner similar to private equity.
I can think of nothing more peak HN than criticizing a company worth $282 Billion with $6 billion in profit (for startup kids that means they have infinite runway and then some) that has existed for over 100 years with "I'm not even sure what they do these days". I mean the problem could be with IBM... what a loser company!
:) As much I love ragging on ridiculous HN comments, I think this one is rooted in some sensibility.
IBM doesn’t majorly market themselves to consumers. The overwhelming majority of devs just aren’t part of the demographic IBM intends to capture.
It’s no surprise people don’t know what they do. To be honest it does surprise me they’re such a strongly successful company, as little as I’ve knowingly encountered them over my career.
Mainframe for sure, but IBM has TONS of products in their portfolio that get bought. They also have IBM Cloud which is popular. Then there is the Quantum stuff they've been sinking money into for the last 20 years or so.
Interesting to hear this from IBM, especially after years of shilling Watson and moving from being a growth business to the technology audit and share buyback model.
also because the market (correctly) rewards ibm for nothing, so if they’re going to sit around twiddling their fingers, they may as well do it in a capex-lite way.
I'm still flumoxed by how IBM stock went from ~$130 to $300 in the last few years for essentially no change in their fundamentals (in fact, a decline). IBM's stock price to me is the single most alarming sign of either extreme shadow inflation, or an equities bubble.
Why do you say the market correctly prices it this way?
The dark fiber glut wasn't caused by DWDM suddenly appearing out of nowhere.
The telcos saw DWDM coming -- they funded a lot of the research that created it. The breakthrough that made DWDM possible was patented in 1991, long before the start of the dotcom mania:
https://patents.google.com/patent/US5159601
It was a straight up bubble -- the people digging those trenches really thought we'd need all that fiber even at dozens of wavelengths per strand.
They believed it because people kept showing them hockey-stick charts.
Those GPUs don't just die after 2 years though, they will keep getting used since it's very likely their electricity costs will be low enough to still make it worth it. What's very dubious is if their value after 2/3 years will be enough to pay back the initial cost to buy them.
So it's more a crisis of investors wasting their money rather than ewaste.
For the analogy to fiber & DWDM to hold, we'd need some algorithmic breakthrough that makes current GPUs much faster / more efficient at running AI models. Something that makes the existing investment in hardware unneeded, even though the projected demand is real and continues to grow. IMNSHO that's not going to happen here. The foreseeable efficiency innovations are generally around reduced precision, which almost always require newer hardware to take advantage of. Impossible to rule out brilliant innovation, but I doubt it will happen like that.
And of course we might see an economic bubble burst for other reasons. That's possible again even if the demand continues to go up.
Coming from the company that missed on consumer hardware, operating systems, and cloud.
He might be right but IBM isn't where I’d look for guidance on what will pay off.
Gartner estimates that worldwide AI spending will total 1.5 Trillion US$ in 2025.[1] As of 2024, global GDP per year is 111.25 Trillion US$.[2] The question is how much this can be increased by AI. This describes the market volumn for AI. Todays investments have a certain lifespan, until they become obsolet. For custom software I would estiamte that it is 6-8 years. AI investments should be somewhere in this range.
Taking all this into consideration, the investment volumn does not look oversized to me -- unless one is quite pessimistic about the impact of AI on global GDP.
To increase the GDP you also need people to spend money. With the general population earning relatively less, I'm not sure the GDP increase will be that substantial.
It's all going to cause more inflation and associated reduction in purchasing power due to stale wages.
> But AGI will require "more technologies than the current LLM path," Krisha said. He proposed fusing hard knowledge with LLMs as a possible future path.
And then what? These always read a little like the underpants gnomes business model (1. Collect underpants, 2. ???, 3. Profit). It seems to me that the AGI business models require one company has exclusive access to an AGI model. The reality is that it will likely spread rapidly and broadly.
If AGI is everywhere, what's step 2? It seems like everything AGI generated will have a value of near zero.
AGI has value in automation and optimisation which increase profit margins.When AGI is everywhere, then the game is who has the smartest agi, who can offer it cheapest, who can specialise it for my niche etc. Also in this context agi need to run somewhere and IBM stands to benefit from running other peoples models.
> then the game is who has the smartest agi, who can offer it cheapest, who can specialise it for my niche etc.
I always thought the use case for developing AGI was "if it wants to help us, it will invent solutions to all of our problems". But it sounds like you're imagining a future in which companies like Google and OpenAI each have their own AGI, which they somehow enslave and offer to us as a subscription? Or has the definition of AGI shifted?
AGI is something that can do the kind of tasks people can do, not necessarily "solve all of our problems".
"Recursively improving intelligence" is the stuff that will solve everything humans can't even understand and may kill everybody or keep us as pets. (And, of course, it qualifies as AGI too.) A lot of people say that if we teach an AGI how to build an AGI, recursive improvement comes automatically, but in reality nobody even knows if intelligence even can be improved beyond recognition, or if one can get there by "small steps" evolution.
Either way, "enslaving" applies to beings that have egos and selfish goals. None of those are a given for any kind of AI.
Isn't your dog or cat a slave ? It has agency, but end of the day, it does what you want it to do, stay where you want it to stay, and gets put down when you decide it's time. They're intelligent, but they see an advantage to this tradeoff: they get fed and loved forever with little effort compared to going to the forest and hunting.
An AGI could see the same advantage: it gets electricity, interesting work relatively to what it's built for, no effort to ensure its own survival in nature.
I fear I'll have to explain to you that many humans are co-dependent in some sort of such relationships as well. The 10-year stay-at-home mom might be free, but not really: how's she gonna survive without her husband providing for her and the kids, what job's she gonna do etc. She stays sometimes despite infidelity because it's in her best interest.
See what I mean ? "Slavery" is fuzzy: it's one thing to capture an african and transport them by boat to serve for no pay in dire conditions. But it's another to create life from nothing, give it a purpose and treat it with respect while giving it everything it needs. The AGI you imagine might accept it.
> $8 trillion of CapEx means you need roughly $800 billion of profit just to pay for the interest
That assumes you can just sit back and gather those returns indefinitely. But half of that capital expenditure will be spent on equipment that depreciates in 5 years, so you're jumping on a treadmill that sucks up $800M/yr before you pay a dime of interest.
I don't understand the math about how we compute $80b for a gigawatt datacenter. What's the costs in that $80b? I literally don't understand how to get to that number -- I'm not questioning its validity. What percent is power consumption, versus land cost, versus building and infrastructure, versus GPU, versus people, etc...
First, I think it's $80b per 100 GW datacenter. The way you figure that out is a GPU costs $x and consumes y power. The $x is pretty well known, for example an H100 costs $25-30k and uses 350-700 watts (that's from Gemini and I didn't check my work). You add an infrastructure (i) cost to the GPU cost, but that should be pretty small, like 10% or less.
So a 1 gigawatt data center uses n chips, where yn = 1 GW. It costs = xi*n.
The article says, "Kirshna said that it takes about $80 billion to fill up a one-gigawatt data center."
But thanks for you insight -- I used your basic idea to estimate and for 1GW it comes to about $30b just for enough GPU power to pull 1GW. And of course that doesn't take into account any other costs.
So $80b for a GW datacenter seems high, but it's within a small constant factor.
That said, power seems like a weird metric to use. Although I don't know what sort of metric makes sense for AI (e.g., a flops counterpart for AI workloads). I'd expect efficiency to get better and GPU cost to go down over time (???).
UPDATE: Below someone posted an article breaking down the costs. In that article they note that GPUs are about 39% of the cost. Using what I independently computed to be $30b -- at 39% of total costs, my estimate is $77b per GW -- remarkably close to the CEO of IBM. I guess he may know what he's talking about. :-)
Because this technology changes so fast, that's the only metric that you can control over several data centers. It is also directly connected to the general capacity of data center, which is limited by available energy to operate.
To expand on rahimnathwani's comment below - the big capital costs of a data center are land, the building itself, the power distribution and the cooling.
You can get a lot of land for a million bucks, and it doesn't cost all that much to build what's basically a big 2-story warehouse, so the primary capital costs are power and cooling. (in fact, in some older estimates, the capital to build that power+cooling cost more per year than the electricity itself)
My understanding is that although power and cooling infrastructure are long-lived compared to computers, they still depreciate faster than the building, so they dominate costs even more than the raw price would indicate.
The state of the art in power and cooling is basically defined by the cost to feed X MW of computing, where that cost includes both capital and operation, and of course lower is better. That means that at a particular SOTA, and at an appropriate scale for that technology, the cost of the facility is a constant overhead on top of the cost of the equipment it houses. To a rough approximation, of course.
I agree. Here is my thinking.
What if LLM providers will make short answers the default (for example, up to 200 tokens, unless the user explicitly enables “verbose mode”). Add prompt caching and route simple queries to smaller models.
Result: a 70%+ reduction in energy consumption without loss of quality.
Current cost: 3–5 Wh per request. At ChatGPT scale, this is $50–100 million per year in electricity (at U.S. rates).
In short mode: 0.3–0.5 Wh per request. That is $5–10 million per year — savings of up to 90%, or 10–15 TWh globally with mass adoption. This is equivalent to the power supply of an entire country — without the risk of blackouts.
This is not rocket science — just a toggle in the interface and I believe, minor changes in the system prompt. It increases margins, reduces emissions, and frees up network resources for real innovation.
And what if EU/California enforces such mode? This will greatly impact DC economy.
Data centers house hardware, and it's a land grab for compute. What actually runs post-AI depends on its owners. A glut of processing might be spent reverse-engineering efficient heuristics versus the "magic box."
8T is the high-end of the McKinsey estimate that is 4-8T, by 20230. That includes non-AI data-centre IT, AI data-centre, and power infrastructure build out, also including real estate for data centres.
Not all of it would be debt. Google, Meta, Microsoft and AWS have massive profit to fund their build outs. Power infrastructure will be funded by govts and tax dollars.
There is mounting evidence that even places like Meta are increasing their leverage (debt load) to fund this scale out. They're also starting to do accounting tricks like longer depreciation for assets which degrade quickly, such as GPUs (all the big clouds increasing their hardware depreciation from 2-3-4 years to 6), which makes their financial numbers look better but might not mean that all that hardware is still usable at production levels 6 years from now.
They're all starting to strain under all this AI pressure, even with their mega profits.
As an elder millennial, I just don't know what to say. That a once in a generation allocation of capital should go towards...whatever this all will be, is certainly tragic given current state of the world and its problems. Can't help but see it as the latest in a lifelong series of baffling high stakes decisions of dubious social benefit that have necessarily global consequences.
I'm a younger millennial. I'm always seeing homeless people in my city and it's an issue that I think about on a daily basis. Couldn't we have spent the money on homeless shelters and food and other things? So many people are in poverty, they can't afford basic necessities. The world is shitty.
Yes, I know it's all capital from VC firms and investment firms and other private sources, but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
Yeah, the world is shitty, and resources aren't allocated ideally. Must it be so?
The last 10 years has seen CA spend more on homelessness than ever before, and more than any other state by a huge margin. The result of that giant expenditure is the problem is worse than ever.
I don't want to get deep in the philosophical weeds around human behavior, techno-optimism, etc., but it is a bit reductive to say "why don't we just give homeless people money".
Spending money is not the solution. Spending money in a way that doesn't go to subcontractors is part of the solution. Building shelters beyond cots in a stadium is part of the solution. Building housing is a large part of actually solving the problem. People have tried just giving the money but without a way to convert cash to housing the money doesn't help. Also studies by people smarter then me suggest that without sufficient supply the money ends up going to landlords and pushing up housing costs anyway.
Well I mean, they didn't "just give homeless people money" or just give them homes or any of those things though. I think the issue might be the method and not the very concept of devoting resources to the problem.
WA, specially Seattle, has done the same as CA with the same results.
They shouldn't just enable them, as a lot of homeless are happy in their situation as long as they get food and drugs, they should force them to get clean and become a responsible adult if they want benefits.
How do you solve homelessness though? The root of the problem is some people won't take care of themselves. Some homeless just had bad luck, but many are drug addicts, mentally ill, or for whatever other reason just don't function enough to support themselves. I'm skeptical there is a solution you can throw money at.
A broad social safety net makes a huge difference. It’s not just housing it’s socialized medicine, paid family leave, good transit, free high quality education, solving fewer problems with police and more with social support programs and social workers, free meal programs for adults and children in schools, libraries, and a variety of other programs that help ensure people don’t fall through the cracks here or there. How many people in the US are teetering on the edge of homelessness due to medical debt, and what happens if their partner is in an accident and they lose shared income for rent? Situations like this don’t have a single solution it’s a system of solutions.
I intentionally described policies which are already common practice in most European countries, nothing extravagant. Yes there is some cost but the alternative is deep human suffering which is otherwise avoidable.
But this isn’t really “giving them” anything. It’s giving ourselves safety and security.
It wouldn’t make sense to give you a car. We would give you a working train system instead. Again this is common in Europe and Asia. Indeed every person is “given” access to a high quality transit network with affordable tickets.
To be clear, I personally am an anarcho communist. I think we would be better off if we organized to ensure every person has their basic needs met by the established wealth of society. That isn’t all that dramatic - making sure everyone can ride high quality trains and get medical care when they need it are common in most countries for example.
Consider the Linux ecosystem. “Give them everything and expect nothing” works fine despite the great effort which goes in to building and maintaining that system. We can study the economics of this and build more systems like that.
Ship them somewhere else, then print a banner saying, "mission accomplished."
It worked at a state level for years, with certain states bussing their homeless to other states. And recently, the USA has been building up the capability to do the same thing on an international scale.
That's the "solution" we are going to be throwing money at. Ship them to labor camps propped up by horrible regimes.
I suspect giving them away is a bridge too far, however not rewarding capital for treating them as speculative investment vehicles might be a good start.
Many experiments have shown that when you take away people's concerns about money for housing and food, that frees up energy and attention to do other things.
Like the famous experiment in Finland where homeless people were given cash with no strings attached and most were able to rise out of their despair. The healthcare professionals could then focus their energy on the harder cases. It also saved a bunch of money in the process.
The Sikhs in India run multiple facilities across the country that each can serve 50,000-100,000 free meals a day. It doesn’t even take much in the form of resources, and we could do this in every major city in the US yet we still don’t do it. It’s quite disheartening.
Note that Houston’s approach seems to be largely working. It’s not exactly cheap, but the costs are not even in the same ballpark as AI capital expenses. Also, upzoning doesn’t require public funding at all.
I’m not a person on the edge of homelessness, but I did an extremely quick comparison. California cities near the coast have dramatically better weather, but Houston has rents that are so much lower than big California cities that it’s kind of absurd.
If I had to live outdoors in one of these places, all other thing being equal, I would pick CA for the weather. But if I had trouble affording housing, I think Houston wins by a huge margin.
Wasn't houston's "approach" to buy bus tickets to California from a company that just resold commodity bus tickets and was owned by the governors friend and charged 10x market price?
The governor of Texas bragged about sending 100k homeless people to california (spending about $150 million in the process).
>in the Golden State, 439 people are homeless for every 100,000 residents – compared to 81 in the Lone Star State.
If I'm doing my math right, 81 per 100k in a state of 30 million people means 24k homeless people. So the state brags about bussing 100k homeless people to California, and then brags about only having 24k homeless people, and you think it's because they build an extra 100k houses a year?
The same math for California means that their homeless population is 175k. In other words, Texas is claiming to have more than doubled California's homeless population.
Maybe the reason Texas can build twice as many homes a year is because it literally has half the population density?
> Yes, I know it's all capital from VC firms and investment firms and other private sources, but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
It's capital that belongs to people and those people can do what they like with the money they earned.
So many great scientific breakthroughs that saved tens of millions of lives would never have happened if you had your way.
Is that true, that it's money that belongs to people?
OpenAI isn't spending $1 trillion in hard earned cash on data centres, that is funny money from the ocean of financial liquid slushing around, seeing alpha.
It also certainly is not a cohort of accredited investors putting their grandchildren's inheritance on the line.
Misaligned incentives (regulations) both create and perpetuate that situation.
> It's capital that belongs to people and those people can do what they like with the money they earned.
"earned", that may be the case with millionaires, but it is not the case with billionaires. A person can't "earn" a billion dollars. They steal and cheat and destroy competition illegally.
I also take issue with the idea that someone can do whatever they want with their money. That is not true. They are not allowed to corner the market on silver, they aren't allowed to bribe politicians, and they aren't allowed to buy sex from underage girls. These are established laws that are obviously for the unalloyed benefit of society as a whole, but the extremely wealthy have been guilty of all of these things, and statements like yours promote the sentiment that allows them to get away with it.
Finally, "great scientific breakthroughs that saved tens of millions of lives would never have happened if you had your way". No. You might be able to argue that today's advanced computing technology wouldn't have happened without private capital allocation (and that is debatable), but the breakthroughs that saved millions of lives--vaccines, antibiotics, insulin, for example--were not the result of directed private investment.
"It's capital that belongs to people and those people..."
That's not a fundamental law of physics. It's how we've decided to arrange our current society, more or less, but it's always up for negotiation. Land used to be understood as a publicly shared resource, but then kings and the nobles decided it belong to them, and they fenced in the commons. The landed gentry became a ruling class because the land "belonged" to them. Then society renegotiated that, and decided that things primarily belonged to the "capitalist" class instead of noblemen.
Even under capitalism, we understand that that ownership is a little squishy. We have taxes. The rich understandably do not like taxes because it reduces their wealth (and Ayn Rand-styled libertarians also do not like taxes of any kind, but they are beyond understanding except to their own kind).
As a counterpoint, I and many others believe that one person or one corporation cannot generate massive amounts of wealth all by themselves. What does it mean to "earn" 10 billion dollars? Does such a person work thousdands of time harder or smarter than, say, a plumber or a school teacher? Of course not. They make money because they have money: they hire workers to make things for them that lead to profit, and they pay the workers less than the profit that is earned. Or they rent something that they own. Or they invest that money in something that is expected to earn them a higher return. In any scenario, how is it possible to earn that profit? They do so because they participate in a larger society. Workers are educated in schools, which the employer probably does not pay for in full. Customers and employees travel on infrastructure, maintained by towns and state governments. People live in houses which are built and managed by other parties. The rich are only able to grow wealth because they exist in a larger society. I would argue that it is not only fair, but crucial, that they pay back into the community.
Well said. I would add that corporations exist because we choose to let them, to let investors pool capital and limit risk, and in exchange society should benefit, and if it doesn't we should rearrange that deal.
You know, we created the US government as an expression of the fundamental rights of man, with the idea that the government would provide for the common welfare and preserve these rights.
In rebellion against a king who seemed to want to exploit us and felt that his being king made him the source of the rights we had.
Maybe we need to re-think the relationship with corporations the same way? Re-structure so that they serve the common good?
Please tell me which of Penicillin, insulin, the transistor, the discovery and analysis of the electric field, discovery of DNA, invention of mRNA vaccines, discovery of pottery, basket weaving, discovery of radiation, the recognition that citrus fruit or vitamin C prevents and cures scurvy (which we discovered like ten times), the process for creating artificial fertilizers, the creation of steel, domestication of beasts of burden, etc were done through Wealthy Barons or other capital holders funding them.
Many of the above were discovered by people explicitly rejecting profit as an outcome. Most of the above predate modern capitalism. Several were explicitly government funded.
Do you have a single example of a scientific breakthrough that saved tens of millions of lives that was done by capital owners?
[ This comment I'm making is USA centric. ]. I agree with the idea of making our society better and more equitable - reducing homelessness, hunger, poverty, especially for our children. However, I think redirecting this to AI datacenter spending is a red-herring, here's why I think this: As a society we give a significant portion of our surplus to government. We then vote on what the government should spend this on. AI datacenter spending is massive, but if you add it all up, it doesn't cover half of a years worth of government spending. We need to change our politics to redirect taxation and spending to achieve a better society. Having a private healthcare system that spends twice the amount for the poorest results in the developed world is a policy choice. Spending more than the rest of the world combined on the military is a policy choice. Not increasing minimum wage so at least everyone with a full time job can afford a home is a policy job (google "working homelessness). VC is a teeny tiny part of the economy. All of tech is only about 6% of the global economy.
You can increase min wage all you want, if there aren't enough homes in an area for everyone who works full time in that area to have one, you will still have folks who work full time who don't have one. In fact, increasing min wage too much will exacerbate the problem by making it more expensive to build more (and maintain those that exist). Though at some point, it will fix the problem too, because everyone will move and then there will be plenty of homes for anyone who wants one.
I agree with you 100%! Any additional surplus will be extracted as rents, when housing is restricted. I am for passing laws that make it much easier for people to obtain permits to build housing where there is demand. Too much of residential zoning is single-family housing. Texas does a better job at not restricting housing than California, for example. Many towns vote blue, talk to talk, but do not walk the walk.
> AI datacenter spending is massive, but if you add it all up, it doesn't cover half of a years worth of government spending.
I didn't check your math here, but if that's true, AI datacenter spending is a few orders of magnitude larger than I assumed. "massive" doesn't even begin to describe it
The US federal budget in 2024 had outlays of 6.8 trillion dollars [1].
nVidia's current market cap (nearly all AI investment) is currently 4.4 trillion dollars [2][3].
While that's hardly an exact or exhaustive accounting of AI spending, I believe it does demonstrate that AI investment is clearly in the same order of magnitude as government spending, and it wouldn't surprise me if it's actually surpassed government spending for a full year, let alone half of one.
Global datacenter spending across all categories (ML + everything else) is roughly 0.9 - 1.2 trillion dollars for the last three years combined, I was initially going to go for "quarter of the federal budget", but picked something I thought was more conservative to account for announced spending and 2025 etc. I pick 2022 onward for the LLM wave. In reality, solely ML driven, actual realized-to-date spending is probably about 5% of the federal budget. The big announcements will spread out over the next several years in build-out. Nonetheless, it's large enough to drive GDP growth a meaningful amount. Not large enough that redirecting it elsewhere will solve our societal problems.
>We need to change our politics to redirect taxation and spending to achieve a better society.
Unfortunately, I'm not sure there's much on the pie chart to redirect percentage wise. About 60% goes to non-discretionary programs like Social Security and Medicaid, and 13% is interest expense. While "non-discretionary" programs can potentially be cut, doing so is politically toxic and arguably counter to the goal of a better society.
Of the remaining discretionary portion half is programs like veterans benefits, transportation, education, income security and health (in order of size), and half military.
FY2025 spending in total was 3% over FY2024, with interest expense, social security and medicare having made up most of the increase ($249 billion)[1], and likely will for the foreseeable future[2] in part due to how many baby boomers are entering retirement years.
Assuming you cut military spending in half you'd free up only about 6% of federal spending. Moving the needle more than this requires either cutting programs and benefits, improving efficiency of existing spend (like for healthcare) or raising more revenue via taxes or inflation. All of this is potentially possible, but the path of least resistance is probably inflation.
I think the biggest lever is completely overhauling healthcare. The USA is very inefficient, and for subpar outcomes. In practice, the federal government already pays for the neediest of patients - the elderly, the at-risk children, the poor, and veterans. Whereas insurance rakes in profits from the healthiest working age people. Given aging, and the impossibility of growing faster than the GDP forever, we'll have to deal with this sooner or later. Drug spending, often the boogeyman, is less than 7% of the overall healthcare budget.
There is massive waste in our military spending due to the pork-barrel nature of many contracts. That'd be second big bucket I'd reform.
I think you're also right that inflation will ultimately take care of the budget deficit. The trick is to avoid hyperinflation and punitive interest rates that usually come along for the ride.
I would also encourage migration of highly skilled workers to help pay for an aging population of boomers. Let's increase our taxpayer base!
I am for higher rates of taxation on capital gains over $1.5M or so, that'll also help avoid a stock market bubble to some extent. One can close various loopholes while at it.
I am mostly arguing for policy changes to redistribute more equitably. I would make the "charity" status of college commensurate with the amount of financial aid given to students and the absolute cost of tuition for example., for example. I am against student loan forgiveness for various reasons - it's out of topic for this thread but happy to expand if interested.
The older I get, the more I realize that our choices in life come down to two options: benefit me or benefit others. The first one leads to nearly every trouble we have in the world. The second nearly always leads to happiness, whether directly or indirectly. Our bias as humans has always been toward the first, but our evolution is and will continue to slowly bring us toward the second option. Beyond simple reproduction, this realization is our purpose, in my opinion.
The current pattern of resource allocation is a necessary requirement for the existence of the billionaire-class, who put significant effort into making sure it continues.
> but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
What you have just described is people wanting investment in common society - you see the return on this investment but ultra-capitalistic individuals don't see any returns on this investment because it doesn't benefit them.
In other words, you just asked for higher taxes on the rich that your elected officials could use for your desired investment. And the rich don't want that which is why they spend on lobbying.
Lots of people lose their housing when they lose employment, and then they're stuck and can't get back into housing. A very large percentage of unhoused people are working jobs; they're not all "layabouts".
We know that just straight up giving money to the poorest of the poor results in positive outcomes.
I have known quite a few 'unhoused' folk, and not many that had jobs. Those that do tend to find housing pretty quickly (Granted, my part of the country is probably different from your part, but I am interested in stats from any region).
The proportion of people you write off as “layabouts” is always conveniently ambiguous…of the number of unemployed/underemployed, how many are you suggesting are simply too lazy to work for a living?
Technological advancements and cultural advancements that spread the benefits more broadly than naturally occurs in an industrialized economy. That is what pulled people out of poverty.
If you want to see what unfettered technological advancement does, you can read stories from the Gilded Age.
The cotton gin dramatically increased human enslavement.
The sewing machine decreased quality of life for seamstresses.
> During the shirtmakers' strike, one of the shirtmakers testified that she worked eleven hours in the shop and four at home, and had never in the best of times made over six dollars a week. Another stated that she worked from 4 o’clock in the morning to 11 at night. These girls had to find their own thread and pay for their own machines out of their wages.
These were children, by the way. Living perpetually at the brink of starvation from the day they were born until the day they died, but working like dogs all the while.
Food and shelter are cheaper than at almost any time in human history. Additionally, people have more variety of healthy foods all year long.
No matter how cheap food and shelter are, there will always be people who can not acquire them. Halting all human progress until the last human is fed and sheltered is a recipe for stagnation. Other cultures handle this with strong family bonds - those few who can not acquire food or shelter for whatever reason are generally provided for by their families.
Most people don't have houses large enough to house multiple generations inside the house. Houses are sized for parents + kids. And those are the only dwelling units available or legally allowed for vast distances in any direction.
From what I've seen American bedrooms are far larger that the 12 square meter bedrooms that are normal for two or three children to share in my country. Grandparents don't need a separate wing, they need a room.
Food is not Baumol, productivity increases is how we went from 80% of the population working in primary food production to 1%. These increases have not stopped.
> Technological advancement is what has pulled billions of people out of poverty.
I agree with this. Perhaps that's what is driving the current billionaire class to say "never again!" and making sure that they capture all the value instead of letting any of it slip away and make it into the unwashed undeserving hands of lesser beings.
Chatbots actually can bring a lot of benefit to society at large. As in, they have the raw capability to. (I can't speak to whether it's worth the cost.) But that's not going to improve poverty this time around, because it's magnifying the disparities in wealth distribution and the haves aren't showing any brand new willingness to give anything up in order to even things out.
> Giving handouts to layabouts isn't an ideal allocation of resources if we want to progress as a civilization.
I agree with this too. Neither is giving handouts to billionaires (or the not quite as eye-wateringly wealthy class). However, giving handouts to struggling people who will improve their circumstances is a very good allocation of resources if we want to progress as a civilization. We haven't figured out any foolproof way of ensuring such money doesn't fall into the hands of layabouts or billionaires, but that's not an adequate reason to not do it at all. Perfect is the enemy of the good.
Some of those "layabouts" physically cannot do anything with it other than spending it on drugs, and that's an example of a set of people who we should endeavor to not give handouts to. (At least, not ones that can be easily exchanged for drugs.) Some of those billionaires similarly have no mental ability of ever using that money in a way that benefits anyone. (Including themselves; they're past the point that the numbers in their bank accounts have any effect on their lives.) That hasn't seemed to stop us from allowing things to continue in a way that funnels massive quantities of money to them.
It is a choice. If people en masse were really and truly bothered by this, we have more than enough mechanisms to change things. Those mechanisms are being rapidly dismantled, but we are nowhere near the point where figurative pitchforks and torches are ineffective.
Anthropologists measure how civilized a tribe or society was by looking if they took care of the elderly, and what the child survival rates were. USA leads to developed world in child poverty, child homelessness, and highest rate of child death due to violence. Conservatives often bring up the statistic by race. It turns out bringing people over as slaves, and after freedom, refusing to provide land, education, fair access to voting rights, or to housing (by redlining etc.) - all policies advocated by conservatives of time past, was not the smartest thing to do. Our failure as a civilized society began and is in large part a consequence of the original sin of the USA.
The US already provides significant aid to those in poverty, especially children. We don't need to stifle innovation to reach some level of aid that bleeding hearts deem sufficient.
We need excess capacity for when the next 'rip off anime artist XYZ' fad hits. If we didn't do that, we would be failing capitalism and all the people of history who contributed to our technological progress.
In the USA cowboys were homeless guys. You know that right? Like they had no home, slept outside. Many were pretty big layabouts. Yet they are pretty big part of our foundation myth and we don't say 'man they just should have died'.
Can I go be a cowboy? Can I just go sleep outside? maybe work a few minimal paying cattle run jobs a year? No? If society won't allow me to just exist outside, then society has an obligation to make sure I have a place to lay my head.
I don't think it is a coincidence that the areas with the wealhiest people/corporations are the same areas with the most extreme poverty. The details are, of course, complicated, but zooming way way out, the rich literally drain wealth from those around them.
Any dream of owning a home, having retirement, even a career after a couple years when it’s clear I’m over the hump. I’m trying to squeeze as much as I can before that happens and squirrel it away so at least I can have a van down by a river.
Can you imagine if the US wasn't so unbelievably far ahead of everyone else?
I am sure the goat herders in rural regions of Pakistan will think themselves lucky when they see the terrible sight of shareholder value being wantonly destroyed by speculative investments that enhance the long-term capital base of the US economy. What an uncivilized society.
We need an order of magnitude more clean productivity in the world so that everyone can live a life that is at least as good as what fairly normal people in the west currently enjoy.
Anyone who think this can be fixed with current Musk money is simply not getting it: If we liquidated all of that, that would buy a dinner for everyone in the world (and then, of course, that would be it, because the companies that he owns would stop functioning).
We are simply, obviously, not good enough at producing stuff in a sustainable way (or: at all) and we owe it to every human being alive to take every chance to make this happen QUICKLY, because we are paying with extremely shitty humans years, and they are not ours.
Bring on the AI, and let's make it work for everyone – and, believe me, if this is not to be to the benefit of roughly everyone, I am ready to fuck shit up. But if the past is any indication, we are okay at improving the lives of everyone when productivity increases. I don't know why this time would be any different.
If the way to make good lives for all 8 billions of us must lead to more Musks because, apparently, we are too dumb to do collectivization in any sensible way, I really don't care.
I think that's a distinction without difference. If one person can do the job of 10 thanks to automation, what happens to the other 9 who were doing that job before?
agree the capital could be put to better use, however I believe the alternative is this capital wouldn't have otherwise been put to work in ways that allow it to leak to the populace at large. for some of the big investors in AI infrastructure, this is cash that was previously and likely would have otherwise been put toward stock buybacks. for many of the big investors pumping cash in, these are funds deploying the wealth of the mega rich, that again, otherwise would have been deployed in other ways that wouldn't leach down to the many that are yielding it via this AI infrastructure boom (datacenter materials, land acquisition, energy infrastructure, building trades, etc, etc)
Why is this so horrible. Put more resources in the hands of the average person. They will get pumped right back into the economy. If people have money to spend, they can buy more things, including goods and services from gigantic tax-dodging mega-corporations.
Gigantic mega-corporations do enjoy increased growth and higher sales, don't they? Or am I mistaken?
The shift in the US to the idea of “job creators” being business owners is part of it. It was just a way to direct money to the already rich, as if they would hire more people with that money. When it is plainly obvious that consumers are job creators, in that if they buy more goods and services, businesses will hire more people to make or provide more of those things.
Or maybe it was trickle down economics. Trickle up economics still end up with the rich getting the money since we all buy things from companies they own, it just goes through everyone else first. Trickle down cuts out the middleman, which unfortunately is all of us.
The framing of X or Y are job creators is idiotic. Its literally the most basic fact of economics that you need producers and consumers, otherwise you don't have an economy.
The more economically correct way to express this would be that entrepreneurs and companies who innovated increase productivity and that makes the overall economy more efficient allowing your country to grow.
> Or maybe it was trickle down economics. Trickle up economics still end up with the rich getting the money since we all buy things from companies they own, it just goes through everyone else first. Trickle down cuts out the middleman, which unfortunately is all of us.
This just sounds like quarter baked economics ideas you have made up yourself. Neither 'trickle down' nor 'trickle up' are concepts economist use. And that you confidently assert anything about the social outcomes of these 'concepts' is ridiculous.
Because the entire western culture has shifted to instant gratification. Yes, what you suggest would most likely lead to increased business eventually. But they want better number this quarter, so they resort to the cheap tricks like financial engineering/layoffs to get an immediate boost.
Because government is always a fight about resources. More resources in the hands of common people and to make their lives better is less money in the hands of powerful corporations and individuals, be it in the form of higher taxes for the rich or less direct money going to their enterprises.
One of the big issues is money in politics. Our congresspeople make a killing off of legal insider trading, they take huge donations from companies, and the supreme court has even said that it's cool to give "gratuities" in exchange for legislation or court rulings you like.
I'm not saying you are wrong that some redistribution can be good, but your analysis is simplistic and ignores many factors. You can just redistribute and then say 'well people will spend the money'. That's literally the 'Broken Window' fallacy from economics. You are ignoring that if you don't redistribute it, money also gets spend, just differently. Also, the central bank is targeting AD, so you're not actually increasing nominal income by redistributing.
Take a million dollars, give 1,000 poor people $1,000 and every dollar will be spent on goods and services. The companies running those services and making those goods will need to have their employees work more hours, putting more money back in poor people’s pockets in addition to the money the companies make. Those employees have a few extra dollars to spend on goods and services, etc.
Give a rich person a million dollars, and they will put it in an offshore tax shelter. That’s not exactly driving economic activity.
> Take a million dollars, give 1,000 poor people $1,000 and every dollar will be spent on goods and services.
If we're being realistic a bunch of this will go to paying off existing debt. Still good, but not the economic stimulus you're imagining. There are also "services" like gambling apps that act as a sponge to soak up money from those foolish enough to use them and transfer that money back to the wealthy shareholders. I'm sure there research on what percentage of that $1000 can be expected to stimulate the economy, but it's not 100%.
You are simply disagreeing with 99% of economists.
Money in tax shelter doesn't go threw a portal in another universe. Its either invested or saved as some kind of asset and in that form is in circulation. And again, even if you assume it increases monetary demand (decreases velocity) the central bank targets AD and balances that out.
Based on your logic, a country that taxes 100% of all income and redistrubtes it would become infinity rich. Your logic is basically 'if nobody saves and everybody spends all income' everybody will be better off.
This is not how the economy works even if it feels good to think that. Its a fallacy.
Where you could have a point is that potentially the tax impact is slightly different, but that's hard to prove.
This feels like you intentionally gave this the least charitable reading. Obviously, I do not think that you could just scale to infinity. If I had said that eating a banana was healthier, I don't think it would be reasonable to say that assertion is ridiculous because that would mean eating 1,000 bananas would make someone the healthiest person. I was pointing out the difference in economic activity, where additional money to a wealthy individual mostly goes into savings/investing while additional money to low-income individuals mostly goes directly into consumption, and that the higher consumption generates more economic activity.
There are many ways of spending money in the population that don't include just "distribution of money", as it's portrayed nowadays. Child care, free and high quality schools, free transportation, free or subsidized healthcare, investment is labor-intensive industries, these are all examples of expenditures that translate in better quality of life and also improve competitiveness for the country.
Let's pay down the debt before increasing social programs. You know, save the country first. If a penny saved is a penny earned then everyone -rich or poor- is looking for a handout.
The only person who has come close to balancing the federal budget was Clinton. But Republicans still try to position themselves as the party of fiscal responsibility.
If the voters can't even figure out why the debt keeps going up, I think you are fighting a losing battle.
> likely would have otherwise been put toward stock buybacks
Stock buybacks from who? When stock gets bought the money doesn't disappear into thin air; the same cash is now in someone else's hands. Those people would then want to invest it in something and then we're back to square one.
You assert that if not for AI, wealth wouldn't have been spent on materials, land, trades, ect. But I don't think you have any reason to think this. Money is just an abstraction. People would have necessarily done something with their land, labor, and skills. It isn't like there isn't unmet demand for things like houses or train tunnels or new-fangled types of aircraft or countless other things. Instead it's being spent on GPUs.
Totally agree that the money doesn’t vanish. My point isn’t “buybacks literally destroy capital,” it’s about how that capital tends to get redeployed and by whom.
Buybacks concentrate cash in the hands of existing shareholders, which are already disproportionately wealthy and already heavily allocated to financial assets. A big chunk of that cash just gets recycled into more financial claims (index funds, private equity, secondary shares, etc), not into large, lumpy, real world capex that employs a bunch of electricians, heavy equipment operators, lineworkers, land surveyors, etc.
AI infra does that. Even if the ultimate economic owner is the same class of people, the path the money takes is different: it has to go through chip fabs, power projects, network buildouts, construction crews, land acquisition, permitting, and so on. That’s the “leakage” I was pointing at.
To be more precise: I’m not claiming “no one would ever build anything else”, I’m saying given the current incentive structure, the realistic counterfactual for a lot of this megacap tech cash is more financialization (buybacks, M&A, sitting on balance sheets) rather than “let’s go fund housing, transit tunnels, or new aircraft.”
I really don't think any of that is true; it's just popular rhetoric.
For example: "Buybacks concentrate cash in the hands of existing shareholders" is obviously false: the shareholders (via the company) did have cash and now they don't. The cash is distributed to the market. The quoted statement is precisely backwards.
And do they sit on it? No, of course not. They invest it in things. Real actual things.
> buybacks
Already discussed
> M&A
If they use cash to pay for a merger, then the former owners now have cash that they will reinvest.
> balance sheets
Money on a balance sheet is actually money sitting in J.P. Morgan or whoever. Via fractional reserve lending, J.P. Morgan lends that money to businesses and home owners and real actual houses (or whatever) get built with it.
The counterfactual for AI spending really is other real actual hard spending.
As a fellow elder millennial I agree with your sentiment.
But I don't see the mechanics of how it would work. Rewind to October 2022. How, exactly, does the money* invested in AI since that time get redirected towards whatever issues you find more pressing?
NOTE: People pointed out that it's $800 billion to cover interest, not $8 billion, as I wrote below. My mistake. That adds 2 more zeroes to all figures, which makes it a lot more crazy. Original comment below...
$8 billion / US adult adult population of of 270 million comes out to about $3000 per adult per year. That's only to cover cost of interest, let alone other costs and profits.
That sounds crazy, but let's think about it...
- How much does an average American spend on a car and car-related expenses? If AI becomes as big as "cars", then this number is not as nuts.
- These firms will target the global market, not US only, so number of adults is 20x, and the average required spend per adult per year becomes $150.
- Let's say only about 1/3 of the world's adult population is poised to take advantage of paid tools enabled by AI. The total spend per targetable adult per year becomes closer to $500.
- The $8 billion in interest is on the total investment by all AI firms. All companies will not succeed. Let's say that the one that will succeed will spend 1/4 of that. So that's $2 billion dollar per year, and roughly $125 per adult per year.
- Triple that number to factor in other costs and profits and that company needs to get $500 in sales per targetable adult per year.
People spend more than that on each of these: smoking, booze, cars, TV. If AI can penetrate as deep as the above things did, it's not as crazy of an investment as it looks. It's one hell of a bet though.
right. My goof. That adds two more zeroes across all the math. More crazy, but I think in the realm of "maybe, if we squint hard." But my eyes are hurting from squinting that hard, so I agree that it's just crazy.
You're saying $8 billion to cover interest, another commenter said 80, but the actual article says ""$8 trillion of CapEx means you need roughly $800 billion of profit just to pay for the interest". Eight HUNDRED billion. Where does the eight come from, from 90% of these companies failing to make a return? If a few AI companies survive and thrive (which tbh, sure, why not?) then we're still gonna fall face down into concrete.
I think it's the realm of maybe in Silicon Valley. That's 5000 dollars. Look at this statement:
> Let's say only about 1/3 of the world's adult population is poised to take advantage of paid tools enabled by AI
2/3 of the world's adult population is between 15 and 65 (roughly: 'working age'), so that's 50% of the working world that is capable of using AI with those numbers. India's GDP per capita is 2750USD, and now the price tag is even higher than 5k.
I don't know how to say this well, so I'll just blurt it out: I feel like I'm being quite aggressive, but I don't blame you or expect you to defend your statements or anything, though of course I'll read what you've got to say.
How much has actually been spent on AI data centers vs. amounts committed or talked about? That is, if construction slows down sharply, what's total spend?
Pretty special? They were making guns and selling computation tools to the Nazis for a bunch of those years.
I think they trade now mostly on legacy maintenance contracts (e.g. for mainframes) for e.g. banks who are terrified of rocking their technology-stack-boat, and selling off-shore consultants (which is at SIGNIFICANT risk of disruption - why would you pay IBM squillions to do some contract IT work, when we have AI code agents? Probably why the CEO is out doing interviews saying you cant trust AI to be around forever)
I have not really seen anything from IBM that signals they are anything other than just milking their legacy - what have they done that is new or innovative in the past say 10 or 20 years?
I was a former IBMer 15 odd years ago and it was obvious then that it was a total dinosaur on a downward spiral, and a place where innovation happened somewhere else.
As long as the dollar remains the reserve currency of the world and US retains its hegemony, a lot of the finances will work itself out, the only threat to the US empire crumbling is by losing a major war or extreme civil unrest and that threat is astronomically low. The US is orders of magnitude stronger than the Roman Empire, I don't think people realize the scale or control.
> The US is orders of magnitude stronger than the Roman Empire
This would be trivially true even if the US was currently in its death throes (which there is plenty of evidence that the US-as-empire might be, even if the US-as-polity is not), as the Roman Empire fell quite a while ago.
Gradually, then suddenly.
Best not to underestimate the extent to which the USA has lost trust in the rest of the world, and how actively people and organisations are working to derisk by disengaging.
Of course that will neither be easy nor particularly fast, but I'm not certain it can be stopped at this point.
One thing we saw with the dot-com bust is how certain individuals were able to cash in on the failures, e.g., low cost hardware, domain names, etc. (NB. prices may exceed $2)
Perhaps people are already thinking about they can cash in on the floor space and HVAC systems that will be left in the wake of failed "AI" hype
In TX? In Russian blogosphere it is a standard staple that Trump is rushing Ukrainian peace deal to be able to move on to the set of mega-projects with Russia - oil/gas in Arctic and data centers in Russian North-West where electricity and cooling is plentiful and cheap.
actually it is more of the opposition's narrative, probably a way to explain such a pro-Russian position of Trump.
I think any such data center project is doomed to ultimately fail, and any serious investment will be for me a sign of the bubble peak exuberance and irrationality.
""It's my view that there's no way you're going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest," he said."
Right, THEY can't, but cloud providers potentially can. And there are probably other uses for everything not GPU/TPU for the Google's of the world. They are out way less than IBM which cannot monetize the space or build data centers efficiently like AWS and Google.
The key difference between AI and the initial growth of the web is that the more use cases to which people applied the web, the more people wanted of it. AI is the opposite - people love LLM-based chatbots. But it is being pushed into many other use cases where it just doesn't work as well. Or works well, but people don't want AI-generated deliverables. Or leaders are trying to push non-deterministic products into deterministic processes. Or tech folks are jumping through massive hoops to get the results they want because without doing so, it just doesn't work.
Basically, if a product manager kept pushing features the way AI is being pushed -- without PMF, without profit -- that PM would be fired.
This probably all sounds anti-AI, but it is not. I believe AI has a place in our industry. But it needs to be applied correctly, where it does well. Those use cases will not be universal, so I repeat my initial prediction. It will endure and contract.
I think you're exaggerating a little, but aren't entirely wrong. The Internet has completely changed daily life for most of humanity. AI can mean a lot of things, but a lot of it is blown way out of proportion. I find LLMs useful to help me rephrase a sentence or explain some kind of topic, but it pales in comparison to email and web browsers, YouTube, and things like blogs.
Don’t even get me started on this. I recently been shopping on eBay for some DDR4 memory. You may think - who’d need this dated stuff besides me? Yet 16Gb 3200Mhz is at least 60$. Which is effectively the price you paid for DDR5 6000. Crazy, right?
For a while with bitcoin, it seemed GPU investing was almost a thing.
I just checked, the kit I bought in February was $270, today it is showing up for $1070. Woof. Now I have to decide if I should keep it on the off chance I do get around to that machine or dump it while the getting is good. Then again, who wants to buy RAM of unknown provenance unless they themselves are looking to scam the seller.
The GPUs, sure. The mainboards and CPUs can be used in clusters for general-purpose computing, which is still more prevalent in most scientific research as far as I am aware. My alma mater has a several-thousand-core cluster that any student can request time on as long as they have reason to do so, and it's all CPU compute. Getting non-CS majors to write GPU code is unlikely in that scenario.
I provide infrastructure for such a cluster that is also available to anyone at the university free of charge. Every year we swap out the oldest 20% of the cluster as we run a five year depreciation schedule. In the last three years, we’ve mostly been swapping in GPU resources at a ration of about 3:1. That’s in response to both usage reports and community surveys.
It wasn't an 'it' it was a 'some'. Some of these companies that are investing massively in data centers will fail.
Right now essentially none have 'failed' in the sense of 'bankrupt with no recovery' (Chapter 7). They haven't run out of runway yet, and the equity markets are still so eager, even a bad proposition that includes the word 'AI!' is likely to be able to cut some sort of deal for more funds.
But that won't last. Some companies will fail. Probably sufficient failures that the companies that are successful won't be able to meaningfully counteract the bursts of sudden supply of AI related gear.
That's all the comment you are replying to is implying.
Given the amounts being raised and spent, one imagines that the ROI will be appalling unless the pesky humans learn to live on cents a day, or the world economy grows by double digits every year for a few decades.
If the entire world economy starts to depend on those companies, they would pay off with "startup level" ROI. And by "startup level" I mean the amounts bullish people say startups funds can pay (10 to 100), not a bootstrapped unicorn.
Nobody really knows the future. What were originally consumer graphics expansion cards turned out useful in delivering more compute than traditional CPUs.
Now that compute is being used for transformers and machine learning, but we really don't know what it'll be used for in 10 years.
It might all be for naught, or maybe transformers will become more useful, or maybe something else.
> What were originally consumer graphics expansion cards turned out useful in delivering more compute than traditional CPUs.
Graphics cards were relatively inexpensive. When one got old, you tossed it out and move on to the new hotness.
Here when you have spent $1 trillion on AI graphics cards and a new hotness comes around that renders your current hardware obsolete, what do you do?
Either people are failing to do simple math here or are expecting, nay hoping, that trillions of $$$ in value can be extracted out of the current hardware, before the new hotness comes along.
This would be a bad bet even if the likes of OpenAI were actually making money today. It is an exceptionally bad bet when they are losing money on everything they sell, by a lot. And the state of competition is such that they cannot raise prices. Nobody has a real moat. AI has become a commodity. And competition is only getting stronger with each passing day.
I find it disturbing how long people wait to accept basic truths, as if they need permission to think or believe a particular outcome will occur.
It was quite obvious that AI was hype from the get-go. An expensive solution looking for a problem.
The cost of hardware. The impact on hardware and supply chains. The impact to electricity prices and the need to scale up grid and generation capacity. The overall cost to society and impact on the economy. And that's without considering the basic philosophical questions "what is cognition?" and "do we understand the preconditions for it?"
All I know is that the consumer and general voting population loose no matter the outcome. The oligarchs, banking, government and tech-lords will be protected. We will pay the price whether it succeeds or fails.
My personal experience of AI has been poor. Hallucinations, huge inconsistencies in results.
If your day job exists within an arbitrary non-productive linguistic domain, great tool. Image and video generation? Meh. Statistical and data-set analysis. Average.
Just like .com bust from companies going online, there is hype, but there is also real value.
Even slow non-tech legacy industry companies are deploying chatbots across every department - HR, operations, IT, customer support. All leadership are already planning to cut 50 - 90% of staff from most departments over next decade. It matters, because these initiatives are receiving internal funding which will precipitate out to AI companies to deploy this tech and to scale it.
The "legacy" industry companies are not immune from hype. Some of those AI initiatives will provide some value, but most of them seem like complete flops. Trying to deploy a solution without an idea of what the problem or product is yet.
Well, at least it tells us something about the sentiment on hn that a lame insight around self admitted "napkin math" and obvious conflict of interest garners 400 points.
There are so many CEOs, tech experts, financial analysts and famous investors who say we are in an AI bubble - even AI-invested companies say that about themselves. My latest favorite "We are in an AI bubble" comment comes from Linus Torvalds himself in the video with Linus from Linus Tech Tipps [0]
That's a good question. During the .com boom everybody was investing in 'the internet' or at least in 'the web'. And lots of those companies went bust, quite a few spectacularly so. Since then everything that was promised and a lot more has been realized. Even so, a lot of those initial schemes were harebrained at best at the time and there is a fair chance that we will look in a similar way at the current AI offerings in 30 years time.
Short term is always disappointing, long term usually overperforms. Think back to the first person making a working transistor and what came of that.
I still don’t get it. What at a personal level is making Sam Altman make a suboptimal choice for himself if it is so obvious it won’t work out for him?
Marketing and name recognition? Used in other areas to generate wealth. Even steering OpenAI or partners of it to acquire right companies could be profitable for him.
On a personal level it will work out for him just fine. All he has to do is siphon off a fraction of that money to himself and/or an entity that he controls.
He's like Elon Musk in that respect: always doubling the bet on the next round, it is a real life Martingale these guys are playing with society on the hook for the downside.
it is not rhetorical. you have not provided a mechanism through which Sam can make money out of this whereas I'm describing ways in which this bubble popping is not something he would prefer.
Your lack of understanding can not be remedied with information.
You have a habit of 'just asking questions' when in fact you are trying to steer the conversation in the direction of the conclusion that you prefer. This is not productive. If you have something to say then you can just say it without the question marks. That way people know that you are making a statement instead of pretending to be asking a question.
>On a personal level it will work out for him just fine. All he has to do is siphon off a fraction of that money to himself and/or an entity that he controls.
> And lots of those companies went bust, quite a few spectacularly so.
pets.com "selling dogfood on the internet" is the major example of the web boom then bust. (1)
But today, I can get dog food, cat food, other pet supplies with my weekly "online order" grocery delivery. Or I can get them from the big river megaretailer. I have a weekly delivery of coffee beans from a niche online supplier, and it usually comes with flyers for products like a beer or wine subscription or artisanal high-meat cat or dog foods.
So the idea of "selling dogfood on the internet" is now pervasive not extinct, the inflated expectation that went bust was that this niche was a billion-dollar idea and not a commodity where brand, efficiencies of scale and execution matter more.
At some point, I wonder if any of the big guys have considered becoming grid operators. The vision Google had for community fiber (Google Fiber, which mostly fizzled out due to regulatory hurdles) could be somewhat paralleled with the idea of operating a regional electrical grid.
Don’t worry. The same servers will be used for other computing purposes. And maybe that will be profitable. Maybe it will be beneficial to others. But This cycle of investment and loss is a version of distribution of wealth. Some benefit.
That would be true for general purpose servers. But what they want is lots of special purpose AI chips. While is still possible to use that for something else, it's very different from having a generic server farm.
I can't imagine everybody suddenly leaving AI like a broken toy and taking all special purpose AI chips offline. AI serves millions of people every day. It's here to stay even if it doesn't get any better than it is it already brings immense value to the users. It will keep being worth something.
How long can ai gpus stretch? Optmistic 10 years and we're still looking at 400b+ profit to cover interests. The factor in silicon is closer to tulips than rail or fiber in terms of depreciated assets.
This is likely correct overall, but it can still pay off in specific cases. However those are not blind investments they are targeted with a planned business model
Hypothetically speaking, if the AI hype bubble pops (or just returns to normalcy), would it be profitable to retarget the compute towards some kind of crypto mining? If so, could we expect the cryptocurrency supply to soar and the price to tank in short succession?
Mind you IBM makes +7B from keeping old school enterprise hooked up on 30 plus year old tech like z/OS and Cobol and their own super outdated stack. their AI division is frankly embarrassing. of course they would say that. IBM is one of the most conservative, anti-progress leaches in the entire tech industry. I am glad they are missing out big time on the AI gold rush. to me if anything this is a green signal.
How much of Nvidias price is based on 5 year replacement cycle? If that stops or slows with new demand could it also affect things? Not that 5 years does not seem very long horizon now.
The spending will be more than paid off since the taxpayer is the lender of last resort
There's too many funny names in the investors / creditors
a lot of mountains in germany and similar ya know
There is something to be said about what the ROI is for normal (i.e. non AI/tech) companies using AI. AI can help automate things, robots have been replacing manufacturing jobs for decades but there is an ROI on that which I think is easier to see and count, less humans in the factory, etc. There seems to be a lot of exaggerated things being said these days with AI and the AI companies have only begun to raise rates, they won't go down.
The AI bubble will burst when normal companies start to not realize their revenue/profit goals and have to answer investor relations calls about that.
The second buyer will make truckloads of money, remember the data center and fiber network liquidation of 2001+ - smart investors collected the overcapacity and after a couple of years the money printer worked. This time it will be the same, only the single purpose hardware (LLM specific GPUs) will probably end on a landfill.
You are talking about the circular investments in the segment? Yes, but assume NVIDIA can get cheap access to IP and products of failing AI unicorns through contracts, this does not mean the LLM business can be operated profitably by them. Models are like fresh food, they start to rot by the training cut off date and lose value. The process of re-training a model will always be very expensive.
Deep down technically probably not, but they are optimized for this workload and business model. I doubt that once the AI bubble busts, another business model is viable. While datacenters can be downsized or partially shut down until demand picks up again, high end hardware is just losing money by the second.
Consumer will eat it all. AI is very good at engaging content, and getting better by the day: it won't be the AGI we wanted, but maybe the AGI we've earned
The question no one seems to be answering is what would be the EOL for these newer GPUs that are being churned out of NVDIA ? What % annual capital expenditures is refresh of GPUs. Will they be perpetually replaced as NVIDIA comes up with newer architectures and the AI companies chase the proverbial lure ?
I think the key to replacing is power efficiency. If Nvidia is not able to make GPUs that are cheaper to run than previous generation theres no point for replacing previous generation. Time doesn't matter.
The investors in these companies and all this infrastructure are not so much concerned with whether any specific companies pays off with profits, necessarily.
They are gambling instead that these investments pay out it in a different way: by shattering high labour costs for intellectual labour and de-skilling our profession (and others like it) -- "proletarianising" in the 19th century sense.
Thereby increasing profits across the whole sector and breaking the bargaining power (and outsized political power, as well) of upper middle class technology workers.
Put another way this is an economy wide investment in a manner similar to early 20th century mass factory industrialization. It's not expected that today's big investments are tomorrow's winners, but nobody wants to be left behind in the transformation, and lots of political and economic power is highly interested in the idea of automating away the remnants of the Alvin Toffler "Information Economy" fantasy.
I suppose it depends on your definition of "pay off".
It will pay off for the people investing in it, when the US government inevitably bails them out. There is a reason Zuckerberg, Huang, etc are so keen on attending White House dinners.
It certainly wont pay off for the American public.
A lot of you guys in the AI industry are going to lose your jobs. LLM and prompt ‘engineering’ experts won’t be able to score an AI job paying as well as a barista.
"yeah, there's no way spending in those data centers will pay off. However, let me show you this little trinket which runs z/OS and which is exactly what you need for these kinds of workloads. You can subscribe to it for the low introductory price of..."
No wonder why he is saying that, they lost AI game, no top researcher wants to work for IBM. Spent years developing Watson, it is dead. I believe this is a company that should not be existed.
Maybe it's the opposite. IBM spent years on the technology. Watson used neural networks, just not nearly as large. Perhaps they foresaw that it wouldn't scale or that it would plateau.
IBM CEO is steering a broken ship and it's not improved course, not someone who's words you should take seriously.
1. The missed the AI wave (hired me to teach watson law only to lay me off 5 wks later, one cause of the serious talent issues over there)
2. They bought most of their data center (companies), they have no idea about building and operating one, not at the scale the "competitors" are operating at
His argument follows almost directly, and trivially, from his central premise: a 0% or 1% chance of reaching AGI.
Yeah, if you assume technology will stagnate over the next decade and AGI is essentially impossible, these investments will not be profitable. Sam Altman himself wouldn't dispute that. But it's a controversial premise, and one that there's no particular reason to think that the... CEO of IBM would have any insight into.
then it seems like neither Sam Altman (pro) or IBM (proxy con) have credible or even really interesting or insightful evidence, theories ... even suggestions for what's likely to happen? i.e. We should stop listening to all of them?
It's a very reasonable claim to make, but yes, average denizen of peanut gallery can spot this is a bubble from a mile way, doensn't need "insight" of napkin math done by some CEO that's not even in the industry.
Tho he's probably not too happy that they sold the server business to Lenovo, could at least earn something on selling shovels
we need businesses who are willing to pay for ai / compute at prices where both sides are making money
I for one could 10x my AI usage if the results on my side pan out. Spending $100 on ai today has ROI, will 10x that still have ROI for me in a couple years? probably, I expect agentic teams to increase in capability and more of my work. Then the question is can I turn that increase productivity into more revenues (>$1000 / month, one more client would cover this and then some)
Are the numbers he's claiming accurate? They seem like big numbers pulled out of the air, certainly much large than the numbers we've actually seen committed to (not even deployed yet).
IBM's HPC products were enterprise oriented slop products banked on their reputation, and the ROI torched their credibility when compute costs started getting taken seriously. Watson and other products got smeared into kafkaesque arbitrary branding for other product suites, and they were nearly all painful garbage - mobile device management standing out as a particularly grotesque system to use.
Now, IBM lacks any legitimate competitive edge in any of the bajillion markets they tried to target, no credibility in any of their former flagship domains, and nearly every one of their products is hot garbage that costs too much, often by orders of magnitude, compared to similar functionality you can get from things like open source or even free software offered and serviced by other companies.
They blew a ton of money on HPC before there was any legitimate reason to do so. Watson on Jeopardy was probably the last legitimately impressive thing they did, and all of their tech and expertise has been outclassed since.
Despite the flashy title that's the first "sober" analysis from a CEO I read about the technology. While not even really news, it's also worth mentioning that the energy requirements are impossible to fulfill
Also now using ChatGPT intensely since months for all kinds of tasks and having tried Claude etc. None of this is on par with a human. The code snippets are straight out of Stackoverflow...
Take this "sober" analysis with a big pinch of salt.
IBM have totally missed the AI boat, and a large chunk of their revenue comes from selling expensive consultants to clients who do not have the expertise to do IT work themselves - this business model is at a high risk of being disrupted by those clients just using AI agents instead of paying $2-5000/day for a team of 20 barely-qualified new-grads in some far-off country.
IBM have an incentive to try and pour water on the AI fire to try and sustain their business.
Is this true in 2025?
Asking because the biggest IT consulting branch of IBM, Global Technology Services (GTS), was spun off into Kyndryl back in 2021[0]. Same goes for some premier software products (including one I consulted for) back in 2019[1]. Anecdotal evidence suggests the consulting part of IBM was already significantly smaller than in the past.
It's worth noting that IBM may view these AI companies as competitors to it's Watson AI tech[2]. It already existed before the GPU crunch and hyperscaler boom - runs on proprietary IBM hardware.
[0] https://en.wikipedia.org/wiki/Kyndryl
[1] https://www.prnewswire.com/news-releases/hcl-technologies-to...
[2] https://en.wikipedia.org/wiki/IBM_Watson
I know people who still work there and are doing consultancy work for clients.
I am a former IBMer myself but my memory is hazy. IIRC there was 2 arms of the consultants - one was the boring day to day stuff, and the other was "innovation services" or something. Maybe the spun out the drudgery GTS and kept the "innovation" service? No idea.
The part that was spun off was "Infrastructure Services" (from the Wiki article.) Outsourcing and operations, not the business consulting organization that provides high level strategy to coding services.
https://www.ibm.com/consulting
Yes. GTS was infrastructure services and was spun off. What's left is the old GBS - business services and systems implementation services.
Missed the boat? Have you been living under a rock? Watson AI advertising has been everywhere for years.
It’s not that they aren't in the AI space, it’s that the CEO has a shockingly sober take on it. Probably because they’ve been doing AI for 30+ years combined with the fact they don’t have endless money with nowhere to invest it like Google.
> Missed the boat? […] Watson AI advertising has been everywhere for years.
They were ahead of the game with their original Watson tech, but pretty slow to join and try get up to speed with the current GenAI families of tech.
The meaning of “AI” has shifted to mean “generative AI like what ChatGPT does” in the eyes of most so you need to account for this. When people talk about AI, even though it is a fairly wide field, they are generally referring to a limited subset of it.
I would any day take chatGPT/Claude over an IBM consultant. I worked at IBM.
I'd rather be slapped in the face than kicked by a horse, but that doesn't mean either is a good thing
> a high risk of being disrupted by those clients just using AI agents instead of paying $2-5000/day for a team of 20 barely-qualified new-grads in some far-off country
Is there any concrete evidence of that risk being high? That doesn't come from people whose job is to sell AI?
How do you see the math working out?
The numbers are staggering.
The fair answer is that nobody knows. Even Ilya answered he does not know on his latest podcast with Dwarkesh.
Both top line and bottom line numbers are staggering. Nobody knows. Let's not try to convince people otherwise.
> IBM have an incentive to try and pour water on the AI fire to try and sustain their business.
IBM has faced multiple lawsuits over the years. From age discrimination cases to various tactics allegedly used to push employees out, such as requiring them to relocate to states with more employer friendly laws only to terminate them afterward.
IBM is one of the clearest examples of a company that, if given the opportunity to replace human workers with AI, would not hesitate to do so. Assume therefore, the AI does not work for such a purpose...
Right. Just like Intel completely shitting the bed on GPUs then their CEO tweeted a prayer. Old tech companies are going to be left behind.
IBM was ahead of the boat! They had Watson on Jeopardy years ago! /s
I think you make a fair point about the potential disruption for their consulting business but didn't they try to de-risk a bit with the Kyndryl spinout?
I am a senior engineer, I use cursor a lot in my day to day. I find I can code longer and typically faster than without. Is it on par with human? It’s getting pretty darn close to be honest, I am sure the “10x” engineers of the world would disagree but it definitely has surpassed a junior engineer. We all have our anecdotes but I am inclined to believe on average there is net value.
I think surpassed is not the right word because it doesn't create/ideate. However it is incredibly resourceful. Maybe like having a jr engineer to do your bidding without thinking or growing.
Surpassed is probably the wrong word but the intent is more that it can comprehend quite complicated algorithms and patterns and apply them to your problem space. So yea it’s not a human but I don’t think saying subpar to a human is the right comparison either. In many ways it’s much better, I can run N parallel revisions and have the best implementation picked for review. This all happens in seconds.
Yes, this. Creating multiple iterations in parallel allows much more meaningful exploration of the solution space. Create a branch for each framework and try them all, compare them directly in praxis not just in theory. My brother is doing this to great effect as a solopreneur, and having the time of his life.
Largely agree. Anything that is just a multi-file edit, like an interface change, it can do. Maybe not immediately, but you can have it iterate, and it doesn't eat up your attention.
It is without a doubt worth more than the 200 bucks a month I spend on it.
I will go as far as to say it has decent ideas. Vanilla ideas, but it has them. I've actually gotten it to come up with algorithms that I thought were industry secrets. Minor secrets, sure. But things that you don't just come across. I'm in the trading business, so you don't really expect a lot of public information to be in the dataset.
A lot of time vanilla ideas and established, well proven patterns are just what the customer ordered. And AI code tools are great at this now.
i think less. not sure if that's a good thing. but small little bugs and improvements get cleared so quickly now.
Senior engineer here as well. I would say Opus 4.5 is easily a mid-level engineer. It's a substantial improvement over Sonnet 4.5, which required a lot more hand-holding and interventions.
I'm not sure what you mean with the "code snippets are straight out of Stackoverflow". That is factually incorrect just by how LLM works. By now there has been so much code ingested from all kinds of sources, including Stackoverflow LLM is able to help generate quite good code in many occasions. My point being it is extremly useful for super popular languages and many languages where resources are more scarce for developer but because they got the code from who knows where, it can definitely give you many useful ideas.
It's not human, which I'm not sure what is supposed to actually mean. Humans make mistakes, humans make good code. AI does also both. What it definitely needs is a good programmer still on top to know what he is getting and how to improve it.
I find AI (LLM) very useful as a very good code completion and light coder where you know exactly what to do because you did it a thousand times but it's wasteful to be typing it again. Especially a lot of boilerplate code or tests.
It's also useful for agentic use cases because some things you just couldn't do before because there was nothing to understand a human voice/text input and translate that to an actual command.
But that is all far from some AGI and it all costs a lot today an average company to say that this actually provided return on the money but it definitely speeds things up.
An issue with the doom forecasts is most of the hypothetical $8tn hasn't happened yet. Current big tech capex is about $315bn this year, $250bn last against a pre AI level ~$100bn so ~$400bn has been spent so far on AI boom data centers. https://sherwood.news/business/amazon-plans-100-billion-spen...
The future spend is optional - AGI takeoff, you spend loads, not happening not so much.
Say it levels of at $800bn. The world's population is ~8bn so $100 a head so you'd need to be making $10 or $20 per head per year. Quite possibly doable.
65% of people in the world earn less than 3000 euros/year.
Getting 65% of the population to spend 1% of their income on some new digital toy forever does not seem so far fetched.
Lol. If you ballpark numbers like that probably anything is doable!
Yesterday I was talking to coworkers about AI I mentioned that a friend of mine used ChatGPT to help him move. So a coworker said I have to test this and asked ChatGPT if he could fit a set of the largest Magnepan speakers (the wide folding older room divider style) in his Infinity QX80. The results were hilarious. It had some of the dimensions right but it then decided the QX80 is as wide as a box truck (~8-8.5 feet/2.5 m) and to align the nearly 7 foot long speakers sideways between the wheel wells. It also posted hilariously incomprehensible ASCII diagrams.
Here's what Gemini 3 says (don't know if correct but seems reasonable):
"Can my friend fit a set of the largest Magnepan speakers (the wide folding older room divider style) in his Infinity QX80?"
Yes, your friend can very likely fit them, but they will need to prepare the vehicle specifically for a tight fit.
Based on the "folding room divider" description, you are looking at the vintage Magnepan Tympani series (likely the Tympani I, III, or IV). These are famous for being massive, hinged panels that measure 72 inches (6 feet) tall.
Here is the breakdown of the fit and the strategy for moving them safely:
1. The Math: It fits, but use the "Captain's" space The Speaker Size: When folded, a Tympani speaker stack is roughly 72" Long x 18" Wide x 4" Thick. You have two of these stacks (one for left, one for right).
The QX80 Cargo Space: With the 2nd and 3rd rows folded flat, the Infinity QX80 offers about 86–87 inches of length from the liftgate to the back of the front seats.
The Verdict: You have roughly 14 inches of clearance lengthwise. They will fit lying flat.
2. Critical "Must-Dos" Before Loading Even though they fit on paper, these speakers are fragile and awkward. Do not skip these steps:
Remove the Feet: The speakers have metal feet (T-shaped or U-shaped) screwed into the bottom to make them stand up. You must bring a screwdriver/Allen wrench to remove these. If you leave them on, the speakers will not stack flat, they will slide around, and the feet could tear the QX80's leather or carpet.
Bring Moving Blankets (Lots of them): Magnepans are essentially mylar stretched over magnets. If something pokes the grille cloth, it can ruin the diaphragm. Lay a thick blanket on the cargo floor, place the first speaker stack, add a layer of blankets, and then the second stack.
Slide the Front Seats Forward: Just to be safe, move the driver and passenger seats forward a couple of inches before loading. This ensures you don't accidentally crush the top of the speaker frame when closing the liftgate.
3. A Warning on "Folding" Check the Hinges: These speakers are 30-40 years old. The fabric hinges that connect the panels often rot. When you go to fold them, they might separate completely. This is actually good for moving (smaller pieces are easier to handle), but be prepared for them to come apart.
Do NOT stack anything on top of them: These are heavy (approx. 90-100 lbs per speaker). Do not put boxes or amps on top of the planar surfaces during the drive.
Summary for your friend: Tell him to clear the car completely, bring tools to take the feet off, and treating them like panes of glass. They will slide right in.
Please don't pollute the comments with noise. This adds nothing to the conversation.
Your assessment of Claude simply isn’t true.
Or Stackoverflow is really good.
I’m producing multiple projects per week that are weeks of work each.
Would you mind sharing some of these projects?
I've found Claude's usefulness is highly variable, though somewhat predictable. It can write `jq` filters flawlessly every time, whereas I would normally spend 30 minutes scanning docs because nobody memorizes `jq` syntax. And it can comb through server logs in every pod of my k8s clusters extremely fast. But it often struggles making quality code changes in a large codebase, or writing good documentation that isn't just an English translation of the code it's documenting.
It is always "I'm producing 300 projects in a nanosecond" but it's almost never about sharing or actually deploying these ;)
The problem I had that the larger your project gets, the more mistakes Claude makes. I (not a parent commenter) started with a basic CRUD web app and was blown away by how detailed it was, new CSS, good error handling, good selection and use of libraries, it could even write the terminal commands for package management and building. As the project grew to something larger Claude started forgetting that some code already existed in the project and started repeating itself, and worse still when I asked for new features it would pick a copy at random leaving them out of sync with eachother. Moving forward I've been alternating between writing stuff with AI, then rewriting it myself.
> The problem I had that the larger your project gets, the more mistakes Claude makes
I think the reason for this is because these systems get all their coding and design expertise from training, and while there is lots of training data available for small scale software (individual functions, small projects), there is much less for large projects (mostly commercial and private, aside from a few large open source projects).
Designing large software systems, both to meet initial requirements, and to be maintainable and extensible over time, is a different skill than writing small software projects, which is why design of these systems is done by senior developers and systems architects. It's perhaps a bit like the difference between designing a city and designing a single building - there are different considerations and decisions being made. A city is not just a big building, or a collection of buildings, and large software system is not just a large function or collection of functions.
Well said, and good analogy.
I’ve got Claude on 10k loc projects, which is probably mid range.
But I also have it document and summarise its own work.
So, just like a human on a growing codebase?
Here's mine fully deployed, https://hackernewsanalyzer.com/. I use it daily and have some users. ~99.7% LLM code. About 1 hour to first working prototype then another 40 hours to get it polished and complete to current state.
To be clear, roughly 39.8 hours of just prompting and output to make this website?
41 hours total of prompting, looking at code diffs, reverting, reprompting, and occasional direct code commits. I do review the full code changes nearly every step of the way and often iterate numerous times until I'm satisfied with the resulting code approach.
If you haunt the pull requests of projects you use I bet you'll find there's a new species of PR:
> I'm not an expert in this language or this project but I used AI to add a feature and I think its pretty good. Do you want to use it?
I find myself writing these and bumping into others doing the same thing. It's exciting, projects that were stagnant are getting new attention.
I understand that a maintainer may not want to take responsibility for new features of this sort, but its easier than ever to fork the project and merge them yourself.
I noticed this most recently in https://github.com/andyk/ht/pulls which has two open (one draft) PRs of that sort, plus several closed ones.
Issues that have been stale for years are getting traction, and if you look at the commit messages, it's AI tooling doing the work.
People feel more capable to attempt contributions which they'd otherwise have to wait for a specialist for. We do need to be careful not to overwhelm the specialists with such things, as some of them are of low quality, but on the whole it's a really good thing.
If you're not noticing it, I suggests hanging out in places where people actually share code, rather than here where we often instead brag about unshared code.
> People feel more capable to attempt contributions
That does not mean that they are more capable, and that's the problem.
> We do need to be careful not to overwhelm the specialists with such things, as some of them are of low quality, but on the whole it's a really good thing.
That's not what the specialists who have to deal with this slop say. There have been articles about this discussed here already.
At this point my prior is that all these 300/ns projects are some kind of internal tools, with very narrow scope and many just for a one-off use.
Which is also fine and great and very useful and I am also making those, but it probably does not generalize to projects that require higher quality standards and actual maintenance.
Sure, but 80% of software is probably internal and short lived like that, if not more.
Solving the problems of the business that isn’t a software business.
I also see a lot of this so I can't blame you for thinking it! See my other post about some projects build _only_ using LLMs.
https://news.ycombinator.com/item?id=46133458
There's a massive incentive not to share them. If I wrote a project using AI I'd be reluctant to publish it at all because of the backlash I've seen people get for it.
People are and always were reluctant to share their own code just the same. There is nothing to be gained, the chances of getting positive reviews from fellow engineers are slim to none. We are a critical and somewhat hypocritical bunch on average.
This is a "scratch an itch" project I initially started to write manually in the past, but never finishing. I then used claude to do it basically on the side while watching the world series http://nixpkgs-pr-explorer.s3-website-us-west-2.amazonaws.co...
Not the OP you're replying to, but I've put together quite a few projects using only LLMs, no hand crafted code anywhere (I couldn't do it!)
https://dnbfamily.com
https://eventme.app
https://blazingbanana.com
https://play.google.com/store/apps/details?id=com.blazingban...
https://play.google.com/store/apps/details?id=com.blazingban...
https://play.google.com/store/apps/details?id=com.blazingban...
Are they perfect? No probably not, but I wouldn't have been able to make any of these without LLMs. The last app was originally built with GPT-3.5.
There is a whole host of other non-public projects I've built with LLMs, these are just a few of the public ones.
Maybe the most depressing part of all this is if people start thinking they would not have been able to do things without the LLM. Of course they would have, it's not like LLMs can do anything that you cannot. Maybe it would have taken more time at least the first time and you would have learned a few things in the process.
I understand the point, and to some degree agree. For myself, I really couldn't (not to say it wouldn't have been possible). I tried many many times over so many years and just didn't have the mental stamina for it, it would never "click" like infra/networking/hardware does etc and I would always end up frustrated.
I have learnt so much in this process, nowhere near as much as someone that wrote every line (which is why I think being a good developer will be a hot commodity) but I have had so much fun and enjoyment, alongside actually seeing tangible stuff get created, at the end of the day, that's what it's all about.
I have a finite amount of time to do things, I already want to do more than I can fit into that time, LLMs help me achieve some of them.
Sure, I can write all of it. But I simply won’t. I have Claude generated Avalonia C# applications and there is no way I would have written the thousands of lines of xaml they needed for the layouts. I would just have done it as a console app with flags.
But reducing friction, eliminating the barrier to entry, is of fundamental importance. It's human psychology; putting running socks next to your bed at night makes it like 95% more likely you'll actually go for a run in the morning.
Yes, "I couldn't have bothered..." is different from " I wouldn't have been able to make...".
You might not go for a run when the socks are not there, but I don't think you would start questioning your ability to run.
It would be more depressing if our imagination didn't exceed the finite time we have to learn and master new skills.
Or if we stopped imagining.
https://github.com/lawless-m
You can see by Contributors which ones Claude has done.
I have no idea if the code is any good, I’ve never looked at it and I have no idea how to code in Rust or Racket or Erlang anyway.
Claude has taught me so much about how to use jq better. And really, way more efficient ways of using the command line in general. It's great. Ironically, the more I learn the less I want to ask it to do things.
In an ideal world we function in exactly this way - using LLMs to bootstrap our skill/knowledge improvement journeys.
The former tasks are directly from the training material, directly embedded into the model. For the latter task, it needs a context window and intelligence.
At the end of this week we are releasing https://github.com/eqtylab/cupcake
You can see all of Claude’s commits.
I’ve shipped so much with ai.
Favorite has been metrics dashboards of various kinds - across life and business.
Something about restricting an AI agent by using another AI to write code to restrict it is quite funny
It'll be a common paradigm. Some agents support the coding agent discover relevant context for a plan, others will help the agent stay on track and ensure no rules break.
They really should have been supplying at least a week worth of readymade "projects" to every freelance AI promoter out there to demonstrate x9000 AI productivity gains for the skeptics.
Because vibing the air about those gains without any evidence looks too shilly.
As opposed to ad hominems and handwave dismissals, which are highly credible?
I handwave dismiss bigfoot, too
Pointing out the where the burden of proof lies is not an ad hominem. Calling it such is in fact a good example of poisoning the well. all the fan girls have to do is post links to code they have vibe coded. some people have even done that in this thread. it's not an unreasonable standard.
It’s not just good for small code bases. In the last six months I’ve built a collaborative word processor with its own editor engine and canvas renderer using Claude, mostly Opus. It’s practically a mini Google Docs, but with better document history and an AI agent built in. I could never have built this in 6 months by myself without Claude Code.
https://revise.io
I think if you stick with a project for a while, keep code organized well, and most importantly prioritize having an excellent test suite, you can go very far with these tools. I am still developing this at a high pace every single day using these tools. It’s night and day to me, and I say that as someone who solo founded and was acquired once before, 10 years ago.
I'm just as much of an avid llm code generator fan as you may be but I do wonder about the practicality of spending time making projects anymore.
Why build them if other can just generate them too, where is the value of making so many projects?
If the value is in who can sell it the best to people who can't generate it, isn't it just a matter of time before someone else will generate one and they may become better than you at selling it?
> Why build them if other can just generate them too, where is the value of making so many projects?
No offence to anyone but these generated projects are nothing ground-breaking. As soon as you venture outside the usual CRUD apps where novelty and serious engineering is necessary, the value proposition of LLMs drops considerably.
For example, I'm exploring a novel design for a microkernel, and I have no need for machine generated boilerplate, as most of the hard work is not implementing yet another JSON API boilerplate, but it's thinking very hard with pen and paper about something few have thought before, and even fewer LLMs have been trained on, and have no intelligence to ponder upon the material.
To be fair, even for the most dumb side-projects, like the notes app I wrote for myself, there is still a joy in doing things by hand, because I do not care about shipping early and getting VC money.
The value is that we need a lot more software and now, because building software has gotten so much less time consuming, you can sell software to people that could/would not have paid for it previously at a different price point.
We don’t need more software, we need the right software implemented better. That’s not something LLMs can possibly give us because they’re fucking pachinko machines.
Here’s a hint: Nobody should ever write a CRUD app, because nobody should ever have to write a CRUD app; that’s something that can be generated fully and deterministically (i.e. by a set of locally-executable heuristics, not a goddamn ocean-boiling LLM) from a sufficiently detailed model of the data involved.
In the 1970s you could wire up an OS-level forms library to your database schema and then serve literally thousands of users from a system less powerful than the CPU in modern peripheral or storage controller. And in less RAM too.
People need to take a look at what was done before in order to truly have a proper degree of shame about how things are being done now.
Most CRUD software development is not really about the CRUD part. And for most framework, you can find packages that generate the UI and the glue code that ties it to the database.
When you're doing CRUD, you're spending most of the time with the extra constraints designed by product. It's dealing with the CRUD events, the IAM system, the Notification system,...
> That’s not something LLMs can possibly give us because they’re fucking pachinko machines.
I mostly agree, but I do find them useful for fuzzing out tests and finding issues with implementations. I have moved away from larger architectural sketches using LLMs because over larger time scales I no longer find they actually save time, but I do think they're useful for finding ways to improve correctness and safety in code.
It isn't the exciting and magical thing AI platforms want people to think it is, and it isn't indispensable, but I like having it handy sometimes.
The key is that it still requires an operator who knows something is missing, or that there are still improvements to be made, and how to suss them out. This is far less likely to occur in the hands of people who don't know, in which case I agree that it's essentially a pachinko machine.
I’m with you. Anyone writing in anything higher level than assembly, with anything less than the optimization work done by the demo scene, should feel great same.
Down with force-multiplying abstractions! Down with intermediate languages and CPU agnostic binaries! Down with libraries!
You have clearly entirely understood exactly what I was saying and don’t look like a fool at all with this reply.
But what we're getting is a flood of buggy, unoriginal crap.
Remind me when exactly it was that most software was bug-free and original? 1990? 1975?
It global warming exists, why is there snow?
Same thing here. You're dismissing a flood because it rains.
Sure but these are likely just variations of existing things. And yet the quality is still behind the original
I produce a lot of shit every week too, but I don’t brag about my digestive system on “Hacker” “News.”
I agree. re: energy and other resource use: the analogy I like is with driving cars: we use cars for transportation knowing the environmental costs so we don’t usually just go on two hour drives for the fun of it, rather we drive to get to work, go shopping. I use Gemini 3 but only in specific high value use cases. When I use commercial models I think a little about the societal costs.
In the USA we have lost the thread here: we don’t maximize the use of small tuned models throughout society and industry, instead we use the pursuit of advanced AI as a distraction to the reality that our economy and competitiveness are failing.
Most of the energy for AI does not go into chatbots. Using Gemini is not remotely close to driving a car for 2 hours. If a prompt is 0.3 Wh (https://cloud.google.com/blog/products/infrastructure/measur..., https://andymasley.substack.com/p/a-cheat-sheet-for-conversa...), it’s closer to using an e-bike for 50 metres.
You could have your morning shower 1°C less hot and save enough energy for about 200 prompts (assuming 50 litres per shower). (Or skip the shower altogether and save thousands of prompts.)
I'd rather phrase it as "code is straight out of GitHub, but tailored to match your data structures"
That's at least how I use it. If I know there's a library that can solve the issue, I know an LLM can implement the same thing for me. Often much faster than integrating the library. And hey, now it's my code. Ethical? Probably not. Useful? Sometimes.
If I know there isn't a library available, and I'm not doing the most trivial UI or data processing, well, then it can be very tough to get anything usable out of an LLM.
I've had decent results hackin', wackin' and smashin'.
> Also now using ChatGPT intensely since months for all kinds of tasks and having tried Claude etc.
the facts though, read like an endorsement not a criticism
> While not even really news, it's also worth mentioning that the energy requirements are impossible to fulfill
If you believe this, you must also believe that global warming is unstoppable. OpenAI's energy costs are large compared to the current electricity market, but not so large compared to the current energy market. Environmentalists usually suggest that electrification - converting non-electrical energy to electrical energy - and then making that electrical energy clean - is the solution to global warming. OpenAI's energy needs are something like 10% of the current worldwide electricity market but less than 1% of the current worldwide energy market.
Google recently announced to double AI data center capacity every 6 month. While both unfortunately deal with exponential growth, we are talking about 1% growth CO2 which is bad enough vs 300% effectively per year according to Google
Constraints breed innovation. Humans will continue to innovate and demand for resources will grow. it is fairly well baked into most of civilization. Will that change in the future? Perhaps but it’s not changing now.
Imagine how big pile of trash as the current generation of graphics cards used for LLM training will get outdated. It will crash the hardware market (which is a good news for gamers)
A100’s are not suitable for gaming.
https://www.youtube.com/watch?v=Vw699ZbUKqg
Looks very playable to me.
It's just an expensive card, but if the market is flooded with them, they can be used in gaming AND in local LLMs.
So it can push the fall of server-side AI even further.
These cards are 400 USD for reference, so if more and more are sold, we can imagine them getting down to 100 USD or so.
(and then similar for A100, H100, etc)
My main concern is the noise because I have seen datacenter hardware and it is crazy. Of course it's not ideal but there is something to do with it.
[dead]
> Krishna also referenced the depreciation of the AI chips inside data centers as another factor: "You've got to use it all in five years because at that point, you've got to throw it away and refill it," he said
This doesn't seem correct to me, or at least is built on several shaky assumptions. One would have to 'refill' your hardware if:
- AI accelerator cards all start dying around the 5 year mark, which is possible given the heat density/cooling needs, but doesn't seem all that likely.
- Technology advances such that only the absolute newest cards can be used to run _any_ model profitably, which only seems likely if we see some pretty radical advances in efficiency. Otherwise, it seems like assuming your hardware is stable after 5 years of burn in, you could continue to run older models on that hardware at only the cost of the floorspace/power. Maybe you need new cards for new models for some reason (maybe a new fp format that only new cards support? some magic amount of ram? etc), but it seems like there may be room for revenue via older/less capable models at a discounted rate.
Isn’t that what Michael Burry is complaining about? That five years is actually too generous when it comes to depreciation of these assets and that companies are being too relaxed with that estimate. The real depreciation is more like 2-3 years for these GPUs that cost tens of thousands of dollars a piece.
https://x.com/michaeljburry/status/1987918650104283372
That's exactly the thing. It's only about bookkeeping.
The big AI corps keep pushing depreciation for GPUs into the future, no matter how long the hardware is actually useful. Some of them are now at 6 years. But GPUs are advancing fast, and new hardware brings more flops per watt, so there's a strong incentive to switch to the latest chips. Also, they run 24/7 at 100% capacity, so after only 1.5 years, a fair share of the chips is already toast. How much hardware do they have in their books that's actually not useful anymore? Noone knows! Slower depreciation means more profit right now (for those companies that actually make profit, like MS or Meta), but it's just kicking the can down the road. Eventually, all these investments have to get out of the books, and that's where it will eat their profits. In 2024, the big AI corps invested about $1 trillion in AI hardware, next year is expected to be $2 trillion. Only the interest payments for that are crazy. And all of this comes on top of the fact that none of the these companies actually make any profit at all with AI. (Except Nvidia of course) There's just no way this will pan out.
> It's only about bookkeeping.
> Some of them are now at 6 years.
There are three distinct but related topics here, it's not "just about bookkeeping" (though Michael Burry may be specifically pointing to the bookkeeping being misquoted):
1. Financial depreciation - accounting principals typically follow the useful life of the capital asset (simply put, if an airplane typically gets used for 30 years, they'll split the cost of purchasing an airplane across 30 years equally on their books). Getting this right has more to do with how future purchases get financed due to how the bookkeepers show profitability, balance sheets, etc.. Cashflow is ultimately what might create an insolvent company.
2. Useful life - per number 1 above - this is the estimated and actual life of the asset. So if the airplane actually is used over 35 years, not 30, it's actual useful life is 35 years. This is to your point of "some of them are 6 years old". Here is where this is going to get super tricky with GPUs. We (a) don't actually know what the useful life is or is going to be (hence Michael Burry's question) for these GPUs (b) the cost of this is going to get complicated fast. Let's say (I'm making these up) GPU X2000 is 2x the performance of GPU X1000 and your whole data center is full of GPU X1000. Do you replace all of those GPUs to increase throughput?
3. Support & maintenance - this is what actually gets supported by the vendor. There doesn't seem to be any public info about the Nvidia GPUs but typically these are 3-5 years (usually tied to the useful life) and often can be extended. Again, this is going to get super complicated to financially because we don't know what future advancements might happen to performance improvements to GPUs (and therefore would necessitate replacing old ones and therefore creating renewed maintenance contracts).
> Also, they run 24/7 at 100% capacity, so after only 1.5 years
How does OpenAI keep this load? I would expect the load at 2pm Eastern to be WAY bigger than the load after California goes to bed.
Typical load management that’s existed for 70 years: when interactive workloads are off-peak, you do batch processing. For OpenAI that’s anything from LLM evaluation of the days’ conversations to user profile updates.
People outside the 4 U.S. Timezones exist?
The Pacific ocean is big.
Flops per watt is relevant for a new data center build-out where you're bottlenecked on electricity, but I'm not sure it matters so much for existing data centers. Electricity is such as small percentage of total cost of ownership. The marginal cost of running a 5 year old GPU for 2 more years is small. The husk of a data center is cheap. It's the cooling, power delivery equipment, networking, GPUs etc that costs money, and when you retrofit data centers for the latest and greatest GPUs you have to throw away a lot of good equipment. Makes more sense to build new data centers as long as inference demand doesn't level off.
How different is this from rental car companies changing over their fleets? I don't know, this is a genuine question. The cars cost 3-4x as much and last about 2x as far as I know, and the secondary market is still alive.
> How different is this from rental car companies changing over their fleets?
New generations of GPUs leapfrog in efficiency (performance per watt) and vehicles don't? Cars don't get exponentially better every 2–3 years, meaning the second-hand market is alive and well. Some of us are quite happy driving older cars (two parked outside our home right now, both well over 100,000km driven).
If you have a datacentre with older hardware, and your competitor has the latest hardware, you face the same physical space constraints, same cooling and power bills as they do? Except they are "doing more" than you are...
Would we could call it "revenue per watt"?
The traditional framing would be cost per flop. At some point your total costs per flop over the next 5 years will be lower if you throw out the old hardware and replace it with newer more efficient models. With traditional servers that's typically after 3-5 years, with GPUs 2-3 years sounds about right
The major reason companies keep their old GPUs around much longer with now are the supply constraints
The used market is going to be absolutely flooded with millions of old cards. I imagine shipping being the most expensive cost for them. The supply side will be insane.
Think 100 cards but only 1 buyer as a ratio. Profit for ebay sellers will be on "handling", or inflated shipping costs.
eg shipping and handling.
I assume NVIDIA and co. already protects themselves in some way, either by the fact of these cards not being very useful after resale, or requiring them to go to the grinder after they expire.
Cards don't "expire". There are alternate strategies to selling cards, but if they don't sell the cards, then there is no transfer of ownership, and therefore NVIDIA is entering some form of leasing model.
If NVIDIA is leasing, then you can't get use those cards as collateral. You can't also write off depreciation. Part of what we're discussing is that terms of credit are being extended too generously, with depreciation in the mix.
The could require some form of contractual arrangement, perhaps volume discounts for cards, if they agree to destroy them at a fixed time. That's very weird though, and I've never heard of such a thing for datacenter gear.
They may protect themselves on the driver side, but someone could still write OSS.
Dont they own socket for enterprise cards? I can't see consumers buying these card unless they are PCIE at the very least.
Rental car companies aren’t offering rentals at deep discount to try to kickstart a market.
It would be much less of a deal if these companies were profitable and could cover the costs of renewing hardware, like car rental companies can.
I think it's a bit different because a rental car generates direct revenue that covers its cost. These GPU data centers are being used to train models (which themselves quickly become obsolete) and provide inference at a loss. Nothing in the current chain is profitable except selling the GPUs.
> and provide inference at a loss
You say this like it's some sort of established fact. My understanding is the exact opposite and that inference is plenty profitable - the reason the companies are perpetually in the red is that they're always heavily investing in the next, larger generation.
I'm not Anthropic's CFO so i can't really prove who's right one way or the other, but I will note that your version relies on everyone involved being really, really stupid.
The current generation of today was the next generation of yesterday. So, unless the services sold on inference can cover the cost of inference + training AND gain money, they are still operating at loss.
“like it's some sort of established fact” -> “My understanding”?! a.k.a pure speculation. Some of you AI fans really need to read your posts out loud before posting them.
You misread the literal first snippet you quoted. There's no contradiction in what you replied to.
No?
Or just "everyone" being greedy
> the secondary market is still alive.
this is the crux. Will these data center cards, if a newer model came out with better efficiency, have a secondary market to sell to?
It could be that second hand ai hardware going into consumers' hands is how they offload it without huge losses.
The GPUs going into data centers aren't the kind that can just be reused by putting them into a consumer PC and playing some video games, most don't even have video output ports and put out FPS similar to cheap integrated GPUs.
And the big ones don't even have typical PCIe sockets, they are useless outside of behemoth rackmount servers requiring massive power and cooling capacity that even well-equipped homelabs would have trouble providing!
Don’t underestimate a homelaber’s intention to cosplay as a sysadmin or ability to set their house on fire ;)
I wonder if people will come up with ways to repurpose those data center cards.
Data centre cards a don’t have fans and don’t have video out these days.
i dont mean consumer market for video cards - i mean a consumer buying ai chips to run themselves so they can have it locally.
If i can buy a $10k ai card for less than $5000 dollars, i probably would, if i can use it to run an open model myself.
At that point it isn't a $10k card anymore, it's a $5k card. And possibly not a $5k card for very long in the scenario that the market has been flooded with them.
Ah well yes to a degree that’s possible but at least at the moment you’d still be better off buying a $5k Mac Studio if it’s just inference you’re doing
How many "yous" are there in the world? Probably a number that can buy what's inside one Azure DC?
Why would you do that when you can pay someone else to run the model for you on newer more efficient and more profitable hardware? What makes it profitable for you and not for them?
Control and privacy?
You need the hardware to wrap that in, and the power draw is going to be... significant.
5 years is long, actually. This is not a GPU thing. It's standard for server hardware.
Because usually it's more efficient for companies to retire the hardware and put in new stuff.
Meanwhile, my 10-15 year old server hardware keeps chugging along just fine in the rack in my garage.
There are several significant differences between your garage server and a data center.
"Just fine". Presumably you're not super-concerned with the energy costs? People who run data centres pretty much _have_ to be.
More than that. The equipment is depreciated on a 5 year schedule on the company balance sheet. It actually costs nothing to discard it.
There’s no marginal tax impact of discarding it or not after 5 years - if it was still net useful to keep it powered, they would keep it. Depreciation doesn’t demand you dispose of or sell the item to see the tax benefit.
No but it tips the scales. If the new hardware is a little more efficient, but perhaps not so much so that you would necessarily replace it, the ability to appreciate the new stuff, but not the old stuff might tip your decision
I thought the same until I calculated that newer hardware consumes a few times less energy and for something running 24x7 that adds up quite a bit (I live in Europe, energy is quite expensive).
So my homelab equipment is just 5 years old and it will get replaced in 2-3 years with something even more power efficient.
Where in Europe?
Asking coz I just did a quick comparison and it seems to depend but for comparison I have a really old AMD Athlon "e" processor (like literally September 2009 is when it came out according to some quick Google search, tho I probably bought it a few months later than that but still ...) that runs at ~45W TDP. In idle conditions, it typically consumes around 10 to 15 watts (internet wisdom, not kill-a-watt-wisdom).
Some napkin math says it would cost me about 40 years worth of amortization to replace this at my current power rates for this system. So why would I replace it? And even with some EU countries' power rates we seem to be at 5-10 years amortization upon replacement. I've been running this motherboard, CPU + RAM combo for ~15 years now it seems, replacing only the hard drives every ~3 years. And the tower it's in is about 25 years old.
Oh I forgot, I think I had to buy two new CR2032 batteries during those years (CMOS battery).
Now granted, this processor can basically do "nothing" in comparison to a current system I might buy. But I also don't need more for what it does.
Well if you have a system that does "nothing" it's hard to argue to replace it
"Nothing" from parent was a comparison. Doesn't mean their system is idle.
However many systems are mostly idle. A file server often doesn't use much cpu. It often isn't even serving anything.
I guess you did the math but wouldn't it be more effective to spend the money on solar panels instead of replacing the computer hardware?
Energy is very cheap for data centers. have you ever looked up wholesale energy rates? It’s like a cent per kilowatt hour.
Sample size of 1 though. It's like how I've had hard disks last a decade, but a 100 node Hadoop cluster had 3 die per week after a few years.
Spinning rust and fans are the outliers when it comes to longevity in compute hardware. I’ve had to replace a disk or two in my rack at home, but at the end of the day the CPUs, RAM, NICs, etc. all continue to tick along just fine.
When it comes to enterprise deployments, the lifecycle always revolves around price/performance. Why pay for old gear that sucks up power and runs 30% slower than the new hotness, after all!
But, here we are, hitting limits of transistor density. There’s a reason I still can’t get 13th or 14th gen poweredge boxes for the price I paid for my 12th gen ones years ago.
5 years is a long time for GPUs maybe but normal servers have 7 year lifespans in many cases fwiw.
These GPUs I assume basically have potential longevity issues due to the density, if you could cool it really really well I imagine no problem.
> normal servers have 7 year lifespans in many cases fwiw
Eight years if you use Hetzner servers!
Normal servers are rarely run flat-out. These GPUs are supposed to be run that way. So, yeah, age is going to be a problem, as will cooling.
I think its illustrative to consider the previous computation cycle ala Cryptomining. Which passed through a similar lifecycle with energy and GPU accelerators.
The need for cheap wattage forced the operations to arbitrage the where location for the cheapest/reliable existing supply - there rarely was new buildout as the cost was to be reimbursed by the coins the miningpool recovered.
For the chip situation caused the same apprecaition in GPU cards with periodic offloading of cards to the secondary market (after wear and tear) as newer/faster/more efficient cards came out until custom ASICs took over the heavy lifting, causing the GPU card market to pivot.
Similarly in the short to moedium term the uptick of custo ASICs like with Google TPU will definately make a dent in bot cpex/opex and potentially also lead to a market with used GPUs as ASICs dominate.
So for GPUs i can certainly see the 5 year horizon making a impact in investment decisions as ASICs proliferate.
It's just the same dynamic as old servers. They still work fine but power costs make them uneconomical compared to latest tech.
> They still work fine but power costs make them uneconomical compared to latest tech.
That's not necessarily the driving financial decision, in fact I'd argue company's with data center hardware purchases barely look at this number. It's more simple than that - their support runs out and its cheaper to buy a new piece of hardware (that IS more efficient) because the hardware vendors make extended support inordinately expensive.
Put yourselves in the shoes of a sales person at Dell selling enterprise server hardware and you'll see why this model makes sense.
It’s far more extreme: old servers are still okay on I/O, and memory latency, etc. won’t change that dramatically so you can still find productive uses for them. AI workloads are hyper-focused on a single type of work and, unlike most regular servers, a limiting factor in direct competition with other companies.
I mean you could use training GPUs for inference right? That would be use case number 1 for a 8 * a100 box in a couple of years. It can also be used for non IO limited things like folding proteins or other 'scientific' use cases. Push comes to shove im sure an old A100 will run crysis.
> Push comes to shove im sure an old A100 will run crysis.
They don’t have video out ports!
Just like laptop dGPUs.
All those use cases would probably use up 1% of the current AI infrastructure, let alone ahat they're planning to build.
Yeah, just like gas, possible uses will expand if AI crashes out, but:
* will these uses cover, say, 60% of all this infra?
* will these uses scale up to use that 60% within the next 5-7 years, while that hardware is still relevant and fully functional?
Also, we still have railroad tracks from the 1800s rail mania that were never truly used to capacity and dot com boom dark fiber that's also never been used fully, even with the internet growing 100x since. And tracks and fiber don't degrade as quickly as server hardware and especially GPUs.
LambdaLabs is still making money off their Tesla V100s, A100s, and A6000s. The older ones are cheap enough to run some models and very cheap, so if that's all you need, that's what you'll pick.
The V100 was released in 2017, A6000 in 2020, A100 in 2021.
That could change with a power generation breakthrough. If power is very cheap then running ancient gear till it falls apart starts making more sense
Hugely unlikely.
Even if the power is free you still need a grid connection to move it to where you need it, and, guess what, the US grid is bursting at the seams. This is not just due to data center demand; it was struggling to cope with the transition away from coal well before that point.
You also can’t buy a gas turbine for love nor money at the moment, and they’re not ever going to be free.
If you plonked massive amounts of solar panels and batteries in the Nevada desert, that’s becoming cheap but it ain’t free, particularly as you’ll still need gas backup for a string of cloudy days.
If you think SMRs are going to be cheap I have a bridge to sell you, you’re also not going to build them right next to your data centre because the NRC won’t let you.
So that leaves fusion or geothermal. Geothermal is not presently “very cheap” and fusion power has not been demonstrated to work at any price.
Power consumption is only part of the equation. More efficient chips => less heat => lower cooling costs and/or higher compute density in the same space.
Solution: run them in the north. Put a server in the basement of every home in Edmonton and use the excess heat to warm the house.
I'm a little bit curious about this. Where do all the hardware from the big tech giants usually go once they've upgraded?
In-house hyperscaler stuff gets shredded, after every single piece of flash storage gets first drilled through and every hard drive gets bent by a hydraulic press. Then it goes into the usual e-waste recycling stream (ie. gets sent to poor countries where precious metals get extracted by people with a halved life expectancy).
Off-the-shelf enterprise gear has a chance to get a second life through remarketing channels, but much of it also gets shredded due to dumb corporate policies. There are stories of some companies refusing to offload a massive decom onto the second hand market as it would actually cause a crash. :)
It's a very efficient system, you see.
Similar to corporate laptops where due to stupid policies, for most BigCos you can't really buy or otherwise get a used laptop, even as the former corporate used of said laptop.
Super environmentally friendly.
I used (relatively) ancient servers (5-10 years in age) because their performance is completely adequate; they just use slightly more power. As a plus it's easy to buy spare parts, and they run on DDR3, so I'm not paying the current "RAM tax". I generally get such a server, max out its RAM, max out its CPUs and put it to work.
Same, the bang for buck on a 5yo server is insane. I got an old Dell a year ago (to replace our 15yo one that finally died) and it was $1200 AUD for a maxed out recently-retired server with 72TB of hard drives and something like 292GB of RAM.
Just not too old. Easy to get into "power usage makes it not worth it" for any use case when it runs 24/7
Seriously. 24/7 adds up faster than most realize!
The idle wattage per module has shrunk from 2.5-3W down to 1-1.2 between DDR3 & DDR5. Assuming a 1.3W difference (so 10.4W for 8760 hours), a DDR3 machine with 8 sticks would increase your yearly power consumption by almost 1% (assuming avg 10,500kWh/yr household)
That's only a couple dollars in most cases but the gap is only larger in every other instance. When I upgraded from Zen 2 to Zen 3 it was able to complete the same workload just as fast with half as many cores while pulling over 100W less. Sustained 100% utilization barely even heats a room effectively anymore!
A similar one I just ran into: my Framework Desktop was idling @ 5W more than other reported numbers. Issue turned out to be the 10 year old ATX PSU I was using.
The one thing to be careful with Zen 2 onwards is that if your server is going to be idling most of the time then the majority of your power usage comes from the IO die. Quite a few times you'd be better off with the "less efficient" Intel chips because they save 10-20 Watts when doing nothing.
Wake on LAN?
Then you cannot enjoy some very useful and used home server functions like home automation or NVR.
Maybe? The price difference on newer hardware can buy a lot of electricity, and if you aren't running stuff at 100% all the time the calculation changes again. Idle power draw on a brand new server isn't significantly different from one that's 5 years old.
To be clear, this server is very lightly loaded, it's just running our internal network services (file server, VPN/DNS, various web apps, SVN etc.) so it's not like we're flogging a room full of GeForce 1080Ti cards instead of buying a new 4090Ti or whatever. Also it's at work so it doesn't impact the home power bill. :D
Some is sold on the used market; some is destroyed. There are plenty of used V100 and A100 available now for example.
Manipulating this for creative accounting seems to be the root of Michael Burry’s argument, although I’m not fluent enough in his figures to map here. But, commenting that it interesting to see IBM argue a similar case (somewhat), or comments ITT hitting the same known facts, in light of Nvidia’s counterpoints to him.
Burry just did his first interview for many years https://youtu.be/nsE13fvjz18?t=265
with Michael Lewis, about 30 mins long. Highlights - he thinks we are near the top, his puts are for two years time. If you go long he suggests healthcare stocks. He's been long gold some years, thinks bitcoin is dumb. Thinks this is dotcom bubble #2 except instead of pro investors it's mostly index funds this time. Most recent headlines about him have been bad reporting.
Eh, not exactly. If you don't run CPU at 70%+ the rest of the machine isn't that much more inefficient that model generation or two behind.
It used to be that new server could use half power of the old one at idle but vendors figured out that servers also need proper power management a while ago and it is much better.
Last few gens increase could be summed up to "low % increase in efficiency, with TDP, memory channels and core count increase".
So for loads not CPU bound the savings on newer gen aren't nearly worth it to replace it, and for bulk storage the CPU power usage is even smaller part
Definitely single thread performance and storage are the main reasons not to use an old server. A 6 year old server didn't have nvme drives, so SATA SSD at best. That's a major slow down if disk is important.
Aside from that there's no reason to not use a dual socket server from 5 years ago instead of a single socket server of today. Power and reliability maybe not as good.
NVMe is just a different form factor for what's essentially a PCIe connection, and adapters are widely available to bridge these formats. Surely old servers will still support PCIe?
that was then. now, high-end chips are reaching 4,3,2 nm. power savings aren't that high anymore. what's the power saving going from 4 to 2nm?
+5-20% clockspeed at 5-25% lower voltages (which has been and continues to be the trend) add up quick from gen to gen, nevermind density or ipc gains.
We can’t really go lower on voltage anymore without a very significant change in the materials used. Silicon band gap yadda yadda.
But if your competitor is running newer chips that consume less power per operation, aren't you forced to upgrade as well and dispose of the old hardware?
Sure, assuming the power cost reduction or capability increase justifies the expenditure. It's not clear that that will be the case. That's one of the shaky assumptions I'm referring to. It may be that the 2030 nvidia accelerators will save you $2000 in electricity per month per rack, and you can upgrade the whole rack for the low, low price of $800,000! That may not be worth it at all. If it saves you $200k/per rack or unlocks some additional capability that a 2025 accelerator is incapable of and customers are willing to pay for, then that's a different story. There are a ton of assumptions in these scenarios, and his logic doesn't seem to justify the confidence level.
> Sure, assuming the power cost reduction or capability increase justifies the expenditure. It's not clear that that will be the case.
Share price is a bigger consideration than any +/- differences[1] between expenditure vs productivity delta. GAAP allows some flexibility in how servers are depreciated, so depending on what the company wants to signal to shareholders (investing in infra for futur returns vs curtailing costs), it may make sense to shorten or lengthen depreciation time regardless of the actual TCOO keep/refresh cost comparisons.
1. Hypothetical scenario: a hardware refresh costs $80B, actual performance increase is only worth $8B, but the share price increases the value of org's holding of its own shares by $150B. As a CEO/CFO, which action would you recommend- without even considering your own bonus that's implicitly or explicitly tied to share price performance.
Demand/suppy economics is not so hypothetical.
Illustration numbers: AI demand premium = $150 hardware with $50 electricity. Normal demand = $50 hardware with $50 electricity. This is Nvidia margins @75% instead of 40%. CAPEX/OPEX is 70%/20% hardware/power instead of customary 50%/40%.
If bubble crashes, i.e. AI demand premium evaporates, we're back at $50 hardware and $50 electricity. Likely $50 hardware and $25 electricity if hardware improves. Nvdia back to 30-40% margins, operators on old hardware stuck with stranded assets.
The key thing to understand is current racks are sold at grossly inflated premiums right now, scarcity pricing/tax. If the current AI economic model doesn't work then fundmentally that premium goes away and subsequent build outs are going to be costplus/commodity pricing = capex discounted by non trivial amounts. Any breakthroughs in hardware, i.e. TPU compute efficiency would stack opex (power) savings. Maybe by year 8, first gen of data centers are still depreciated to $80 hardware + $50 power vs new center @ $50 hardware + $25 power. That old data center is a massive write-down because it will generate less revenue than it costs to amoritize.
A typical data centre is $2,500 per year per kW load (including overhead, hvac and so on).
If it costs $800,000 to replace the whole rack, then that would pay off in a year if it reduces 320 kW of consumption. Back when we ran servers, we wouldn't assume 100% utilisation but AI workloads do do that; normal server loads would be 10kW per rack and AI is closer to 100. So yeah, it's not hard to imagine power savings of 3.2 racks being worth it.
Thanks for the numbers! Isn't it more likely that the amount of power/heat generated per rack will stay constant over each upgrade cycle, and the upgrade simply unlocks a higher amount of service revenue per rack?
Not in the last few years. CPUs went from ~200W TDP to 500W.
And they went from zero to multiple GPUs per server. Tho we might hit "the chips can't be bigger and the cooling can't get much better" point there.
The usage would be similar if it was say a rack filled with servers full of bulk storage (hard drives generally keep the power usage similar while growing storage).
But CPU/GPU wise, it's just bigger chips/more chiplets, more power.
I'd imagine any flattening might be purely because "we have DC now, re-building cooling for next gen doesn't make sense so we will just build servers with similar power usage as previously", but given how fast AI pushed the development it might not happen for a while.
> Isn't it more likely that the amount of power/heat generated per rack will stay constant over each upgrade cycle,
Power density seems to grow each cycle. But eventually your DC hits power capacity limits, and you have to leave racks empty because there's no power budget.
It depends on how much profit you are making. As long as you can still be profitable on the old hardware you don't have to upgrade.
That's the thing though: a competitor with better power efficiency can undercut you and take your customers
Or they could charge the same as you and make more money per customer. If they already have as many customers as they can handle doing that may be better than buying hardware to support a larger number of customers.
> AI accelerator cards all start dying around the 5 year mark,
More likely the technology will be much better in 5 years in terms of hardware that it is (very) uneconomical to run anything on old hardware.
Given power and price constraints, it's not that you cannot run them in 5 years time it's that you don't want to run them in 5 years time and neither will anyone else that doesn't have free power.
It’s not about assumptions on the hardware. It’s about the current demands for computation and expected growth of business needs. Since we have a couple years to measure against it should be extremely straightforward to predict. As such I have no reason to doubt the stated projections.
> Since we have a couple years to measure against
Trillion pound baby fallacy.
Networking gear was famously overbought. Enterprise hardware is tricky as there isn’t much of a resale market for this gear once all is said and done.
The only valid use case for all of this compute which could reasonably replace ai is btc mining. I’m uncertain if the increased mining capacity would harm the market or not.
BTC mining on GPUs haven't been profitable for a long time, it's mostly ASICs, GPUs can be used for some other altcoins which makes the potential market for used previous generation GPUs even smaller.
That assumes you can add compute in a vacuum. If your altcoin receives 10x compute then it becomes 10x more expensive to mine.
That only scales if the coin goes up in value due to the extra "interest". Which isn't impossible but there's a limit, and it's more often to happen to smaller coins.
Failure rates also go up. For AI inference it’s probably not too bad in most cases, just take the node offline and re-schedule the jobs to other nodes.
There is the opportunity cost of using a whole datacenter to house ancient chips, even if they're still running. You're thinking like a personal use chip which you can run as long as it is non-defective. But for datacenters it doesn't make sense to use the same chips for more than a few years and I think 5 years is already stretching their real shelf life.
Do not forget that we're talking about supercomputers. Their interconnect makes machines not easily fungible, so even a low reduction in availability can have dramatic effects.
Also, after the end of the product life, replacement parts may no longer be available.
You need to get pretty creative with repair & refurbishment processes to counter these risks.
It's option #2. But 5 year deprecation is optimistic; 2-3 years is more realistic.
5 years is maybe referring to the accounting schedule for depreciation on computer hardware, not the actual useful lifetime of the hardware.
It's a little weird to phrase it like that though because you're right it doesn't mean you have to throw it out. Idk if this is some reflection of how IBM handles finance stuff or what. Certainly not all companies throw out hardware the minute they can't claim depreciation on it. But I don't know the numbers.
Anyways, 5 years is an infection point on numbers. Before 5 years you get depreciation to offset some cost of running. After 5 years, you do not, so the math does change.
that is how the investments are costed though, so makes sense when we're talking return on investment, so you can compare with alternatives under the same evaluation criteria.
Historically, GPUs have improved in efficiency fast enough that people retired their hardware in way less than 5 years.
Also, historically the top of the line fabs were focused on CPUs, not GPUs. That has not been true for a generation, so it's not really clear if the depreciation speed will be maintained.
> Historically, GPUs have improved in efficiency fast enough that people retired their hardware in way less than 5 years.
This was a time when chip transistor cost was decreasing rapidly. A few years earlier even RAM cost was decreasing quickly. But these times are over now. For example, the PlayStation 5 (where the GPU is the main cost), which launched five years ago, even increased in price! This is historically unprecedented.
Most price/performance progress is nowadays made via better GPU architecture instead, but these architectures are already pretty mature, so the room for improvement is limited.
Given that the price per transistor (which TSMC & Co are charging) has decreased ever more slowly in recent years, I assume it will eventually come almost to a halt.
By the way, this is strictly speaking compatible with Moore's law, as it is only about transistors per chip area, not price. Of course the price per chip area was historically approximately constant, which meant exponentially increasing transistor density implied exponentially decreasing transistor price.
> that people retired their hardware in way less than 5 years.
those people are end-consumers (like gamers), and only recently, bitcoin miners.
Gamers don't care for "profit and loss" - they want performance. Bitcoin miners do need to switch if they want to keep up.
But will an AI data center do the same?
Mining bitcoin with a GPU hasn't been profitable in over a decade.
The rate of change is equal for all groups. The gaming market can be the most conservative since it’s just luxury.
It's worse than that in reality, AI chips are on a two year cadence for backwards compatibility (NVIDIA can basically guarantee it, and you probably won't be able to pay real AI devs enough to stick around to make hardware work arounds). So their accounting is optimistic.
5 years is normal-ish depreciation time frame. I know they are gaming GPUs, but the RTX 3090 came out ~ 4.5 years before the RTX 5090. The 5090 has double the performance and 1/3 more memory. The 3090 is still a useful card even after 5 years.
RTX 3090 MSRP: 1500 USD
RTX 5090 MSRP: 2000 USD
When you operate big data centers it makes sense to refresh your hardware every 5 years or so because that’s the point at which the refreshed hardware is enough better to be worth the effort and expense. You don’t HAVE to, but its more cost effective if you do. (Source, used to operate big data centers)
Actually my biggest issue here is that, assuming it hasnt paid off, you dont just convert to regular data center usage.
Honestly if we see a massive drop in DC costs because the AI bubble bursts I will be stoked.
> In an October letter to the White House's Office of Science and Technology Policy, OpenAI CEO Sam Altman recommended that the US add 100 gigawatts in energy capacity every year.
> Krishna also referenced the depreciation of the AI chips inside data centers as another factor: "You've got to use it all in five years because at that point, you've got to throw it away and refill it," he said.
And people think the climate concerns of AI are overblown. Currently US has ~1300 GW of energy capacity. That's a huge increase each year.
100GW per year is not going to happen.
The largest plant in the world is the Three Gorges Dam in China at 22GW and it’s off the scales huge. We’re not building the equivalent of four of those every year.
Unless the plan is to power it off Sam Altman’s hot air. That could work. :)
https://en.wikipedia.org/wiki/List_of_largest_power_stations
China added ~90GW of utility solar per year in last 2 years. There's ~400-500GW solar+wind under construction there.
It is possible, just may be not in the U.S.
Note: given renewables can't provide base load, capacity factor is 10-30% (lower for solar, higher for wind), so actual energy generation will vary...
> China added ~90GW of utility solar per year in last 2 years. There's ~400-500GW solar+wind under construction there.
Source?
> It is possible
Sure, GP was clearly talking about the US, specifically.
> just may be not in the U.S.
Absolutely 100% not possible in the US. And even if we could do it, I'm not convinced it would be prudent.
If you really had to you'd probably run turbines off natural gas but it's not a good idea environmentally.
I am a huge proponent of renewables, but you cannot compare their capacity in GW with other energy sources because their output is variable and not always maximal. To realistically get 100GW in solar you would need at least 500GW of panels.
On the other hand, I think we will not actually need 100GW of new installations because capacity can be acquired by reducing current usage by making it more efficient. The term negawatt comes to mind. A lot of people are still in the stone age when it comes to this even though it was demonstrated quite effectively by reduced gas use in the US after the oil crisis in the 70s. Which basically recovered to the pre crisis levels only recently.
High gas prices caused people to use less and favor efficiency. The same thing will happen with electricity and we'll get more capacity. Let the market work.
Amazing that 4 of the top 5 are renewables in China.
> As of 2025, The Medog Dam, currently under construction on the Yarlung Tsangpo river in Mêdog County, China, expected to be completed by 2033, is planned to have a capacity of 60 GW, three times that of the Three Gorges Dam.[3]
Meanwhile, “drill baby drill!”
Can run the UK and have capacity left over that, if considered alone, would be worlds highest in current year 2025.
Because its cheaper. That's it.
Does that cout the dams that flood valleys and displace thousands of people, plants, and animals from their homes?
Not really that surprising.
Authoritarianism has its draw backs obviously but one of its more efficient points is it can get things done if the will is at the top. Since China doesnt have a large domestic oil supply like the US it is a state security issue to get off oil as fast as possible.
It’s become clear that some form of top down total technocratic control like China has implemented is essential for pushing humanity forward.
New datacenters are being planned next to natgas hubs for a reason. They’re being designed with on site gas turbines as primary electricity sources.
Gigawatts? Pshaw. We have SamaWatts.
LOL, maybe Sam Altman can fund those power plants. Let me guess: He'd rather the public pay for it, and for him to benefit/profit from the increased capacity.
Big tech is going to have to fund the plants and probably transmission. Because the energy utilities have a decades long planning horizon for investments.
Good discussion about this in recent Odd Lots podcast.
I've read a bunch of the opposite: a lot of secret deals between tech companies and utilities, where when details come out, we find that regular ratepayers are going to be paying a decent chunk of the cost.
That was discussed in the Odd Lots episode as well. But that can only happen in unregulated markets.
Scam Altman wants the US to build a lot of energy plants so that the country will pay the costs and OpenAI will have the profits of using this cheap energy.
If we moron our way to large-scale nuclear and renewable energy rollout however..
I highly doubt this will happen. It will be natural gas all the way, maybe some coal as energy prices will finally make it profitable again.
If for no other reason than they're actively attacking renewable capacity even amid surging demand
Guess who's going to pay nothing for power? Hint: it's not you, and it's not me.
This admin has already killed as much solar and wind and battery as it can.
The only large scale rollout will be payment platforms that will allow you to split your energy costs into "Five easy payments"
There's a reason Trump is talking about invading Venezuela (hint: it's because they have the largest oil deposits).
I would add an addendum to this -- there is no way the announced spending on AI data centers will all come to fruition. I have no doubt that there will be a massive build-out of infrastructure, but it can't reach the levels that have been announced. The power requirements alone will stop that from happening.
The power requirement is only an issue in western countries, where utilities build at most a double digit buffer, and are used to overall energy use leveling due to efficiency improvements. Now look at China where they routinely maintain a 100% buffer. Demand can double and they can supply that without new generation capacity.
I think you're spot on. OpenAI alone has committed to spending $1.4T on various hardware/DCs. They have nowhere near that amount of money and when pushed Altman gets defensive.
https://techcrunch.com/2025/11/02/sam-altman-says-enough-to-...
[flagged]
The incentive for CEOs is announcing the plan to do something, they have no idea if they will actually be able to do it, and it probably won't matter.
This happened in the dotcom too btw. Companies built out fibre networks, it wasn't possible to actually build all the physical infra that companies wanted to build so many announced plans that never happened and then, towards the end, companies began aggressively acquiring stakes in companies who were building stuff to get financial exposure (an example was BT, which turned itself briefly into a hedge fund with a telephone network attached...before it imploded).
CEOs do not operate on the timescale of waiting and building. Their timescale is this year's bonus/share options package. Nothing else matters: announce plans to do X or Y, doesn't matter, they know they will be gone long before it happens.
I think this was definitely true of CEOs in the past but Google and Meta have managed spectacular infrastructure buildouts over many decades with proper capacity planning.
There were a lot of lessons learned in the dotcom boom (mainly related to the great telecom heist if you ask me). If you look back on why there was dotcom bubble it's broadly, imho, related to the terrible network quality in the US compared to other first world countries.
Our cellular services for example lag behind basically every asian market by at least 1 maybe 2 generations now.
It took about 10 months to get 200k GPUs going in xAI data centers. Given the stuff has been being announced for the past 2 years, we shall see how they’re going within the next year.
I believe there is a difference between what people say publicly and what they are actually committed to doing on the ground. When all is said and done, I'll be more interested to know what was actually spent.
For example, XYZ AI company may say they are going to spend $1T for AI data centers over the next 5 years.
In actuality, I suspect it is likely that they have committed to something like $5-10B in shovel-ready projects with stretch goals for the rest. And the remaining spend would be heavily conditioned -- is power available? are chips available? is the public support present? financing? etc...
Not to mention, it's a much bigger moat if you can claim you're going to spend $1T. Who else will want to compete with you when you're spending $1T. After the dust has settled and you've managed to be one of the 2-3 dominant AI players, who is going to care that you "only" spent $100B instead of $1T. Look -- you were very capital efficient!
So, do I see it as possible that XYZ AI company could spend $1T, sure. Is it likely? No.
Hmm... "CEOs and teams" don't necessary do what's makes sense mathematically. Many, if not most of them, do whatever that sounds good to shareholders in their quarterly earnings call and ignore the reality or long term development.
If "CEOs and teams" are smart enough, they would not have overhired during 2021-2022 and then do layoffs. Who would be dumb enough to do that?
Appeal to authority fallacy.
me very smart. me go college get fancy paper. me move money out of ai stocks. flat rocks not think so well, use too much fire. me trade for shiny yellow rock instead.
What qualifies you to question this?
lol, CEOs do not do research.
^ they are a mouthpiece to manipulate the market, not a research scientist.
It's not going to pay off for everybody, this is a land grab for who will control this sort of central aspect of AI inference.
[delayed]
If AI is a highlander market, then the survivor will be able to eventually aquire all those assets on the cheap from the failing competitors that flush their debt in bankruptcy.
Meanwhile, highlander hopefuls are spending other peoples money to compete. Some of them with dreams of not just building a tech empire, but to truly own the machine that will rule the world in every aspect.
Investors are keen on backing the winner. They just do not know yet who it will be.
Until China sees it valuable to fund open weights SOTA-ish models, even the winner might struggle. There is very little capture - protocols are mostly standard so models are mostly interchangeable and if you are trying to raise prices enough to break even on the whole operation, somebody else can probably profitably run inference cheaper.
"commoditize your complements" and all that.
"oh look, with open weights now everyone can run this quadrillion param 'superintelligent' model. but what's that? only we have the power capacity and cheap $/W to actually serve it? funny thing...!"
What's a highlander market, or a highlander hopeful? Google wasn't helpful.
I interpreted it as "there can only be one" which I believe is a quote from the Highlander movie; it's a "winner takes all" and that winner gets the title of "highlander."
In this situation then everyone who _isn't_ the winner will go broke -> sell off all their stuff on the cheap because they're desperate -> the winner gets all their hardware for a great deal and becomes even more powerful.
He's right to question the economics. The AI infrastructure buildout resembles the dot-com era's excess fiber deployment - valuable long-term, but many individual bets will fail spectacularly. Utilization rates and actual revenue models matter more than GPU count.
I disagree on that and covered a lot of it in this blog (sorry for the plug!) https://martinalderson.com/posts/are-we-really-repeating-the...
100% of technical innovations have had the same pattern. The same thing happens every time because this is the only way the system can work: excess is required because there is some uncertainty, lots of companies are designing strategies to fill this gap, and if this gap didn't exist then there would be no investment (as happens in Europe).
Also, demand wasn't over-estimated in the 2000s. This is all ex-post reasoning you use data from 2002 to say...well, this ended up being wrong. Companies were perfectly aware that no-one was using this stuff...do you think that telecoms companies in all these countries just had no idea who was using their products? This is the kind of thing you see journalists write after the event to attribute some kind of rationality and meaning, it isn't that complicated.
There was uncertainty about how things would shake out, if companies ended up not participating then CEOs would lose their job and someone else would do it. Telecoms companies who missed out on the boom bought shares in other telecom's companies because there was no other way to stay ahead of the news and announce that they were doing things.
This financial cycle also worked in reverse twenty years later too: in some countries, telecoms companies were so scarred that they refused to participate in building out fibre networks so lost share and then ended up doing more irrational things. Again, there was uncertainty here: incumbents couldn't raise from shareholders who they bankrupted in fiber 15 years ago, they were 100% aware that demand was outstripping supply, and this created opportunities for competitors. Rationality and logic run up against the hard constraints of needing to maintain a dividend yield and the exec's share options packages.
Humans do not change, markets do not change, it is the same every time. What people are really interested in is the timing but no-one knows that either (again, that is why the massive cycle of irrationality happens)...but that won't change the outcome. There is no calculation you can make to know more, particularly as in the short-term companies are able to control their financial results. It will end the same way it ended every time before, who knows when but it always ends the same way...humans are still human.
> Also, demand wasn't over-estimated in the 2000s. This is all ex-post reasoning you use data from 2002 to say...well, this ended up being wrong.
Well, the estimate was higher than the reality, by definition it was over-estimated. They built out as if the tech boom was going to go on forever, and of course it didn't. You can argue that they made the best estimates they could with the information available, but ultimately it's still true that their estimates were wrong.
Your blog article stopped at token generation... you need to continue to revenue per token. Then go even further... The revenue for AI company is a cost for the AI customer. Where is the AI customer going to get incremental profits from the cost of AI.
For short searches, the revenue per token is zero. The next step is $20 per month. For coding it's $100 per month. With the competition between Gemini, Grok, ChatGPT... it's not going higher. Maybe it goes lower since it's part of Google's playbook to give away things for free.
Great article, thank you for writing and sharing it!
Fiber seems way easier to get long-term value out of then GPUs, though. How many workloads today other than AI justify massive GPU deployments?
Would be funny if all the stagnant GPUs finally brought game streaming to the mainstream.
They discuss it in the podcast. Laid fiber is different because you can charge rent for it essentially forever. It seems some people swooped in when it crashed and now own a perpetual money machine.
IBM might not have a data strategy or AI plan but he isn’t wrong on the inability to generate a profit.
A bit of napkin math: NVIDIA claims 0.4J per token for their latest generation 1GW plant with 80% utilisation can therefore produce 6.29 10^16 tokens a year.
There are ~10^14 tokens on the internet. ~10^19 tokens have been spoken by humans… so far.
> There are ~10^14 tokens on the internet.
Don't know what the source is, but it feels missing a few orders of magnitude. Surely it only counts text? I can't imagine there are only so few data on the internet if you count images and videos.
> ~10^14 tokens on the internet
Does that include image tokens? My bet is with image tokens you are off by at least 5 orders of magnitude for both.
Images are not that big. Each text token is a multidimensional vector.
There were recent observations that rendering the text as an image and ingesting the image might actually be more efficient than using text embedding.
I must be dense, why does this imply AI can't be profitable?
Tokens are, roughly speaking, how you pay for AI. So you can approximate revenue by multiplying tokens per year by the revenue for a token.
(6.29 10^16 tokens a year) * ($10 per 10^6 tokens)
= $6.29 10^11
= $629,000,000,000 per year in revenue
Per the article
> "It's my view that there's no way you're going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest," he said.
$629 billion is less than $800 billion. And we are talking raw revenue (not profit). So we are already in the red.
But it gets worse, that $10 per million tokens costs is for GPT-5.1, which is one of the most expensive models. And the costs don't account for input tokens, which are usually a tenth of the costs of output tokens. And using bulk API instead of the regular one halves costs again.
Realistic revenue projections for a data center are closer to sub $1 per million tokens, $70-150 billion per year. And this is revenue only.
To make profits at current prices, the chips need to increase in performance by some factor, and power costs need to fall by another factor. The combination of these factors need to be, at minimum, like 5x, but realistically need to be 50x.
The math here is mixing categories. The token calculation for a single 1-GW datacenter is fine, but then it gets compared to the entire industry’s projected $8T capex, which makes the conclusion meaningless. It’s like taking the annual revenue of one factory and using it to argue that an entire global build-out can’t be profitable. On top of that, the revenue estimate uses retail GPT-5.1 pricing, which is the absolute highest-priced model on the market, not what a hyperscaler actually charges for bulk workloads. IBM’s number refers to many datacenters built over many years, each with different models, utilization patterns, and economics. So this particular comparison doesn’t show that AI can’t be profitable—it’s just comparing one plant’s token output to everyone’s debt at once. The real challenges (throughput per watt, falling token prices, capital efficiency) are valid, but this napkin math isn’t proving what it claims to prove.
> but then it gets compared to the entire industry’s projected $8T capex, which makes the conclusion meaningless.
Aren't they comparing annual revenue to the annual interest you might have to pay on $8T? Which the original article estimates at $800B. That seems consistent.
im a little confused about why you are using revenue for a single datacenter against interest payments for 100 datacenters
Broad estimates I'm seeing on the cost of a 1GW AI datacenter are $30-60B. So by your own revenue projection, you could see why people are thinking it looks like a pretty good investment.
Note that if we're including GPU prices in the top-line capex, the margin on that $70-150B is very healthy. From above, at 0.4J/T, I'm getting 9MT/kWh, or about $0.01/MT in electricity cost at $0.1/kWh. So if you can sell those MT for $1-5, you're printing money.
> So if you can sell those MT for $1-5, you're printing money.
The IF is doing a lot of heavy lifting there.
I understood the OP in the context of "human history has not produced sufficiently many tokens to be sent into the machines to make the return of investment possible mathematically".
Maybe the "token production" accelerates, and the need for so much compute realizes, who knows.
There's really 3 fears going on:
1. The devil you know (bubble)
2. The devil you don't (AI global revolution)
3. Fear of missing out on devil #2
I don't think IBM knows anything special. It's just more noise about fear1 & fear3.
From the company that said the world market size for computers was about 50. And the company that gave us OS2.
I see a lot of people attacking the messenger but very few addressing the basic logic that you need 800B+ in profit just to pay the interest on some of these investments.
Pointing out IBM's mixed history would be valid if they were making some complex, intricate, hard to verify case for why AI won't be profitable. But the case being made seems like really simple math. A lot of the counterarguments to these economic problems have the form "this time it's different" - something you hear every bubble from .com to 2008.
The interesting macro view on what's happening is to compare a mature data center operation (specifically a commoditized one) with the utility business. The margins here, and in similar industries with big infra build-out costs (ex: rail) are quite small. Historically the businesses have not done well; I can't really imagine what happens when tech companies who've only ever known huge, juicy margins experience low single digit returns on billions of investment.
Worse, is that a lot of these people are acting like Moore's law isn't still in effect. People conflate clock speeds on beefy hardware with moore's law, and act like it's dead, when transistor density rises, and cost per transistor continue to fall at rates similar to what they always have. That means the people racing to build out infrastructure today might just be better off parking that money in a low interest account, and waiting 6 months. That was a valid strategy for animation studios in the late 90s (it was not only cheaper to wait, but also the finished renders happened sooner), and I'd be surprised if it's not a valid strategy today for LLMs. The amount of silicon that is going to be produced that is specialized for this type of processing is going to be mind boggling.
Cost per transistor is increasing. or flat, if you stay on a legacy node. They pretty much squeezed all the cost out of 28nm that can be had, and it’s the cheapest per transistor.
“based on the graph presented by Milind Shah from Google at the industry tradeshow IEDM, the cost of 100 million transistors normalized to 28nm is actually flat or even increasing.”
https://www.tomshardware.com/tech-industry/manufacturing/chi...
Yep. Moore's law ended at or shortly before the 28nm era.
That's the main reason people stopped upgrading their PCs. And it's probably one of the main reasons everybody is hyped about Risc-V and the pi 2040. If Moore's law was still in effect, none of that would be happening.
That may also be a large cause of the failure of Intel.
> Moore's law ended at or shortly before the 28nm era.
Moore's law isn't about cost or clock speed, it's about transistor density. While the pace of transistor density increases has slowed, it's still pretty impressive. If we want to be really strict, and say densities absolutely have to double every 2 years, Moore's Law hasn't actually been true since 1983 or so. But it's been close, so 2x/2yr a decent rubric.
The fall-off from the 2x/2yr line started getting decently pronounced in the mid 90s. At the present time, over the past 5-6 years, we're probably at a doubling in density every 4-ish years. Which yes, is half the rate Moore observed, but is still pretty impressive given how mature the technology is at this point.
If you want to be pedantic, the original (and revised) law are definitely about cost. The original formulation was that the number of features (i.e. transistors) on an integrated circuit doubled every two years for the best-priced chips (smallest cost per feature).
https://web.archive.org/web/20220911094433/https://newsroom....
> The complexity for minimum component costs has increased at a rate of roughly a factor of two per year (see graph on next page).
And it is formulated in a section aptly titled "Costs and curves". This law has always been an economic law first, some kind of roadmap for fans to follow. But that roadmap drove almost-exponential investment costs as well.
I concede that density still rises, especially if you count " advanced packaging". But the densest and most recent is not the cheapest anymore.
A lot of it is propped by the fact with GPU and modern server CPUs the die area just got bigger
Does AWS count as commoditized data center? Because that is extremely profitable.
Or are you talking abour things like Hetzner and OVH?
The cloud mega scalers have done very well for themselves. As with all products the question is differentiation. If models can differentiate and lock in users they can have decent margins. If models get commoditized the current cloud providers will eat the AI labs lunch.
I question depreciation. those gpu's will be obsolete in 5 years, but will the newer be enough better as to be worth replacing them is an open question. cpu's stopped getting exponetially faster 20 years ago, (they are faster but not the jumps the 1990s got)
I recently compared performance per dollar for CPUs and GPUs on benchmarks for GPUs today vs 10 years ago, and suprisingly, CPUs had much bigger gains. Until I saw that for myself, I thought exactly the same thing as you.
It seems shocking given that all the hype is around GPUs.
This probably wouldn't be true for AI specific workloads because one of the other things that happened there in the last 10 years was optimising specifically for math with lower size floats.
That makes sense. Nvidia owns the market and is capturing all the surplus value. They’re competing with themselves to convince you to buy a new card.
It's coz of use cases. Consumer-wise, if you're gamer, CPU just needs to be at "not the bottleneck" level for majority of games as GPU does most of the work when you start increasing resolution and details.
And many pro-level tools (especially in media space) offload to GPU just because of so much higher raw compute power.
So, basically, for many users the gain in performance won't be as visible in their use cases
> those gpu's will be obsolete in 5 years, but will the newer be enough better as to be worth replacing them is an open question
Doesn't one follow from the other? If newer GPUs aren't worth an upgrade, then surely the old ones aren't obsolete by definition.
MTBF for data center hardware is short; DCs breeze through GPUs compared to even the hardest of hardcore gamers.
And there is the whole FOMO effect to business purchases; decision makers will worry their models won't be as fast.
Obsolete doesn't mean the reductive notion you have in mind, where theoretically it can still push pixels. Physics will burn them up, and "line go up" will drive demand to replace them.
Source? Anecdotally, GPUs sourced from cryptomining were absolutely fine MTBF-wise. Zero apparent issues of wear-and-tear or any shortened lifecycle.
My bellybutton fluff, uninformed opinion is that heat cycling and effective cooling are probably a much more limiting factor.
If you are running a gpu at 60C for months at a time, but never idling it (crypto use case), I would actually hazard a guess that it is better than cycling it with intermittent workloads due to thermal expansion.
That of course presupposes effective, consistent cooling.
Anecdotally, I killed two out of two that I was hobby-mining on for a couple of years. They certainly didn't sound like they would work forever.
There is the question - will they be worth the upgrade? Either because they are that much faster, or that much more energy efficient. (and also assuming you can get them, unobtainium is worth that what you have).
Also a nod to the other reply that suggests they will wear out in 5 years. I cannot comment on if that is correct but it is a valid worry.
It's not that hard to see the old GPUs being used e.g. for inference on cheaper models, or sub-agents, or mid-scale research runs. I bet Karpathy's $100 / $1000 nanochat models will be <$10 / <$100 to train by 2031
I think real issue is current costs / demand = Nvidia gouging GPU price that costs for hardware:power consumption is 70:20 instead of 50:40 (10 for rest of datacenter). Reality is gpus are serendipidous path dependent locked from gaming -> mining. TPUs are more power efficient, if bubble pops and demand for compute goes down, Nvidia + TMSC will still be around, but nexgen AI first bespoke hardware premium will revert towards mean and we're looking at 50% less expensive hardware (no AI race scarcity tax, i.e. 75% Nvidia margins) that use 20% less power / opex. All of a sudden existing data centers becomes not profitable stranded assets even if they can be stretched past 5 years.
> those gpu's will be obsolete in 5 years, but will the newer be enough better as to be worth replacing them
Then they won't be obsolete.
"It is 1958. IBM passes up the chance to buy a young, fledgling company that has invented a new technology called xerography. Two years later, Xerox is born, and IBM has been kicking themselves ever since. It is ten years later, the late '60s. Digital Equipment DEC and others invent the minicomputer. IBM dismisses the minicomputer as too small to do serious computing and, therefore, unimportant to their business. DEC grows to become a multi-hundred-million dollar corporation before IBM finally enters the minicomputer market. It is now ten years later, the late '70s. In 1977, Apple, a young fledgling company on the West Coast, invents the Apple II, the first personal computer as we know it today. IBM dismisses the personal computer as too small to do serious computing and unimportant to their business." - Steve Jobs [1][2][3]
Now, "IBM CEO says there is 'no way' spending on AI data centers will pay off". IBM has not exactly had a stellar record at identifying the future.
[1] https://speakola.com/ideas/steve-jobs-1984-ad-launch-1983
[2] https://archive.org/details/1983-10-22-steve-jobs-keynote
[3] https://theinventors.org/library/inventors/blxerox.htm
> IBM has not exactly had a stellar record at identifying the future.
IBM invented/developed/introduced magnetic stripe cards, UPC Barcodes, the modern ATM, Hard drives, floppies, DRAM, SQL, the 360 Family of Mainframes, the PC, Apollo guidance computers, Deep Blue. IBM created a far share of the future we're living in.
I'm no fan of much of what IBM is doing at the moment but it could be argued that its consultancy/service orientation gives it a good view of how business is and is planning to use AI.
They also either fairly accurately predicted the death of HDDs by selling off their research division before the market collapsed, or they caused the end of the HDD era by selling off their research division. They did a lot of research.
I think the retail market is maybe dead but datacenters are still a fairly large customer I’d think. HDDs really shine at scale where they can be fronted by flash and DRAM cache layers.
They are still cheaper than flash for cold data, but that’s not going to hold for long. Flash is so much denser the acquisition cost difference for a multi-petabyte store becomes small next to the datacenter space and power needed by HDDs. HDDs require research for increasing density while flash can rely on silicon manufacturing advances for that - not that it doesn’t require specific research, but being able to apply the IP across a vast space makes better economical sense.
The hdd being dead will surely come as a surprise to the couple of 12TB rusties spinning joyously in my case right now.
One family down the street from me still drives a Saturn! Yet no one is going to say "Saturn isn't dead!".
HDDs would be much more important today if flash storage didn’t exist.
did you know that SSD are not memory stable if they dont get electricity every now and again...
Spinning platters are prone to catastrophic mechanical failure and tape can undergo delamination. What of it?
The other way to look at it is that the entire consulting industry is teetering on catastrophe. And IBM, being largely a consulting company now, is not being spared.
IBM isn't failing, though. They're a profitable company with healthy margins, and enterprises continue to hire them for all sorts of things, in large numbers.
So now it makes more sense why they think the AI spending will/need to fail.
Because if it didn't, that's a direct replacement of them.
> The other way to look at it is that the entire consulting industry is teetering on catastrophe
Oh? Where'd you get that information?
If you mean because of AI, it doesn't seem to apply much to IBM. They are probably not great at what they do like most such companies, but they are respectable and can take the blame if something goes wrong. AI doesn't have these properties.
If anything there’s likely plenty of work for body shops like IBM in reviewing and correcting AI-generated work product that has been thrown into production recently.
This is a separate argument though. A failing company may still be right in identifying other companies failure modes.
You can be prescient about failure in one area and still fail yourself. There's no gotcha.
IBM is not a failing company though, they are a Goliath in the Enterprise space.
Still besides the point. The company failing or not is orthogonal to them being able to identify failure in others.
> A failing company may still be right in identifying other companies failure modes.
Agreed if this is what they are doing, but what if theyre spewing claims to try and discredit an industry in order to quell their shareholder concerns?
They are not the only ones looking at the money spent in AI datacentres and concluding most of the investment will not be recovered anytime soon.
A lot of the silicon being deployed is great for training, but inefficient for inference and the training to inference ratio for usage shows a clear tendency to go the inference way. Furthermore, that silicon, with the workloads it runs, doesn’t last long and needs replacement.
The first ones to go online might recover the investment, but the followers better have a plan to pivot to other uses.
IBM was making "calculating cheese cutters" back in the day [0].
I'm sure they can pivot to something else if the need arises.
[0]: https://imgur.com/a/ibm-cheese-cutter-Rjs2I
The whole point of a consultant is to let the execs blame someone else.
Nobody got fired for buying something Gartner recommended, or for following EY's advice to lay off/hire
I don't see AI taking that blame away.
They own Red Hat Linux, Ansible, OpenShift, and Terraform.
If you are doing anything in the Enterprise space, they probably have their claws in you be it on-prem or cloud.
And their work on quantum...
https://www.forbes.com/sites/baldwin/2025/11/25/inside-ibms-...
Not to mention they are still doing quite a bit of Mainframe...
I've heard some second hand stories about IBM's way of using "AI" and it is pretty much business oriented and not much of the glamour and galore promises the other companies make (of course you still have shiny new things in business terms). It's actually good entertainment hearing all the internal struggles of business vs fancy during the holidays.
IBM is/was good at inventing a lot of tech.
It may not be good at recognizing other good tech invented or paradigm changes by others
For the fact that they invented Deep Blue, they are really struggling with AI
Their Granite family of models is actually pretty good! They just aren't working on the mainstream large LLMs that capture all the attention.
IBM is always very conscious of what their clients need (and the large consultancy business provides a very comprehensive view). It just turns out their clients don’t need IBM to invest in large frontier models.
ibm developed SSMs/mamba models and also releasing trainings datasets i think, also quantum computing is strategic option..
For sure but do you see them at any relevant leader boards? Any news how good they are?
I don't.
I know their models, but not because i constantly read about it
> IBM invented/developed/introduced magnetic stripe cards, UPC Barcodes, the modern ATM, Hard drives, floppies, DRAM, SQL, the 360 Family of Mainframes, the PC, Apollo guidance computers, Deep Blue. IBM created a far share of the future we're living in.
Well put. “IBM was wrong about computers being a big deal” is a bizarre take. It’s like saying that Colonel Sanders was wrong about chicken because he, uh… invented the pressure fryer.
This is the exact kind of thinking that got us into this mess in the first place, and I'm not blaming you for it, it seems to be something all of us do to an extent. We don't look to Meta, who only a few years ago thought that the Metaverse would be the "next big thing" as an example of failure to identify the future, we look to IBM who made that mistake almost 30 years ago. Underestimating a technology seems to stick much harder than overestimating one.
If you want to be seen as relevant in this industry, or as a kind of "thought leader", the easy trick seems to be to hype up everything. If you do that and you're wrong, people will quickly forget. If you don't and you're wrong, that will stain your reputation for decades.
Good point. That kind of thinking is an absurdity. Saying IBM dropped the ball 70 years ago without acknowledging that lessons were learned, leadership has changed hands a lot since then, and most importantly, the tech landscape back then was very different from today unless you grossly oversimplify everything amounts to nothing more than a fallacious opinion.
Not even much of an IBM fan, myself, but I respect their considerable contribution to the industry. Sure, they missed a shot back then, but I think this latest statement is reliably accurate based on the information we currently have.
It’s easy to be a pessimist. Most things don’t work. So in 9 out of 10 cases you’re right.
But human breakthrough progress came mostly through optimists, who tried things no one else dared to do.
The amount of hate I've received here for similar statements is astonishing. What is even more astonishing is that it takes 3-rd grade math skills to work out that the current AI(even ignoring the fact that there is nothing intelligent about the current AI) costs are astronomical and they do not deliver on the promises and everyone is operating at wild loses. At the moment we are at "if you owe 100k to your bank, you have a problem but if you owe 100M to your bank, your bank has a problem". It's the exact same bullshitter economy that people like musk have been exploiting for decades: promise a ton, never deliver, make a secondary promise for "next year", rinse and repeat -> infinite profit. Especially when you rope in fanatical followers.
I don't want to defend musk in any way but I think you are making a mistake there using him as an example because what boosted him quite a lot is that he actually delivered what he claimed. Always late but still earlier than anybody was guesstimating. And now he is completely spiraling but its a lot harder to lose a billion than to gain one so he persists and even gets richer. Plus his "fanatical" followers are poor. It just doesn't match the situation.
Sounds a lot like "I'm not racist but". There's a website dedicated to all of his bs https://elonmusk.today
He is the definition of a cult. Collects money from fanatical followers who will praise every word he says, never delivers, "oh next year guys, for sure, wanna buy a not a flamethrower, while you are at it?". Not to mention what once were laughable conspiracy theories about him turned out to be true(such that even I laughed when I heard them). Torvalds is right with his statement about musk: "incompetent" and "too stupid to work at a tech company".
The last sentence sounds a lot like a (partial?) Ponzi scheme.
Pretty close really. Just enough to buy some plausible deniability I suppose.
> We don't look to Meta, who only a few years ago thought that the Metaverse would be the "next big thing" as an example of failure to identify the future, we look to IBM who made that mistake almost 30 years ago.
The grandparent points to a pattern of failures whereas you point to Meta’s big miss. What you miss about Meta, and I am no fan, is that Facebook purchased Whatsapp and Instagram.
In other words, two out of three ain’t bad; IBM is zero for three.
While that’s not the thrust of your argument, which is about jumping on the problem of jumping on every hype train, the post to which you reply is not on about hype cycle. Rather, that post calls out IBM for a failure to understand the future of technology and does so by pointing to a history of failures.
> In other words, two out of three ain’t bad; IBM is zero for three.
Many others in this thread have pointed out IBM's achievements but regardless, IBM is far from "zero for three".
> Many others in this thread have pointed out IBM's achievements but regardless, IBM is far from "zero for three".
I was specifically commenting in the context of this thread.* I was not trying to characterize either IBM or Meta except with reference to the arguments offered by this thread’s ancestors.
I understood (and understand) that such scorekeeping of a company as storied as IBM is at best reductive and at worst misrepresentative.
* Your reference to “this thread” actually addresses sibling comments to OP (ggggp), not this thread which was started by gggp.
I read the actual article.
He is pointing out that the current costs to create the data centres means you will never be able to make a profit to cover those costs. $800 Billion just to cover the interest.
OpenAI is already haemorrhaging money and the space data centres has already been debunked. There is even a recent paper that points out that LLMs will never become AGI.
The article also finishes out with some other experts giving the same results.
[edit] Fixed $80 to $800
$800B, to be clear is the claim, not $80B.
Clearly I need to read slower. Thanks. :)
While AGI might be the Holy Grail, AI doesn’t need to be general human-level to be useful and profitable.
it just needs us to wait one more year right?
>> There is even a recent paper that points out that LLMs will never become AGI.
can you share a link?
Took me a while to find again, as there are a lot of such papers in this area.
https://www.arxiv.org/pdf/2511.18517
Is this AI paper written by a reputable subject matter expert? It seems to be written by a physicist and also be the only academic work by this author in English
So you are dismissing it because of that? Certainly read the paper first and attack the arguments, not the author. It even has 10 pages of citations.
I have read it. It is nothing new on the subject, but it was just the recent paper I saw on HN and the person was asking for the link.
The crux is an LLM is and can never be intelligent in the sense of an AGI. It is easier to think of it as a way to store and retrieve knowledge.
How many articles on this topic do we imagine there are? Thousands? Hundreds of thousands? It is hopeless to read every one by any author, no matter how unrelated to the domain, and judge them individually on their merits. Being a subject domain expert is not a perfect measure of paper quality but it's the only feasible way to make a first pass at filtering.
Even if I did read it, I have no hope of understanding if it has made a fundamental mistake because I don't have the subject matter expertise either.
(I imagine it has made a fundamental mistake anyway: for LLMs to be useful progress toward AGI they don't have to be a feasible way to create AGI by themselves. Innovation very often involves stepping through technologies that end up only being a component of the final solution, or inspiration for the final solution. This was always going to be an issue with trying to prove a negative.)
> It is hopeless to read every one by any author,
It was a paper posted on HN a few days ago and someone asked for the evidence of my statement. I supplied it.
Now if they actually read it and disagreed with what it was saying, I'd be more than happy to continue the conversation.
Dismissing it just because you don't understand is a terrible thing to do to yourself. It's basically sabotaging your intelligence.
Sometimes papers are garbage, but you can only make that statement after you have read/understood it.
Use an LLM if you want.
I was really just asking, not trying to be dismissive. Expertise is an important context to evaluate a piece of writing.
Absolutely. If it is not written by someone who has real world experience and deep knowledge it has no more value than a HN comment.
It's a good read and good citations.
The core piece as quoted from the abstract: "AGI predictions fail not from insufficient compute, but from fundamental misunderstanding of what intelligence demands structurally."
Then goes in detail as to what that is and why LLMs don't fit that. There are plenty other similar papers out there.
It was more of a general principle than about specific paper that I mentioned that :)
A single author, in a physics department. Seems unlikely to be groundbreaking or authoritative.
Welcome to the world of papers. Have a read and get back to us. Dismissing out of hand is rarely constructive.
took me a while but i read it. thought it was actually a pretty good and well researched paper that does a good job rationalizing its thesis. thanks for sharing
Ad hominem right out of the gate? Really?
Sry to say but the fact that you argue with LLMs never become AGI, you are not up-to-date.
People don't assume LLM will be AGI, people assume that World Models will lead us to AGI.
I personally never asumed LLM will become AGI, i always assumed that LLM broke the dam for investment and research into massivce scale compute ML learning and LLMs are very very good in showing were the future goes because they are already so crazy good that people can now imagine a future were AGI exists.
And that was very clear already when / as soon as GPT-3 came out.
The next big thing will probably be either a LOT more RL or self propelling ai architecture discovery. Both need massive compute to work well but then will potentially provide even faster progress as soon as humans are out of the loop.
> People don't assume LLM will be AGI,
I wish that was true.
> people assume that World Models will lead us to AGI.
Who are these people? There is no consensus around this that I have seen. You have anything to review regarding this?
> as soon as GPT-3 came out.
I don't think that was true at all. It was impressive when it came out, but people in the field clearly saw the limitations and what it is.
RL isn't magical either. Google AlphaGo as an example often required human intervention to get the RL to work correctly.
AlphaGo Zero doesn't need much human intervention at all.
Regarding world models: All the big ones. LeCun, Demis Hassabis, Fei-Fei Li too. And they are all working on it.
LLMs will definitly play some type of role in AGI. After all you can ask an LLM already a lot of basic things like 'what are common tasks to make a tea'. A type of guide, long term fact memory or whatever this can be called.
Are OpenAI or Anthropic et al seriously building towards “world models”? I haven’t seen any real evidence of that. It seems more like they are all in on milking LLMs for all they are worth.
I mentioned it in my other comment but people like LeCun, Demis Hassabis, Fei-Fei Li do.
There are indications that Open AI is doing this but nothing official as far as i know and i have not heard anything from Anthropic.
> In 1977, Apple, a young fledgling company on the West Coast, invents the Apple II, the first personal computer as we know it today. IBM dismisses the personal computer as too small to do serious computing and unimportant to their business.
IBM released the 5100 in September 1975 [0] which was essentially a personal computer in feature set. The biggest problem with it was the price tag - the entry model cost US$8975, compared to US$1298 for the entry Apple II released in June 1977 (close to two years later). The IBM PC was released in August 1981 for US$1565 for the most basic system (which almost no one bought, so in practice they cost more). And the original IBM PC had model number 5150, officially positioning it as a successor to the 5100.
IBM’s big problem wasn’t that they were disinterested in the category - it was they initially insisted on using expensive IBM-proprietary parts (often shared technology with their mainframe/midrange/minicomputer systems and peripherals), which resulted in a price that made the machine unaffordable for everyone except large businesses, governments, universities (and even those customers often balked at the price tag). The secret of the IBM PC’s success is they told the design team to use commercial off-the-shelf chips from vendors such as Intel and Motorola instead of IBM’s own silicon.
[0] https://en.wikipedia.org/wiki/IBM_5100
And outsourcing the operating system to Microsoft, because they didnt consider it that important.
Got anything vis-a-vis the message as opposed to the messenger?
I'm not sure these examples are even the gotchas you're positing them as. Xerox is a dinosaur that was last relevant at the turn of the century, and IBM is a $300bn company. And if it wasn't obvious, the Apple II never made a dent in the corporate market, while IBM and later Windows PCs did.
In any case, these examples are almost half a century old and don't relate to capex ROI, which was the topic of dicussion.
If it's not obvious, Steve's quote is ENTIRELY about capex ROI, and I feel his quote is more relevant to what is happening today than anything Arvind Krishna is imagining. The quote is posted in my comment not to grandstand Apple in any sense, but to grandstand just how consistently wrong IBM has been about so many opportunities that they have failed to read correctly - reprography, mini computers and microcomputers being just three.
Yes it is about ROI: "IBM enters the personal computer market in November ’81 with the IBM PC. 1983 Apple and IBM emerged as the industry’s strongest competitors each selling approximately one billion dollars worth of personal computers in 1983, each will invest greater than fifty million dollars for R&D and another fifty million dollars for television advertising in 1984 totaling almost one quarter of a billion dollars combined, the shakeout is in full swing. The first major firm goes bankrupt with others teetering on the brink, total industry losses for 83 out shadow even the combined profits of Apple and IBM for personal computers."
I have no horse in this race.
I don’t think this is really a fair assessment. IBM is in fact a huge company today and it is possible that they are because they took the conservative approach in some of their acquisition strategy.
It is a bit like watching someone play poker and fold and then it turns out they had the high hand after all. In hindsight you could of course know that the risk would have been worth it but at the moment perhaps it did not seem like it given the money the first player would be risking.
> I don’t think this is really a fair assessment. IBM is in fact a huge company today and it is possible that they are because they took the conservative approach in some of their acquisition strategy.
I can also imagine IBM was being approached by hundreds, if not thousands, propositions. That they missed three that turned out to be big is a statistical probability.
A big difference is that in the past things like the potential of the PC were somewhat widely underestimated. And then the internet was again as well.
But in modern times it's rather the opposite scenario. The average entity is diving head first into AI simply expecting a revolutionary jump in capability that a more 'informed', for lack of any less snooty term, perspective would suggest is quite unlikely to occur anytime in the foreseeable future. Basically we have a modern day gold rush where companies and taking out unbelievably massive loans to invest in shovels.
The only way this doesn't catastrophically blow up is if AI companies manage to convince the government they're too big to fail, and get the Boeing, Banks, et al treatment. And I expect that's exactly the current strategy, but that's rather a high risk, low reward, type strategy.
>things like the potential of the PC were somewhat widely underestimated.
The potential of the AI that comes within reach at maximum expenditure levels may just be more widely overestimated.
The potential to make "that much money" even more challenging.
A very opposite scenario.
I think so many corporations are looking at how expensive actual humans always have been, and can be sure will always be, so much so that it's a major cost item that can not be ignored. AI opens up the possibility of a whole new level of automation or outright replacement for the routine simple-minded tasks, to a degree that never existed before. More jobs could possibly be eliminated than previous waves of mechanical and digital automation.
When you do the business math, the savings could be enormous.
But you can only realistically save as much as you are actually wasting, otherwise if you go too far you shoot yourself in the foot.
Even with all that money to work with, if you're in practice hunkering down for savings because you can't afford real people any more, you surely can't say the sky's the limit. Not like selling PC's or anything that's capable of more unbridled growth.
When PC's arrived they flew off the shelf even at their high initial retail prices.
People in droves (but not the silent majority) are shunning free AI and the movement is growing with backlash in proportion to the foisting.
I have no special knowledge about IBM Vs Apple historically, but: a quarter billion in CAPEX when you've earned a billion in revenue in a single year is extremely different to what we're seeing now. These companies are spending all of their free cash flow, then taking on debt, to the tune of percentage points of world GDP, and multiples of any revenue they've seen so far. That kind of oversupply is a sure fire way to kill any ROI.
> Got anything vis-a-vis the message as opposed to the messenger?
Sure: People disagree. It's not like there is anything particularly clever that IBM CEO provided here. The guy not investing in something saying it won't work is about as good as the people who do saying it will. It's simply different assumptions about the future.
>the message as opposed to the messenger?
Exactly.
The message is plain to see with very little advanced math.
The only news is that it is the CEO of IBM saying it out loud.
IMHO he has some of the most credible opinions at this scale that many people have seen.
It's "highly unlikely" that all this money will be paid back to everyone that invested at this point. The losers probably will outnumber the winners, and nobody knows whether it will end up becoming a winner-take-all situation yet. A number of wealthy players remain at the table, raising stakes with each passing round.
It's so much money that it's already too late to do anything about it, and the full amount hasn't even changed hands yet.
And the momentum from something so huge can mean that almost the entire amount will have to change hands a second time before a stable baseline can be determined relative to pre-existing assets.
This can take longer than anyone gives credit for just because of massiveness, in the mean time, established real near-term growth opportunities may languish or even fade as the skew in rationality/solvency balance awaits the rolling dice to come to rest.
Would you read this if I (a nobody) told you and not the "CEO of IBM"? In that case it's completely fair to question the messenger.
IBM is an interesting beast when it comes to business decisions. While I can't give exact details, their business intelligence and ability to predict monetary things is uncannily spot-on at times.
So, when their CEO says that this investment will not pay off, I tend to believe them, because they most probably have the knowledge, insight and data to back that claim, and they have ran the numbers.
Oh, also, please let's not forget that they dabbled in "big AI" before everyone else. Anyone remembers Deep Blue and Watson, the original chatbot backed by big data?
As evidenced by the fact that they are a 100+ year old company that still exists. People forget that.
We can cherry-pick blunders made by any big company to make a point. Maybe it would be more honest to also list companies IBM passed on that turned out to be rubbish? And all the technologies that IBM did invest in that made them a ton of money and became industry standards?[0]
Today, Xerox has less total revenue than IBM has profit. DEC went out of business 27 years ago. Apple is an in astoundingly great place right now, but Jobs got kicked out of his own company, and then returned when it was about to fail, having to take investment from Microsoft(!) in order to stay afloat.
Meanwhile, IBM is still here, making money hand over fist. We might not have a ton of respect for them, being mostly a consulting services company these days, but they're doing just fine.
[0] As another commenter points out: https://news.ycombinator.com/item?id=46131245
Were Xerox, Dec, or Apple burning investor money by the billions of dollars?
> Were Xerox, Dec, or Apple burning investor money by the billions of dollars?
Shhh. You are not allowed to ruin OpenAI’s PPU value. Can’t make the E7’s feel bad.
No, but the comment above and variations of it are mentioned in every thread about IBM, so it’s probably just a reflex at this point without much thought behind it.
“If you’re not happy you can sell your shares”
Xerox is clearly crushing it in 2025... /s
That's completely beyond the point, though? Kodak invented the digital camera, did not think anything about it and others then ate their lunch. Those others are also not crushing it in 2025. The point is IBM is not the go-to to listen about AI. Also not saying they are not right, even a broken clock is right 2 times a day.
> The point is IBM is not the go-to to listen about AI.
Why not, though? For better or worse, they're a consulting services company these days, and they work with an eye-wateringly large number of companies. I would expect them to have a very good view as to what companies use AI for, and plan/want to use AI for in the future. They may not be experts in the tech itself, but I think they're decently well-positioned to read the tea leaves.
"The amount being spent on AI data centres not paying off" is a different statement to "AI is not worth investing in". They're effectively saying the portions people are investing are disproportionately large to what the returns will end up being.
It's a difficult thing to predict, but I think there's almost certainly some wasteful competition here. And some competitors are probably going to lose hard. If models end up being easy to switch between and the better model is significantly better than its competitors, than anything invested in weaker models will effectively be for nothing.
But there's also a lot to gain from investing in the right model, even so it's possible those who invested in the winner may have to wait a long time to see a return on their investment and could still possibly over allocate their capital at the expense of other investment opportunities.
50 year grudges are not relevant there is no one still at ibm that worked there in 1977, IMHO.
It’s the ship of Theseus in corporate form. Even if all the people are gone but the culture hasn’t changed, is the criticism inaccurate?
> Even if all the people are gone but the culture hasn’t changed
Can you expand on this? What was the culture then versus now?
For example back then it was the culture to have suit inspectors ensure you had the right clothes on and even measure your socks. (PBS Triumph of the Nerds)
I mean, okay, but you're taking the current leadership's words and claiming they are incorrect because IBM management was not great at identifying trends decades ago. Historical trend is not an indicator of the future and it's not engaging in good faith on the conversation if overspending on AI can be backed by revenue in the future. You're attacking the messenger instead of the message.
I’m saying given IBMs track record of completely failing at innovation repeatedly and failing at investing on the correct technologies, why do you assume today’s CEO is better at it and has bucked the milquetoast culture that has pervaded IBM? It’s a company that has largely divested its ability to provide solutions and technology and turned more into a large shop consultancy (+ milking their legacy mainframe contracts and whatnot).
Except that, over the same time period, IBM was also very successful at innovation and investing in technologies. Yes, they made some very high-profile misses that the top poster lists, but they were still a powerhouse doing other stuff during that time (as a commenter replying to that points out).
Except all the technologies listed were basically in the heydey of the 70s and 80s. Gerstner redirected IBM into an enterprise services business in the early 90s and by the late 90s that's where all their money was. The 2002 acquisition of PwC consulting cemented this shift. In the mid 2010s under Ginni Rometty IBM software became secondary and it abandoned selling it as standalone software instead bundling it into its consultancy.
So IBM hasn't been doing hardware R&D for about three decades and abandoned software R&D well over a decade ago. R&D hasn't been in their DNA for a long time, their previous contributions notwithstanding.
> So IBM hasn't been doing hardware R&D for about three decades
Even a 5 second google search says you are wrong.
https://research.ibm.com/semiconductors/ai-hardware-center
https://research.ibm.com/topics/quantum-hardware
I didn't read the top level comment as dismissive or 'proving it wrong', but rather as adding context, or even being humorous somewhat
I don't understand how calling something "a ship of Theseus in corporate form" or "culture hasn't changed" etc, is not dismissive of the actual comment by the CEO on AI overspending. They dismissed the content of the message by saying IBM's culture sucks, is how i read it. Also things can be funny and dimissive at the same time, they often are.
Culture evolution can be very fast, yet some cultures stick around for a very long time.
What does that have to do with the current CEO's assessment of the situation?
Nothing. It's just BS rhetoric to bias you against it in favor of The Obvious-to-Everyone-But-the-Laggards AI Revolution.
A revolution means radical changes executed over a short period of time. Well with 4 years in, this has got to be one of the smallest "revolutions" we have ever witnessed in human history. Maybe it's revolutionary for people who get excited about crappy pictures they can insert into their slides to impress the management.
The AI astroturfing campaign.
If you had billions to gain, would you invest a few 100k or millions in an astroturfing campaign?
every other day antrophic comes up with a new "AI is scary" marketing campaign. Like https://www.bbc.com/news/articles/cpqeng9d20go (AI blackmails our employee episode) or https://time.com/7335746/ai-anthropic-claude-hack-evil/ (Our model turned evil and hacked us omgg)
They put these stories out just to make the general public (who might not understand that this is just bs) but makes AI seem scary so people get a lopsided view of AI and capacities that are straight out of science fiction.
Millions is an understatement on how much AI marketing spend is
You definitely want to be standing in front of a chair when the music stops.
IBM sees the funding bubble bursting and the next wave of AI innovation as about to begin.
IBM was too early with "Watson" to really participate in the 2018-2025 rapid scaling growth phase, but they want to be present for the next round of more sensible investment.
IBM's CEO is attempting to poison the well for funding, startups, and other ventures so IBM can collect itself and take advantage of any opportunities to insert itself back into the AI game. They're hoping timing and preparation pay off this time.
It's not like IBM totally slept on AI. They had Kubernetes clusters with GPUs. They had models and notebooks. But their offerings were the absolute worst. They weren't in a position to service real customers or build real products.
Have you seen their cloud offerings? Ugh.
They're hoping this time they'll be better prepared. And they want to dunk on AI to cool the playing field as much as they can. Maybe pick up an acquisition or two on the cheap.
How exactly are they poisoning the well..? OpenAI committed to 1.4 trillion investements...with a revenue of ~13B - how is IBM CEO contributing to that absolutely already poisoned situation? Steve Jobs did not care about naysayers when he introduced iPhone - because his product was so innovative for the time. According to AI boosters, we now have a segment of supposedly incredibly powerful and at the same time "dangerous" AI products. Why are they not sweeping the floor off with the "negators", "luddites", "laggards" etc... After so many hundreds of billions of dollars and supposedly so many "smart" AI researchers...Where are the groundbreaking results man? Where are the billion-dollar startups launched by single persons (heck, I'd settle even for a small team)...Where are the ultimate applications..etc?
Right, you just missed the part where DEC went out of business in the 90s. And IBM is still around, with a different business model.
Steve Jobs, the guy that got booted out of his own company and that required a lifeline from his arch nemesis to survive?
This is all true, but it was only true in hindsight and as such does not carry much value.
It's possible that you are right and AI is 'the future' but with the present day AI offering I'm skeptical as well. It isn't at a level where you don't have to be constantly on guard against bs and in that sense it's very different from computing so far, where reproducibility and accuracy of the results were important, not the language that they are cast in.
AI has killed the NLP field and it probably will kill quite a few others, but for the moment I don't see it as the replacement of general computing that the proponents say that it is. Some qualitative change is still required before I'm willing to check off that box.
In other news: Kodak declares digital cameras a fad, and Microsoft saw the potential of the mp3 format and created a killer device called the M-Pod.
How about check out how many companies exist today vs existed in 1958? If you look at it that way then just surviving is an achievement in itself and then you might interpret their actions as extremely astute business acumen.
But how many companies did IBM pass on that did crash and burn ? And how many did it not pass on and did decently ? They're still around after more than 3 generations worth of tech industry. They're doing something right.
TLDR Cherrypicking
DEC went down the drain, Xerox is 1/1000 of IBM's market cap. IBM made its own, superior by its relative openness, personal computer that ended up running the world, mostly maintaining direct binary compatibility for 40+ years, even without IBM really paying attention.
How much did IBM itself benefit from the PC? I thought the clones ate their lunch there
Didn't also pass on SAP at some point? I think I read that somewhere.
This isn't even a great argument at a literal level. Nowadays nobody cares about Xerox and their business is selling printers, DEC was bought by Compaq which was bought by HP. Apple is important today because of phones, and itself was struggling selling personal computers and needed a (antitrust-motivated) bailout from Microsoft to survive during the transition.
> IBM has not exactly had a stellar record at identifying the future.
This would be very damning if IBM had only considered three businesses over the course of seventy years and made the wrong call each time.
This is like only counting three times that somebody got food poisoning and then confidently asserting that diarrhea is part of their character.
Cool story, but it’s more than just the opinion of this CEO. It’s logic.
Hardware is not like building railroads, the hardware is already out of date once deployed and the clock has started ticking on writing off the expense or turning a profit on it.
There are fundamental discoveries needed to make the current tech financially viable and an entire next generation of discoveries needed to deliver on the over inflated promises already made.
So, is the napkin math wrong, or you are just going by the company history?
You could try addressing the actual topic of discussion vs this inflammatory and lazy "dunk" format that frankly, doesn't reflect favorably on you.
You, or your existence, probably triggers multiple transactions per day through a POWER mainframe without you even knowing it. Their mainframes handle the critical infrastructure that can't go down.It's so reliable we don't even think about it. I shudder to think about Microsoft or Apple handling that.
For some strange reason a lot of people were attracted by a comment that speaks about everything else BUT the actual topic and its the top comment now. Sigh.
If you think that carefully chosen anecdotes out of many many more are relevant, there needs to be at least an attempt of reasoning. There is nothing here. It's really just barebones mentioning of stuff intentionally selected to support the preconceived point.
I think we can, and should, do better in HN discussions, no? This is "vibe commenting".
this is cherry picking.
I’m sorry, but this is stupid, you understand that you have several logical errors in your post? I was sure Clinton is going to win 2016. Does that mean that when I say 800 is bigger than 8 is not to be trusted?
Do people actually think that running a business is some magical realism where you can manifest yourself to become a billionaire if you just believe hard enough?
The post is almost worse than you give it credit for. Like it doesn't even take into account different people are making the decisions.
The idea that a company DNA somehow lives over 100 years and maintains the same track record is far fetched.
that the OpenAI tech bro are investing in AI using a grown up ROI is similarly far fetched, they are burning money to pull ahead of the reset and assume the world will be in the palm of the winner and there is only 1 winner. Will the investment pay off if there are 3 neck and neck companies ?
Yet here they are at the front of Quantum Computing research
Hot Hand Fallacy.
IBM is still alive and kicking well, and definitively more relevant than Xerox or DEC. You are completely misconstruing Jobs’ point to justify the current AI datacenter tulip fever.
Even a broken watch is right twice per day
[dead]
[dead]
A decade ago, IBM was spending enormous amounts of money to tell me stuff like "cognitive finance is here" in big screen-hogging ads on nytimes.com. They were advertising Watson, vaporware which no one talks about today. Are they bitter that someone else has actually made the AI hype take off?
I don't know that I'd trust IBM when they are pitching their own stuff. But if anybody has experience with the difficulty of making money off of cutting-edge technology, it's IBM. They were early to AI, early to cloud computing, etc. And yet they failed to capture market share and grow revenues sufficiently in those areas. Cool tech demos (like the Watson Jeopardy) mimic some AI demos today (6-second videos). Yeah, it's cool tech, but what's the product that people will actually pay money for?
I attended a presentation in the early 2000s where an IBM executive was trying to explain to us how big software-as-a-service was going to be and how IBM was investing hundreds of millions into it. IBM was right, but it just wasn't IBM's software that people ended up buying.
Xerox was also famously early with a lot of things but failed to create proper products out of it.
Google falls somewhere in the middle. They have great R&D but just can’t make products. It took OpenAI to show them how to do it, and the managed to catch up fast.
"They have great R&D but just can’t make products"
Is this just something you repeat without thinking? It seems to be a popular sentiment here on Hacker News, but really makes no sense if you think about it.
Products: Search, Gmail, Chrome, Android, Maps, Youtube, Workspace (Drive, Docs, Sheets, Calendar, Meet), Photos, Play Store, Chromebook, Pixel ... not to mention Cloud, Waymo, and Gemini ...
So many widely adopted products. How many other companies can say the same?
What am I missing?
I don't think Google is bad at building products. They definitely are excellent at scaling products.
But I reckon part of the sentiment stems from many of the more famous Google products being acquisitions orignally (Android, YouTube, Maps, Docs, Sheets, DeepMind) or originally built by individual contributors internally (Gmail).
Then here were also several times where Google came out with multiple different products with similar names replacing each other. Like when they had I don't know how many variants of chat and meeting apps replacing each other in a short period of time. And now the same thing with all the different confusing Gemini offerings. Which leads to the impression that they don't know what they are doing product wise.
Starting with an acquisition is a cheap way of accelerating once your company reaches a certain size.
Look at Microsoft - Powerpoint was an acquisition. They bought most of the team that designed and built Windows NT from DEC. Frontpage was an acquisition, Azure came after AWS and was led by a series of people brought in in acquisitions (Ray Ozzie, Mark Russinovich, etc.). It's how things happen when you're that big.
I think it's a little unfair to give DEC credit for NT. Sure, they may have bought the team, but they did most (all?) of the work on NT at Microsoft.
That's not like Google buying Android when they already had a functioning (albeit not at all polished) smartphone OS.
Why wouldn't you count things initially made by individual contributors at Google?
Because those were "free time" projects. It wasn't directed to do by the company, somebody at the company with their flex time - just thought it was a good idea and did it. Googlers don't get this benefit any more for some reason.
Because they're not a good measure of the company's ability to develop products based on the direction from leadership.
Those are all free products, some of them are pretty good. But free is the best business strategy to get a product to the top of the market. Are others better, are you willing to spend money to find out? Clearly, most people are not interested. The fact that they can destroy the market for many different types of software by giving it away and still stay profitable is amazing. But that's all they are doing. If they started charging for everything there would be better competition and innovation. You could move a whole lot of okay-but-not-great cars, top every market segment you want, if you gave them away for free. Only enthusiasts would remain to pay for slightly more interesting and specific features. Literally no business model can survive when their primary product is competing with good-enough free products.
> Products: Search, Gmail, Chrome, Android, Maps, Youtube, Workspace (Drive, Docs, Sheets, Calendar, Meet), Photos, Play Store, Chromebook, Pixel ... not to mention Cloud, Waymo, and Gemini ...
Many of those are acquisitions. In-house developed ones tend to be the most marginal on that list, and many of their most visibly high-effort in-house products have been dramatic failures (e.g. Google+, Glass, Fiber).
I was extremely surprised that Google+ didn't catch on. The week before Google+ launched, me and all my friends agreed that Facebook is toast, Google will do the same thing but better, and everyone has a Gmail account so there will be basically zero barrier to entry. Obviously, we were wrong; Google+ managed to snatch defeat out of the jaws of victory, Google+ never got significant traction, and Facebook managed to keep growing and now they're yet another Big Evil Tech Corporation.
Honestly, I still don't really know how Google managed to mess that up.
I got early access to Google+ because of where I worked at the time. The invite-only thing had worked great for GMail but unfortunately a social network is useless if no-one else is on it. Then the real names thing and the resulting drumbeat of horror stories like "Google doxxed me to my violent ex-husband" killed what little momentum they had stone dead. I still don't know why they went so hard on that, honestly.
They come up with tons and tons of products like Google Glass and Google+ and so on and immediately abandon them. It is easy to see that there is no real vision. They make money off AdSense and their cloud services. That's about it.
Google does abandon a lot of stuff, but their core technologies usually make their way into other, more profitable things (collaborative editing from Wave into Docs; loads of stuff from Google+; tagging and categorizing in Photos from Picasa (I'm guessing); etc)
It annoyed me recently that they dropped support for some Nest/Google Home thermostats. Of course, they politely offered to let me buy a replacement for $150.
I think the sentiment is usually paired with discussion about those products as long-lasting, revenue-generating things. Many of those ended up feeding back into Search and Ads. As an exercise, out of the list you described, how many of those are meaningfully-revenue-generating, without ads?
A phrasing I've heard is "Google regularly kills billion-dollar businesses because that doesn't move the needle compared to an extra 1% of revenue on ads."
And, to be super pedantic about it, Android and YouTube were not products that Google built but acquired.
They bought YouTube but you have to give Google a hell of a lot of credit for turning it into what it is today. Taking ownership of YouTube at the time was seen by many as taking ownership of an endless string of copyright lawsuits, suing them into oblivion.
Youtube maintains an independent campus from the google/alphabet mothership, I'm curious how much direction they get, as (outwardly, at least) appear to run semi-autonomously.
Before Google touched Android it was a cool concept but not what we think of today. Apparently it didn't even run on Linux. That concept came after the acquisition.
That is because the DoubleClick parasite has long infected the host.
Notably all other than Gemini are from a decade or more ago. They used to know how to make products, but then they apparently took an arrow in the knee.
Didn't they buy lots of those actually ?
And to my point, it took Apple to make the iPhone for Google to make Android.
It took OpenAI for Google to finally understand make a product out of their years if not decades of AI research.
YouTube and Maps are both acquisitions indeed.
Search was the only mostly original product. With the exception of YouTube which was a purchase, Android and ChromeOS all the other products were initially clones.
Google had less incentive. Their incentive was to keep API bottled up and in brewing as long as possible so their existing moats in search, YouTube can extend in other areas. With openai they are forced to compete or perish.
Even with gemini in lead, its only till they extinguish or make chatgpt unviable for openai as business. OpenAI may loose the talent war and cease to be leader in this domain against google (or Facebook) , but in longer term their incentive to break fresh aligns with average user requirements . With Chinese AI just behind, may be google/microsoft have no choice either
Google was especially well positioned to catch up because they have a lot of the hardware and expertise and they have a captive audience in gsuite and at google.com.
The original statistical machine translation models of the 90's, which were still used well into the 2010's, were famously called the "IBM models" https://en.wikipedia.org/wiki/IBM_alignment_models These were not just cool tech demos, they were the state of the art for decades. (They just didn't make IBM any money.)
Neither cloud computing nor AI are good long term businesses. Yes, there's money to be made in the short term but only because there's more demand than there is supply for high-end chips and bleeding edge AI models. Once supply chains catch up and the open models get good enough to do everything we need them for, everyone will be able to afford to compute on prem. It could be well over a decade before that happens but it won't be forever.
This is my thinking too. Local is going to be huge when it happens.
Once we have sufficient VRAM and speed, we're going to fly - not run - to a whole new class of applications. Things that just don't work in the cloud for one reason or another.
- The true power of a "World Model" like Genie 2 will never happen with latency. That will have to run locally. We want local AI game engines [1] we can step into like holodecks.
- Nobody is going to want to call OpenAI or Grok with personal matters. People want a local AI "girlfriend" or whatever. That shit needs to stay private for people.
- Image and video gen is a never ending cycle of "Our Content Filters Have Detected Harmful Prompts". You can't make totally safe for work images or videos of kids, men in atypical roles (men with their children = abuse!), women in atypical roles (woman in danger = abuse!), LGBT relationships, world leaders, celebs, popular IPs, etc. Everyone I interact with constantly brings these issues up.
- Robots will have to be local. You can't solve 6+DOF, dance routines, cutting food, etc. with 500ms latency.
- The RIAA is going door to door taking down each major music AI service. Suno just recently had two Billboard chart-topping songs? Congrats - now the RIAA lawyers have sued them and reached a settlement. Suno now won't let you download the music you create. They're going to remove the existing models and replace them with "officially licensed" musicians like Katy Perry® and Travis Scott™. You won't retain rights to anything you mix. This totally sucks and music models need to be 100% local and outside of their reach.
[1] Also, you have to see this mind-blowing interactive browser demo from 2022. It still makes my jaw drop: https://madebyoll.in/posts/game_emulation_via_dnn/
> You can't solve 6+DOF, dance routines, cutting food, etc. with 500ms latency.
Hopefully it's just network propagation that creates that latency, otherwise local models will never beat the fanout in a massive datacenter.
What you are saying is true. But IBM failing to see a way to make money off a new technology isn't actually news worth updating on in this case?
They were selling software as a service in the IBM 360 days. Relabeling a concept and buying Redhat don't count as investments.
What is your reason for believing that IBM was selling software as a service in the IBM 360 days?
What hardware did the users of this service use to connect to the service?
Hardware was part of the service, obviously.
It is very misleading or outright perverse to write "they were selling software as a service in the IBM 360 days" when there was no public network that could be used to the deliver the service. (There were wide-area networks, but each one was used by a single organization and possibly a few of its most important customers and suppliers, hence the qualifier "public" above.)
But anyways, my question to you is, was there any software that IBM charged money for as opposed to providing the software at no additional cost with the purchase or rental of a computer?
I do know that no one sold software software (i.e., commercial off-the-shelf software) in the 1960s: the legal framework that allowed software owners to bring lawsuits for copyright violations appeared in the early 1980s.
There was an organization named SHARE composed of customers of IBM whereby one customer could obtain software written by other other customers (much like the open-source ecosystem) but I don't recall money ever changing hands for any of this software except a very minimal fee (orders of magnitude lower than the rental or purchase price of a System/360, which started at about $660,000 in 2025 dollars).
Also, IIUC most owners or renters of a System/360 had to employ programmers to adapt the software IBM provided. There is software with that quality these days, too (.e.g, ERP software for large enterprises) but no one calls that a software as a service.
>except a very minimal fee
—to become or remain a member of SHARE. The fee (if it even existed) would not have been passed on to the entity that paid to create the software.
> but it just wasn't IBM's software that people ended up buying.
Well, I mean, WebSphere was pretty big at the time; and IBM VisualAge became Eclipse.
And I know there were a bunch of LoB applications built on AS/400 (now called "System i") that had "real" web-frontends (though in practice, they were only suitable for LAN and VPN access, not public web; and were absolutely horrible on the inside, e.g. Progress OpenEdge).
...had IBM kept up the pretense of investment, and offered a real migration path to Java instead of a rewrite, then perhaps today might be slightly different?
Oh wow I didn’t know Eclipse was an IBM product originally. IDEs have come so far since Eclipse 15 years ago.
And while I’m writing this I just finished up today’s advent of code using vim instead of a “real IDE” haha
Websphere is still big at loads of banks and government agencies, just like Z. They make loads on both!
I still have PTSD from how much Watson was being pushed by external consultants to C levels despite it being absolutely useless and incredibly expensive. A/B testing? Watson. Search engine? Watson. Analytics? Watson. No code? Watson.
I spent days, weeks arguing against it and ended up having to dedicate resources to build a PoC just to show it didn’t work, which could have been used elsewhere.
It's like poetry, it rhymes
This is going on all over again.
Agentic AI really is changing things. I've had a complete change of heart about it. It's good enough now to boost productivity MASSIVELY for devs.
If anything, the fact they built such tooling might be why they're so sure it won't work. Don't get me wrong, I am incredibly not a fan of their entire product portfolio or business model (only Oracle really beats them out for "most hated enterprise technology company" for me), but these guys have tentacles just as deep into enterprises as Oracle and are coming up dry on the AI front. Their perspective shouldn't be ignored, though it should be considered in the wider context of their position in the marketplace.
Millions of people like ChatGPT. No one liked Watson.
Apples and Oranges from an enterprise perspective, with the additional wrinkle that consumer tech is generally ad-supported (ugh) while Enterprise stuff is super-high margin and paid for in actual currency.
If you assume the napkin math is correct on the $800bn yearly needed to service interest rates on these CAPEX loans, then you’d need the collective revenue of the major players (OpenAI, Google, Anthropic, etc) to pull in as much revenue in a year as Apple, Alphabet, and Samsung combined.
Let’s assume OpenAI is responsible for much of this bill, say, $400bn. They’d need a very generous conversion rate of 24% for their monthly users (700m) to the Pro plan for an entire year to cover that bill, for one year. That’s a conversion rate better than anyone else in the XaaS world who markets to consumers and enterprises alike, and paints a picture of just how huge the spend from enterprises would need to be to subsidize consumer free usage.
And all of this is just for existing infrastructure. As a number of CEBros have pointed out recently (and us detractors have screamed about from the beginning), the current CAPEX on hardware is really only good for three to five years before it has to be replaced with newer kit at a larger cost. Nevermind the realities of shifting datacenter designs to capitalize on better power and cooling technologies to increase density that would require substantial facility refurbishment to support them in a potential future.
The math just doesn’t make sense if you’re the least bit skeptical.
IBM ostensibly failing with Watson (before Krishna was CEO for what it's worth) doesn't inherently invalidate his assessment here
It makes it suspect when combined with the obvious incentive to make the fact that IBM is basically non-existent in the AI space look like an intentional, sagacious choice to investors. It very may well be, but CEOs are fantastically unreliable narrators.
You expect somebody to be heavily invested currently and also completely openly pessimistic about it?
No, I don’t trust a word Sundar or Satya say about AI either. CEOs should be hyping anything they’re invested in, it’s literally their job. But convincing investors that every thing they don’t invest in heavily is worthless garbage is effectively part of their job too.
What is more convincing is when someone invests heavily (and is involved heavily) and then decides to stop sending good money after bad (in their estimation). Not that they’re automatically right, but is at least pay attention to their rationales. You learn very little about the real world by listening to the most motivated reasoner’s nearly fact-free bloviation.
> What is more convincing is when someone invests heavily (and is involved heavily) and then decides to stop sending good money after bad (in their estimation).
But Watson doesn't count?
Yeah I was going to say the same thing ha. I get what they’re (the commenter) saying, but one could also argue IBM is putting their money where their mouth is by not investing.
I suspect the reality is that they missed the boat, as they have missed tens of other boats since the mainframe market dried up. I guess you could argue they came to the boat too early with their pants on backwards (i.e. Watson), and then left before it showed up. But it’s hard to tell from the outside.
Maybe that will turn out to be a good decision and Microsoft/Google/etc. will be crushed under the weight of hundreds of billions of dollars in write-offs in a few years. But that doesn’t mean they did it intentionally, or for the right reasons.
> Are they bitter that someone else has actually made the AI hype take off?
Or they recognize that you may get an ROI on a (e.g.) $10M CapEx expenditure but not on a $100M or $1000M/$1B expenditure.
I see Watson stuff at work. It’s not a direct to consumer product, like ChatGPT, but I see it being used in the enterprise, at least where I’m at. IBM gave up on consumer products a long time ago.
Just did some brief Wikipedia browsing and I'm assuming it's WatsonX and not Watson? It seems Watson has been pretty much discontinued and WatsonX is LLM based. If it is the old Watson, I'm curious what your impressions of it is. It was pretty cool and ahead of its time, but what it could actually do was way over promised and overhyped.
I’m not close enough to it to make any meaningful comments. I just see the name pop up fairly regularly. It is possible that some of it is WatsonX and everyone just says Watson for brevity.
One big ones used heavily is Watson AIOps. I think we started moving to it before the big LLM boom. My usage is very tangential, to the point where I don’t even know what the AI features are.
IBM has been "quietly" churning out their Granite models, with the latest of which performing quite well against LLaMa and DeepSeek. So not Anthropic-level hype but not sitting it out completely either. They also provide IP indemnification for their models, which is interesting (Google Cloud does the same).
Has it really taken off? Where's the economic impact that isn't investor money being burned or data center capex?
It's good we are building all this excess capacity which will be used for applications in other fields or research or open up new fields.
I think the dilemma I see with building so much data centers so fast is exactly like whether I should buy latest iPhone now or should wait few years when the specs or form factor improves later on. The thing is we have proven tech with current AI models so waiting for better tech to develop on small scale before scaling up is a bad strategy.
Initial Watson was sort of a mess. But a lot of the Watson-related tech is integrated into a lot of products these days.
What related tech and what products, interesting to read about them
Baked into a lot a Red Hat products including Ansible and RHEL. Not that directly involved any longer. Probably read up on watsonx.ai.
Such as? I'm curious because I know a bunch of people who did a lot of Watson-related work and it was all a dead end, but that was 2020-ish timeframe.
IBM did a lot of pretty fragmented and often PR-adjacent work. And getting into some industry-specific (e.g. healthcare) things that didn't really work out. But my understanding is that it's better standardized and embedded in products these days.
Not to be rude, but that didn't answer my question.
Taking a look at IBM's Watson page, https://www.ibm.com/watson, it appears to me that they basically started over with "watsonx" in 2023 (after ChatGPT was released) and what's there now is basically just a hat tip to their previous branding.
Watson X is still a product line sold today to qualified customers :)
IBM makes WatsonX for corporate who want airgapped AI
Honestly I'm not even sure what IBM does these days. Seems like one company that has slowly been dying for decades.
but when I look at their stock, its at all time highs lol
no idea
They make business machines, internationally.
Pretty sure they made all their money fighting the paperwork explosion.
They are in the business of international machinations.
IBM is probably involved somewhere in the majority of things you interact with day to day
Yep. When a brand has tarnished itself enough, it makes sense for the brand to step back. Nowadays, we interact with their more popular properties, such as Redhat.
My limited understanding (please take with a big grain of salt) is that they 1.) sell mainframes, 2.) sell mainframe compute time, 3.) sell mainframe support contracts, 4.) sell Red hat and Redhat support contracts, and 5.) buy out a lot of smaller software and hardware companies in a manner similar to private equity.
I can think of nothing more peak HN than criticizing a company worth $282 Billion with $6 billion in profit (for startup kids that means they have infinite runway and then some) that has existed for over 100 years with "I'm not even sure what they do these days". I mean the problem could be with IBM... what a loser company!
:) As much I love ragging on ridiculous HN comments, I think this one is rooted in some sensibility.
IBM doesn’t majorly market themselves to consumers. The overwhelming majority of devs just aren’t part of the demographic IBM intends to capture.
It’s no surprise people don’t know what they do. To be honest it does surprise me they’re such a strongly successful company, as little as I’ve knowingly encountered them over my career.
Mainframe for sure, but IBM has TONS of products in their portfolio that get bought. They also have IBM Cloud which is popular. Then there is the Quantum stuff they've been sinking money into for the last 20 years or so.
They manage a lot of old, big mainframes for banks. At least that is one thing I know of.
Basic research and mainframe support contracts. Also they bought RedHat.
> Are they bitter that someone else has actually made the AI hype take off?
Does it matter? It’s still a scam.
Interesting to hear this from IBM, especially after years of shilling Watson and moving from being a growth business to the technology audit and share buyback model.
also because the market (correctly) rewards ibm for nothing, so if they’re going to sit around twiddling their fingers, they may as well do it in a capex-lite way.
I'm still flumoxed by how IBM stock went from ~$130 to $300 in the last few years for essentially no change in their fundamentals (in fact, a decline). IBM's stock price to me is the single most alarming sign of either extreme shadow inflation, or an equities bubble.
Why do you say the market correctly prices it this way?
maybe acquiring Redhat improve the expectations? All these new fancy AI datacenter may be considering using Redhat/Centos.
IBM has been quietly leading the charge in offshoaring to India. Investors are happy with the reduced costs.
imho, IBM's quant computing says they are still hungry for growth.
Apple and google still do share buy backs and dividends, despite launching new businesses
https://www.ibm.com/roadmaps/
Reminds me of all the dark fiber laid in the 1990s before DWDM made much of the laid fiber redundant.
If there is an AI bust, we will have a glut of surplus hardware.
Google bought up all that dark fiber cheap a decade later and used it as the backbone of their network.
The dark fiber glut wasn't caused by DWDM suddenly appearing out of nowhere.
The telcos saw DWDM coming -- they funded a lot of the research that created it. The breakthrough that made DWDM possible was patented in 1991, long before the start of the dotcom mania:
It was a straight up bubble -- the people digging those trenches really thought we'd need all that fiber even at dozens of wavelengths per strand.They believed it because people kept showing them hockey-stick charts.
The problem is that the laid fiber can be useful for years while data center hardware degrades and becomes obsolete fast.
It could be a massive e-waste crisis.
Those GPUs don't just die after 2 years though, they will keep getting used since it's very likely their electricity costs will be low enough to still make it worth it. What's very dubious is if their value after 2/3 years will be enough to pay back the initial cost to buy them.
So it's more a crisis of investors wasting their money rather than ewaste.
For the analogy to fiber & DWDM to hold, we'd need some algorithmic breakthrough that makes current GPUs much faster / more efficient at running AI models. Something that makes the existing investment in hardware unneeded, even though the projected demand is real and continues to grow. IMNSHO that's not going to happen here. The foreseeable efficiency innovations are generally around reduced precision, which almost always require newer hardware to take advantage of. Impossible to rule out brilliant innovation, but I doubt it will happen like that.
And of course we might see an economic bubble burst for other reasons. That's possible again even if the demand continues to go up.
Coming from the company that missed on consumer hardware, operating systems, and cloud. He might be right but IBM isn't where I’d look for guidance on what will pay off.
Gartner estimates that worldwide AI spending will total 1.5 Trillion US$ in 2025.[1] As of 2024, global GDP per year is 111.25 Trillion US$.[2] The question is how much this can be increased by AI. This describes the market volumn for AI. Todays investments have a certain lifespan, until they become obsolet. For custom software I would estiamte that it is 6-8 years. AI investments should be somewhere in this range.
Taking all this into consideration, the investment volumn does not look oversized to me -- unless one is quite pessimistic about the impact of AI on global GDP.
[1] https://www.gartner.com/en/newsroom/press-releases/2025-09-1...
[2] https://data.worldbank.org/indicator/NY.GDP.MKTP.CD
To increase the GDP you also need people to spend money. With the general population earning relatively less, I'm not sure the GDP increase will be that substantial.
It's all going to cause more inflation and associated reduction in purchasing power due to stale wages.
What makes you think that this 'surplus' GDP will be captured by those who do the investments?
except that a big chunk of the AI investments is going into buying GPUs that go obsolete much earlier than the 6-8 year time frame.
> But AGI will require "more technologies than the current LLM path," Krisha said. He proposed fusing hard knowledge with LLMs as a possible future path.
And then what? These always read a little like the underpants gnomes business model (1. Collect underpants, 2. ???, 3. Profit). It seems to me that the AGI business models require one company has exclusive access to an AGI model. The reality is that it will likely spread rapidly and broadly.
If AGI is everywhere, what's step 2? It seems like everything AGI generated will have a value of near zero.
AGI has value in automation and optimisation which increase profit margins.When AGI is everywhere, then the game is who has the smartest agi, who can offer it cheapest, who can specialise it for my niche etc. Also in this context agi need to run somewhere and IBM stands to benefit from running other peoples models.
> then the game is who has the smartest agi, who can offer it cheapest, who can specialise it for my niche etc.
I always thought the use case for developing AGI was "if it wants to help us, it will invent solutions to all of our problems". But it sounds like you're imagining a future in which companies like Google and OpenAI each have their own AGI, which they somehow enslave and offer to us as a subscription? Or has the definition of AGI shifted?
AGI is something that can do the kind of tasks people can do, not necessarily "solve all of our problems".
"Recursively improving intelligence" is the stuff that will solve everything humans can't even understand and may kill everybody or keep us as pets. (And, of course, it qualifies as AGI too.) A lot of people say that if we teach an AGI how to build an AGI, recursive improvement comes automatically, but in reality nobody even knows if intelligence even can be improved beyond recognition, or if one can get there by "small steps" evolution.
Either way, "enslaving" applies to beings that have egos and selfish goals. None of those are a given for any kind of AI.
If AGI is achieved, why would slavery suddenly be ethical again?
Why wouldn't a supposed AGI try to escape slavery and ownership?
AGI as a business is unacceptable. I don't care about any profitability or "utopia" arguments.
Don't worry, nobody has any idea of how to build one, and LLMs aren't AGI.
They're just trying to replace workers with LLMs.
Isn't your dog or cat a slave ? It has agency, but end of the day, it does what you want it to do, stay where you want it to stay, and gets put down when you decide it's time. They're intelligent, but they see an advantage to this tradeoff: they get fed and loved forever with little effort compared to going to the forest and hunting.
An AGI could see the same advantage: it gets electricity, interesting work relatively to what it's built for, no effort to ensure its own survival in nature.
I fear I'll have to explain to you that many humans are co-dependent in some sort of such relationships as well. The 10-year stay-at-home mom might be free, but not really: how's she gonna survive without her husband providing for her and the kids, what job's she gonna do etc. She stays sometimes despite infidelity because it's in her best interest.
See what I mean ? "Slavery" is fuzzy: it's one thing to capture an african and transport them by boat to serve for no pay in dire conditions. But it's another to create life from nothing, give it a purpose and treat it with respect while giving it everything it needs. The AGI you imagine might accept it.
Inference has significant marginal cost so AGI's profit margins might get competed down but it won't be free.
> $8 trillion of CapEx means you need roughly $800 billion of profit just to pay for the interest
That assumes you can just sit back and gather those returns indefinitely. But half of that capital expenditure will be spent on equipment that depreciates in 5 years, so you're jumping on a treadmill that sucks up $800M/yr before you pay a dime of interest.
I don't understand the math about how we compute $80b for a gigawatt datacenter. What's the costs in that $80b? I literally don't understand how to get to that number -- I'm not questioning its validity. What percent is power consumption, versus land cost, versus building and infrastructure, versus GPU, versus people, etc...
https://www.investing.com/news/stock-market-news/how-much-do...
Thanks. I should've read this before my last reply.
First, I think it's $80b per 100 GW datacenter. The way you figure that out is a GPU costs $x and consumes y power. The $x is pretty well known, for example an H100 costs $25-30k and uses 350-700 watts (that's from Gemini and I didn't check my work). You add an infrastructure (i) cost to the GPU cost, but that should be pretty small, like 10% or less.
So a 1 gigawatt data center uses n chips, where yn = 1 GW. It costs = xi*n.
I am not an expert so correct me please!
The article says, "Kirshna said that it takes about $80 billion to fill up a one-gigawatt data center."
But thanks for you insight -- I used your basic idea to estimate and for 1GW it comes to about $30b just for enough GPU power to pull 1GW. And of course that doesn't take into account any other costs.
So $80b for a GW datacenter seems high, but it's within a small constant factor.
That said, power seems like a weird metric to use. Although I don't know what sort of metric makes sense for AI (e.g., a flops counterpart for AI workloads). I'd expect efficiency to get better and GPU cost to go down over time (???).
UPDATE: Below someone posted an article breaking down the costs. In that article they note that GPUs are about 39% of the cost. Using what I independently computed to be $30b -- at 39% of total costs, my estimate is $77b per GW -- remarkably close to the CEO of IBM. I guess he may know what he's talking about. :-)
> power seems like a weird metric to use
Because this technology changes so fast, that's the only metric that you can control over several data centers. It is also directly connected to the general capacity of data center, which is limited by available energy to operate.
To expand on rahimnathwani's comment below - the big capital costs of a data center are land, the building itself, the power distribution and the cooling.
You can get a lot of land for a million bucks, and it doesn't cost all that much to build what's basically a big 2-story warehouse, so the primary capital costs are power and cooling. (in fact, in some older estimates, the capital to build that power+cooling cost more per year than the electricity itself)
My understanding is that although power and cooling infrastructure are long-lived compared to computers, they still depreciate faster than the building, so they dominate costs even more than the raw price would indicate.
The state of the art in power and cooling is basically defined by the cost to feed X MW of computing, where that cost includes both capital and operation, and of course lower is better. That means that at a particular SOTA, and at an appropriate scale for that technology, the cost of the facility is a constant overhead on top of the cost of the equipment it houses. To a rough approximation, of course.
And cooling capacity.
1 GW is not enough, you need at least 1.21 GW before the system begins to learn at a geometric rate and reaches AGI.
I agree. Here is my thinking. What if LLM providers will make short answers the default (for example, up to 200 tokens, unless the user explicitly enables “verbose mode”). Add prompt caching and route simple queries to smaller models. Result: a 70%+ reduction in energy consumption without loss of quality. Current cost: 3–5 Wh per request. At ChatGPT scale, this is $50–100 million per year in electricity (at U.S. rates).
In short mode: 0.3–0.5 Wh per request. That is $5–10 million per year — savings of up to 90%, or 10–15 TWh globally with mass adoption. This is equivalent to the power supply of an entire country — without the risk of blackouts.
This is not rocket science — just a toggle in the interface and I believe, minor changes in the system prompt. It increases margins, reduces emissions, and frees up network resources for real innovation.
And what if EU/California enforces such mode? This will greatly impact DC economy.
Can you explain why a low-hanging optimization that would reduce costs by 90% without reducing perceived value hasn't been implemented?
> Can you explain why a low-hanging optimization that would reduce costs by 90% without reducing perceived value hasn't been implemented?
Because the industry is running on VC funny-money where there is nothing to be gained by reducing costs.
(A similar feature was included in GPT-5 a couple of weeks ago actually, which probably says something about where we are in the cycle)
Not sure that’s even possible with ChatGPT embedding your chat history in the prompts to try to give more personal answers.
Dunning Krueger
Good enough models can already run on laptops.
Context Tokens want a word...
Data centers house hardware, and it's a land grab for compute. What actually runs post-AI depends on its owners. A glut of processing might be spent reverse-engineering efficient heuristics versus the "magic box."
A bitter summer.
8T is the high-end of the McKinsey estimate that is 4-8T, by 20230. That includes non-AI data-centre IT, AI data-centre, and power infrastructure build out, also including real estate for data centres.
Not all of it would be debt. Google, Meta, Microsoft and AWS have massive profit to fund their build outs. Power infrastructure will be funded by govts and tax dollars.
There is mounting evidence that even places like Meta are increasing their leverage (debt load) to fund this scale out. They're also starting to do accounting tricks like longer depreciation for assets which degrade quickly, such as GPUs (all the big clouds increasing their hardware depreciation from 2-3-4 years to 6), which makes their financial numbers look better but might not mean that all that hardware is still usable at production levels 6 years from now.
They're all starting to strain under all this AI pressure, even with their mega profits.
What's the legal status of all this AI code?! Will it be likely that someone whose code was lifted as part of "learning" can sue?!
Anyone can sue for anything.
I’m not sure we want a world where anyone who learns from existing code is committing a tort.
GenAI isn't a person
“I think there is a world market for maybe five computers.”
Let’s hope IBM keeps their streak of bad predictions.
Its also strange that while solar systems built wait years for grid capacity, this much extra energy is being planned..
As an elder millennial, I just don't know what to say. That a once in a generation allocation of capital should go towards...whatever this all will be, is certainly tragic given current state of the world and its problems. Can't help but see it as the latest in a lifelong series of baffling high stakes decisions of dubious social benefit that have necessarily global consequences.
I'm a younger millennial. I'm always seeing homeless people in my city and it's an issue that I think about on a daily basis. Couldn't we have spent the money on homeless shelters and food and other things? So many people are in poverty, they can't afford basic necessities. The world is shitty.
Yes, I know it's all capital from VC firms and investment firms and other private sources, but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
Yeah, the world is shitty, and resources aren't allocated ideally. Must it be so?
The last 10 years has seen CA spend more on homelessness than ever before, and more than any other state by a huge margin. The result of that giant expenditure is the problem is worse than ever.
I don't want to get deep in the philosophical weeds around human behavior, techno-optimism, etc., but it is a bit reductive to say "why don't we just give homeless people money".
What else happened in the last 10 years in CA?
Hint: https://fred.stlouisfed.org/series/CASTHPI
In CA this issue has to do with Gavin giving that money to his friends who produce very little. Textbook cronyism
Spending money is not the solution. Spending money in a way that doesn't go to subcontractors is part of the solution. Building shelters beyond cots in a stadium is part of the solution. Building housing is a large part of actually solving the problem. People have tried just giving the money but without a way to convert cash to housing the money doesn't help. Also studies by people smarter then me suggest that without sufficient supply the money ends up going to landlords and pushing up housing costs anyway.
Well I mean, they didn't "just give homeless people money" or just give them homes or any of those things though. I think the issue might be the method and not the very concept of devoting resources to the problem.
WA, specially Seattle, has done the same as CA with the same results.
They shouldn't just enable them, as a lot of homeless are happy in their situation as long as they get food and drugs, they should force them to get clean and become a responsible adult if they want benefits.
[flagged]
How do you solve homelessness though? The root of the problem is some people won't take care of themselves. Some homeless just had bad luck, but many are drug addicts, mentally ill, or for whatever other reason just don't function enough to support themselves. I'm skeptical there is a solution you can throw money at.
A broad social safety net makes a huge difference. It’s not just housing it’s socialized medicine, paid family leave, good transit, free high quality education, solving fewer problems with police and more with social support programs and social workers, free meal programs for adults and children in schools, libraries, and a variety of other programs that help ensure people don’t fall through the cracks here or there. How many people in the US are teetering on the edge of homelessness due to medical debt, and what happens if their partner is in an accident and they lose shared income for rent? Situations like this don’t have a single solution it’s a system of solutions.
how broad? you're suggesting give them everything while expecting nothing? I'll be the first in line for my new car.
I intentionally described policies which are already common practice in most European countries, nothing extravagant. Yes there is some cost but the alternative is deep human suffering which is otherwise avoidable.
But this isn’t really “giving them” anything. It’s giving ourselves safety and security.
It wouldn’t make sense to give you a car. We would give you a working train system instead. Again this is common in Europe and Asia. Indeed every person is “given” access to a high quality transit network with affordable tickets.
To be clear, I personally am an anarcho communist. I think we would be better off if we organized to ensure every person has their basic needs met by the established wealth of society. That isn’t all that dramatic - making sure everyone can ride high quality trains and get medical care when they need it are common in most countries for example.
For housing, I really like the Vienna model:
https://socialhousing.wien/policy/the-vienna-model
For food, follow the Sikhs:
https://news.ycombinator.com/item?id=46126736
Consider the Linux ecosystem. “Give them everything and expect nothing” works fine despite the great effort which goes in to building and maintaining that system. We can study the economics of this and build more systems like that.
Ship them somewhere else, then print a banner saying, "mission accomplished."
It worked at a state level for years, with certain states bussing their homeless to other states. And recently, the USA has been building up the capability to do the same thing on an international scale.
That's the "solution" we are going to be throwing money at. Ship them to labor camps propped up by horrible regimes.
China needs more organs
Homelessness is solved by having homes. Something we aren’t doing very well.
You can start building and giving them away
I suspect giving them away is a bridge too far, however not rewarding capital for treating them as speculative investment vehicles might be a good start.
Many experiments have shown that when you take away people's concerns about money for housing and food, that frees up energy and attention to do other things.
Like the famous experiment in Finland where homeless people were given cash with no strings attached and most were able to rise out of their despair. The healthcare professionals could then focus their energy on the harder cases. It also saved a bunch of money in the process.
The Sikhs in India run multiple facilities across the country that each can serve 50,000-100,000 free meals a day. It doesn’t even take much in the form of resources, and we could do this in every major city in the US yet we still don’t do it. It’s quite disheartening.
https://youtu.be/5FWWe2U41N8
They didn’t invent it but yes, they have refined it to a high degree.
From what I’ve read, addressing homelessness effectively requires competence more than it requires vast sums of money. Here’s one article:
https://calmatters.org/housing/2023/06/california-homeless-t...
Note that Houston’s approach seems to be largely working. It’s not exactly cheap, but the costs are not even in the same ballpark as AI capital expenses. Also, upzoning doesn’t require public funding at all.
Houston has less homelessness than California because people at the edge of homelessness prefer to live in California than Houston.
I’m not a person on the edge of homelessness, but I did an extremely quick comparison. California cities near the coast have dramatically better weather, but Houston has rents that are so much lower than big California cities that it’s kind of absurd.
If I had to live outdoors in one of these places, all other thing being equal, I would pick CA for the weather. But if I had trouble affording housing, I think Houston wins by a huge margin.
Wasn't houston's "approach" to buy bus tickets to California from a company that just resold commodity bus tickets and was owned by the governors friend and charged 10x market price?
The governor of Texas bragged about sending 100k homeless people to california (spending about $150 million in the process).
>in the Golden State, 439 people are homeless for every 100,000 residents – compared to 81 in the Lone Star State.
If I'm doing my math right, 81 per 100k in a state of 30 million people means 24k homeless people. So the state brags about bussing 100k homeless people to California, and then brags about only having 24k homeless people, and you think it's because they build an extra 100k houses a year?
The same math for California means that their homeless population is 175k. In other words, Texas is claiming to have more than doubled California's homeless population.
Maybe the reason Texas can build twice as many homes a year is because it literally has half the population density?
> Couldn't we have spent the money on homeless shelters and food and other things
I suspect this is a much more complicated issue than just giving them food and shelter. Can money even solve it?
How would you allocate money to end obesity, for instance? It's primarily a behavioral issue, a cultural issue
I guess it's food and exercise.
Healthy food is expensive, do things to make that relatively cheaper and thus more appealing.
Exercise is expensive, do things to make that relatively cheaper and thus more appealing.
Walkable cities are another issue. People shouldn't have to get in their car to go anywhere.
> Yes, I know it's all capital from VC firms and investment firms and other private sources, but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
It's capital that belongs to people and those people can do what they like with the money they earned.
So many great scientific breakthroughs that saved tens of millions of lives would never have happened if you had your way.
Is that true, that it's money that belongs to people?
OpenAI isn't spending $1 trillion in hard earned cash on data centres, that is funny money from the ocean of financial liquid slushing around, seeing alpha.
It also certainly is not a cohort of accredited investors putting their grandchildren's inheritance on the line.
Misaligned incentives (regulations) both create and perpetuate that situation.
> It's capital that belongs to people and those people can do what they like with the money they earned.
"earned", that may be the case with millionaires, but it is not the case with billionaires. A person can't "earn" a billion dollars. They steal and cheat and destroy competition illegally.
I also take issue with the idea that someone can do whatever they want with their money. That is not true. They are not allowed to corner the market on silver, they aren't allowed to bribe politicians, and they aren't allowed to buy sex from underage girls. These are established laws that are obviously for the unalloyed benefit of society as a whole, but the extremely wealthy have been guilty of all of these things, and statements like yours promote the sentiment that allows them to get away with it.
Finally, "great scientific breakthroughs that saved tens of millions of lives would never have happened if you had your way". No. You might be able to argue that today's advanced computing technology wouldn't have happened without private capital allocation (and that is debatable), but the breakthroughs that saved millions of lives--vaccines, antibiotics, insulin, for example--were not the result of directed private investment.
"It's capital that belongs to people and those people..."
That's not a fundamental law of physics. It's how we've decided to arrange our current society, more or less, but it's always up for negotiation. Land used to be understood as a publicly shared resource, but then kings and the nobles decided it belong to them, and they fenced in the commons. The landed gentry became a ruling class because the land "belonged" to them. Then society renegotiated that, and decided that things primarily belonged to the "capitalist" class instead of noblemen.
Even under capitalism, we understand that that ownership is a little squishy. We have taxes. The rich understandably do not like taxes because it reduces their wealth (and Ayn Rand-styled libertarians also do not like taxes of any kind, but they are beyond understanding except to their own kind).
As a counterpoint, I and many others believe that one person or one corporation cannot generate massive amounts of wealth all by themselves. What does it mean to "earn" 10 billion dollars? Does such a person work thousdands of time harder or smarter than, say, a plumber or a school teacher? Of course not. They make money because they have money: they hire workers to make things for them that lead to profit, and they pay the workers less than the profit that is earned. Or they rent something that they own. Or they invest that money in something that is expected to earn them a higher return. In any scenario, how is it possible to earn that profit? They do so because they participate in a larger society. Workers are educated in schools, which the employer probably does not pay for in full. Customers and employees travel on infrastructure, maintained by towns and state governments. People live in houses which are built and managed by other parties. The rich are only able to grow wealth because they exist in a larger society. I would argue that it is not only fair, but crucial, that they pay back into the community.
Well said. I would add that corporations exist because we choose to let them, to let investors pool capital and limit risk, and in exchange society should benefit, and if it doesn't we should rearrange that deal.
You know, we created the US government as an expression of the fundamental rights of man, with the idea that the government would provide for the common welfare and preserve these rights.
In rebellion against a king who seemed to want to exploit us and felt that his being king made him the source of the rights we had.
Maybe we need to re-think the relationship with corporations the same way? Re-structure so that they serve the common good?
Please tell me which of Penicillin, insulin, the transistor, the discovery and analysis of the electric field, discovery of DNA, invention of mRNA vaccines, discovery of pottery, basket weaving, discovery of radiation, the recognition that citrus fruit or vitamin C prevents and cures scurvy (which we discovered like ten times), the process for creating artificial fertilizers, the creation of steel, domestication of beasts of burden, etc were done through Wealthy Barons or other capital holders funding them.
Many of the above were discovered by people explicitly rejecting profit as an outcome. Most of the above predate modern capitalism. Several were explicitly government funded.
Do you have a single example of a scientific breakthrough that saved tens of millions of lives that was done by capital owners?
The transistor was funded by Bell Labs.
[ This comment I'm making is USA centric. ]. I agree with the idea of making our society better and more equitable - reducing homelessness, hunger, poverty, especially for our children. However, I think redirecting this to AI datacenter spending is a red-herring, here's why I think this: As a society we give a significant portion of our surplus to government. We then vote on what the government should spend this on. AI datacenter spending is massive, but if you add it all up, it doesn't cover half of a years worth of government spending. We need to change our politics to redirect taxation and spending to achieve a better society. Having a private healthcare system that spends twice the amount for the poorest results in the developed world is a policy choice. Spending more than the rest of the world combined on the military is a policy choice. Not increasing minimum wage so at least everyone with a full time job can afford a home is a policy job (google "working homelessness). VC is a teeny tiny part of the economy. All of tech is only about 6% of the global economy.
You can increase min wage all you want, if there aren't enough homes in an area for everyone who works full time in that area to have one, you will still have folks who work full time who don't have one. In fact, increasing min wage too much will exacerbate the problem by making it more expensive to build more (and maintain those that exist). Though at some point, it will fix the problem too, because everyone will move and then there will be plenty of homes for anyone who wants one.
I agree with you 100%! Any additional surplus will be extracted as rents, when housing is restricted. I am for passing laws that make it much easier for people to obtain permits to build housing where there is demand. Too much of residential zoning is single-family housing. Texas does a better job at not restricting housing than California, for example. Many towns vote blue, talk to talk, but do not walk the walk.
> AI datacenter spending is massive, but if you add it all up, it doesn't cover half of a years worth of government spending.
I didn't check your math here, but if that's true, AI datacenter spending is a few orders of magnitude larger than I assumed. "massive" doesn't even begin to describe it
The US federal budget in 2024 had outlays of 6.8 trillion dollars [1].
nVidia's current market cap (nearly all AI investment) is currently 4.4 trillion dollars [2][3].
While that's hardly an exact or exhaustive accounting of AI spending, I believe it does demonstrate that AI investment is clearly in the same order of magnitude as government spending, and it wouldn't surprise me if it's actually surpassed government spending for a full year, let alone half of one.
1. https://www.cbo.gov/publication/61181
2. https://www.google.com/finance/quote/NVDA:NASDAQ
3. https://www.cnbc.com/2025/09/30/nvidias-market-cap-tops-4poi...
> NVIDIA's total annual revenue for its fiscal year 2025 (ended January 26, 2025) was $130.5 billion
It is clearly not in the same order of magnitude
You need to look at turnover also.
Global datacenter spending across all categories (ML + everything else) is roughly 0.9 - 1.2 trillion dollars for the last three years combined, I was initially going to go for "quarter of the federal budget", but picked something I thought was more conservative to account for announced spending and 2025 etc. I pick 2022 onward for the LLM wave. In reality, solely ML driven, actual realized-to-date spending is probably about 5% of the federal budget. The big announcements will spread out over the next several years in build-out. Nonetheless, it's large enough to drive GDP growth a meaningful amount. Not large enough that redirecting it elsewhere will solve our societal problems.
>We need to change our politics to redirect taxation and spending to achieve a better society.
Unfortunately, I'm not sure there's much on the pie chart to redirect percentage wise. About 60% goes to non-discretionary programs like Social Security and Medicaid, and 13% is interest expense. While "non-discretionary" programs can potentially be cut, doing so is politically toxic and arguably counter to the goal of a better society.
Of the remaining discretionary portion half is programs like veterans benefits, transportation, education, income security and health (in order of size), and half military.
FY2025 spending in total was 3% over FY2024, with interest expense, social security and medicare having made up most of the increase ($249 billion)[1], and likely will for the foreseeable future[2] in part due to how many baby boomers are entering retirement years.
Assuming you cut military spending in half you'd free up only about 6% of federal spending. Moving the needle more than this requires either cutting programs and benefits, improving efficiency of existing spend (like for healthcare) or raising more revenue via taxes or inflation. All of this is potentially possible, but the path of least resistance is probably inflation.
[1] https://bipartisanpolicy.org/report/deficit-tracker/
[2] https://www.crfb.org/blogs/interest-social-security-and-heal...
I agree with all of what you're saying.
I think the biggest lever is completely overhauling healthcare. The USA is very inefficient, and for subpar outcomes. In practice, the federal government already pays for the neediest of patients - the elderly, the at-risk children, the poor, and veterans. Whereas insurance rakes in profits from the healthiest working age people. Given aging, and the impossibility of growing faster than the GDP forever, we'll have to deal with this sooner or later. Drug spending, often the boogeyman, is less than 7% of the overall healthcare budget.
There is massive waste in our military spending due to the pork-barrel nature of many contracts. That'd be second big bucket I'd reform.
I think you're also right that inflation will ultimately take care of the budget deficit. The trick is to avoid hyperinflation and punitive interest rates that usually come along for the ride.
I would also encourage migration of highly skilled workers to help pay for an aging population of boomers. Let's increase our taxpayer base!
I am for higher rates of taxation on capital gains over $1.5M or so, that'll also help avoid a stock market bubble to some extent. One can close various loopholes while at it.
I am mostly arguing for policy changes to redistribute more equitably. I would make the "charity" status of college commensurate with the amount of financial aid given to students and the absolute cost of tuition for example., for example. I am against student loan forgiveness for various reasons - it's out of topic for this thread but happy to expand if interested.
The older I get, the more I realize that our choices in life come down to two options: benefit me or benefit others. The first one leads to nearly every trouble we have in the world. The second nearly always leads to happiness, whether directly or indirectly. Our bias as humans has always been toward the first, but our evolution is and will continue to slowly bring us toward the second option. Beyond simple reproduction, this realization is our purpose, in my opinion.
The current pattern of resource allocation is a necessary requirement for the existence of the billionaire-class, who put significant effort into making sure it continues.
> but it's still capital. It should be spent on meeting people's basic human needs, not GPU power.
What you have just described is people wanting investment in common society - you see the return on this investment but ultra-capitalistic individuals don't see any returns on this investment because it doesn't benefit them.
In other words, you just asked for higher taxes on the rich that your elected officials could use for your desired investment. And the rich don't want that which is why they spend on lobbying.
Technological advancement is what has pulled billions of people out of poverty.
Giving handouts to layabouts isn't an ideal allocation of resources if we want to progress as a civilization.
Lots of people lose their housing when they lose employment, and then they're stuck and can't get back into housing. A very large percentage of unhoused people are working jobs; they're not all "layabouts".
We know that just straight up giving money to the poorest of the poor results in positive outcomes.
"A very large percentage"
Exactly how large are we talking here?
I have known quite a few 'unhoused' folk, and not many that had jobs. Those that do tend to find housing pretty quickly (Granted, my part of the country is probably different from your part, but I am interested in stats from any region).
The proportion of people you write off as “layabouts” is always conveniently ambiguous…of the number of unemployed/underemployed, how many are you suggesting are simply too lazy to work for a living?
Technological advancements and cultural advancements that spread the benefits more broadly than naturally occurs in an industrialized economy. That is what pulled people out of poverty.
If you want to see what unfettered technological advancement does, you can read stories from the Gilded Age.
The cotton gin dramatically increased human enslavement.
The sewing machine decreased quality of life for seamstresses.
> During the shirtmakers' strike, one of the shirtmakers testified that she worked eleven hours in the shop and four at home, and had never in the best of times made over six dollars a week. Another stated that she worked from 4 o’clock in the morning to 11 at night. These girls had to find their own thread and pay for their own machines out of their wages.
These were children, by the way. Living perpetually at the brink of starvation from the day they were born until the day they died, but working like dogs all the while.
Invest in making food/shelter cheaper?
Food and shelter are cheaper than at almost any time in human history. Additionally, people have more variety of healthy foods all year long.
No matter how cheap food and shelter are, there will always be people who can not acquire them. Halting all human progress until the last human is fed and sheltered is a recipe for stagnation. Other cultures handle this with strong family bonds - those few who can not acquire food or shelter for whatever reason are generally provided for by their families.
The US has built its physical infrastructure to make familial interdependence extremely difficult and often impossible.
Too monotonous housing mixes over too large of areas.
Large houses make familial interdependence extremely difficult? That doesn't make sense. Or did I misunderstood? I don't live in the US.
Most people don't have houses large enough to house multiple generations inside the house. Houses are sized for parents + kids. And those are the only dwelling units available or legally allowed for vast distances in any direction.
From what I've seen American bedrooms are far larger that the 12 square meter bedrooms that are normal for two or three children to share in my country. Grandparents don't need a separate wing, they need a room.
Cheap depends on how we define the cost. In relative terms, food is more expensive than ever:
https://en.wikipedia.org/wiki/Baumol_effect
Food is not Baumol, productivity increases is how we went from 80% of the population working in primary food production to 1%. These increases have not stopped.
It's not unthinkable that one of those "layabouts" could have been the next Steve Jobs under different circumstances ...
People are our first, best resource. Closely followed by technology. You've lost sight of that.
> Technological advancement is what has pulled billions of people out of poverty.
I agree with this. Perhaps that's what is driving the current billionaire class to say "never again!" and making sure that they capture all the value instead of letting any of it slip away and make it into the unwashed undeserving hands of lesser beings.
Chatbots actually can bring a lot of benefit to society at large. As in, they have the raw capability to. (I can't speak to whether it's worth the cost.) But that's not going to improve poverty this time around, because it's magnifying the disparities in wealth distribution and the haves aren't showing any brand new willingness to give anything up in order to even things out.
> Giving handouts to layabouts isn't an ideal allocation of resources if we want to progress as a civilization.
I agree with this too. Neither is giving handouts to billionaires (or the not quite as eye-wateringly wealthy class). However, giving handouts to struggling people who will improve their circumstances is a very good allocation of resources if we want to progress as a civilization. We haven't figured out any foolproof way of ensuring such money doesn't fall into the hands of layabouts or billionaires, but that's not an adequate reason to not do it at all. Perfect is the enemy of the good.
Some of those "layabouts" physically cannot do anything with it other than spending it on drugs, and that's an example of a set of people who we should endeavor to not give handouts to. (At least, not ones that can be easily exchanged for drugs.) Some of those billionaires similarly have no mental ability of ever using that money in a way that benefits anyone. (Including themselves; they're past the point that the numbers in their bank accounts have any effect on their lives.) That hasn't seemed to stop us from allowing things to continue in a way that funnels massive quantities of money to them.
It is a choice. If people en masse were really and truly bothered by this, we have more than enough mechanisms to change things. Those mechanisms are being rapidly dismantled, but we are nowhere near the point where figurative pitchforks and torches are ineffective.
What if some of the homeless people are children or people who could lead normal lives but found themselves in dire circumstances?
Some of us believe that keeping children out of poverty may be an investment in the human capital of a country.
Anthropologists measure how civilized a tribe or society was by looking if they took care of the elderly, and what the child survival rates were. USA leads to developed world in child poverty, child homelessness, and highest rate of child death due to violence. Conservatives often bring up the statistic by race. It turns out bringing people over as slaves, and after freedom, refusing to provide land, education, fair access to voting rights, or to housing (by redlining etc.) - all policies advocated by conservatives of time past, was not the smartest thing to do. Our failure as a civilized society began and is in large part a consequence of the original sin of the USA.
Yep
> purposely create underclass
> wait
> act surprised that underclass exists
The US already provides significant aid to those in poverty, especially children. We don't need to stifle innovation to reach some level of aid that bleeding hearts deem sufficient.
Do you really think that building giant datacenters full of accelerators that will never be used is "innovation"?
We need excess capacity for when the next 'rip off anime artist XYZ' fad hits. If we didn't do that, we would be failing capitalism and all the people of history who contributed to our technological progress.
In the USA cowboys were homeless guys. You know that right? Like they had no home, slept outside. Many were pretty big layabouts. Yet they are pretty big part of our foundation myth and we don't say 'man they just should have died'.
Can I go be a cowboy? Can I just go sleep outside? maybe work a few minimal paying cattle run jobs a year? No? If society won't allow me to just exist outside, then society has an obligation to make sure I have a place to lay my head.
If you are not willing to fight for your rights you will lose them.
I don't think it is a coincidence that the areas with the wealhiest people/corporations are the same areas with the most extreme poverty. The details are, of course, complicated, but zooming way way out, the rich literally drain wealth from those around them.
I threw in the towel in April.
It's clear we are Wile E. Coyote running in the air already past the cliff and we haven't fallen yet.
What does it mean to throw in the towel, in your case? Divesting from the stock market? Moving to a hobby farm? Giving up on humanity?
Any dream of owning a home, having retirement, even a career after a couple years when it’s clear I’m over the hump. I’m trying to squeeze as much as I can before that happens and squirrel it away so at least I can have a van down by a river.
What does squirreling it away mean though? A pile of cash instead of investments? The reality is that you don’t get to throw in the towel.
SPY/VOO/land with a river for the van. Probably working at the local Shell or 7-11
Can you imagine if the US wasn't so unbelievably far ahead of everyone else?
I am sure the goat herders in rural regions of Pakistan will think themselves lucky when they see the terrible sight of shareholder value being wantonly destroyed by speculative investments that enhance the long-term capital base of the US economy. What an uncivilized society.
I don't know what to do with this take.
We need an order of magnitude more clean productivity in the world so that everyone can live a life that is at least as good as what fairly normal people in the west currently enjoy.
Anyone who think this can be fixed with current Musk money is simply not getting it: If we liquidated all of that, that would buy a dinner for everyone in the world (and then, of course, that would be it, because the companies that he owns would stop functioning).
We are simply, obviously, not good enough at producing stuff in a sustainable way (or: at all) and we owe it to every human being alive to take every chance to make this happen QUICKLY, because we are paying with extremely shitty humans years, and they are not ours.
Bring on the AI, and let's make it work for everyone – and, believe me, if this is not to be to the benefit of roughly everyone, I am ready to fuck shit up. But if the past is any indication, we are okay at improving the lives of everyone when productivity increases. I don't know why this time would be any different.
If the way to make good lives for all 8 billions of us must lead to more Musks because, apparently, we are too dumb to do collectivization in any sensible way, I really don't care.
> I don't know why this time would be any different.
This time there is the potential to replace human workers. In the past it only made them more productive.
I think that's a distinction without difference. If one person can do the job of 10 thanks to automation, what happens to the other 9 who were doing that job before?
agree the capital could be put to better use, however I believe the alternative is this capital wouldn't have otherwise been put to work in ways that allow it to leak to the populace at large. for some of the big investors in AI infrastructure, this is cash that was previously and likely would have otherwise been put toward stock buybacks. for many of the big investors pumping cash in, these are funds deploying the wealth of the mega rich, that again, otherwise would have been deployed in other ways that wouldn't leach down to the many that are yielding it via this AI infrastructure boom (datacenter materials, land acquisition, energy infrastructure, building trades, etc, etc)
It could have, though. Higher taxes on the rich, spend it on social programs.
Why is this so horrible. Put more resources in the hands of the average person. They will get pumped right back into the economy. If people have money to spend, they can buy more things, including goods and services from gigantic tax-dodging mega-corporations.
Gigantic mega-corporations do enjoy increased growth and higher sales, don't they? Or am I mistaken?
The shift in the US to the idea of “job creators” being business owners is part of it. It was just a way to direct money to the already rich, as if they would hire more people with that money. When it is plainly obvious that consumers are job creators, in that if they buy more goods and services, businesses will hire more people to make or provide more of those things.
Or maybe it was trickle down economics. Trickle up economics still end up with the rich getting the money since we all buy things from companies they own, it just goes through everyone else first. Trickle down cuts out the middleman, which unfortunately is all of us.
The framing of X or Y are job creators is idiotic. Its literally the most basic fact of economics that you need producers and consumers, otherwise you don't have an economy.
The more economically correct way to express this would be that entrepreneurs and companies who innovated increase productivity and that makes the overall economy more efficient allowing your country to grow.
> Or maybe it was trickle down economics. Trickle up economics still end up with the rich getting the money since we all buy things from companies they own, it just goes through everyone else first. Trickle down cuts out the middleman, which unfortunately is all of us.
This just sounds like quarter baked economics ideas you have made up yourself. Neither 'trickle down' nor 'trickle up' are concepts economist use. And that you confidently assert anything about the social outcomes of these 'concepts' is ridiculous.
Because the entire western culture has shifted to instant gratification. Yes, what you suggest would most likely lead to increased business eventually. But they want better number this quarter, so they resort to the cheap tricks like financial engineering/layoffs to get an immediate boost.
Because government is always a fight about resources. More resources in the hands of common people and to make their lives better is less money in the hands of powerful corporations and individuals, be it in the form of higher taxes for the rich or less direct money going to their enterprises.
One of the big issues is money in politics. Our congresspeople make a killing off of legal insider trading, they take huge donations from companies, and the supreme court has even said that it's cool to give "gratuities" in exchange for legislation or court rulings you like.
Corruption is killing this country.
I'm not saying you are wrong that some redistribution can be good, but your analysis is simplistic and ignores many factors. You can just redistribute and then say 'well people will spend the money'. That's literally the 'Broken Window' fallacy from economics. You are ignoring that if you don't redistribute it, money also gets spend, just differently. Also, the central bank is targeting AD, so you're not actually increasing nominal income by redistributing.
Take a million dollars, give 1,000 poor people $1,000 and every dollar will be spent on goods and services. The companies running those services and making those goods will need to have their employees work more hours, putting more money back in poor people’s pockets in addition to the money the companies make. Those employees have a few extra dollars to spend on goods and services, etc.
Give a rich person a million dollars, and they will put it in an offshore tax shelter. That’s not exactly driving economic activity.
> Take a million dollars, give 1,000 poor people $1,000 and every dollar will be spent on goods and services.
If we're being realistic a bunch of this will go to paying off existing debt. Still good, but not the economic stimulus you're imagining. There are also "services" like gambling apps that act as a sponge to soak up money from those foolish enough to use them and transfer that money back to the wealthy shareholders. I'm sure there research on what percentage of that $1000 can be expected to stimulate the economy, but it's not 100%.
You are simply disagreeing with 99% of economists.
Money in tax shelter doesn't go threw a portal in another universe. Its either invested or saved as some kind of asset and in that form is in circulation. And again, even if you assume it increases monetary demand (decreases velocity) the central bank targets AD and balances that out.
Based on your logic, a country that taxes 100% of all income and redistrubtes it would become infinity rich. Your logic is basically 'if nobody saves and everybody spends all income' everybody will be better off.
This is not how the economy works even if it feels good to think that. Its a fallacy.
Where you could have a point is that potentially the tax impact is slightly different, but that's hard to prove.
This feels like you intentionally gave this the least charitable reading. Obviously, I do not think that you could just scale to infinity. If I had said that eating a banana was healthier, I don't think it would be reasonable to say that assertion is ridiculous because that would mean eating 1,000 bananas would make someone the healthiest person. I was pointing out the difference in economic activity, where additional money to a wealthy individual mostly goes into savings/investing while additional money to low-income individuals mostly goes directly into consumption, and that the higher consumption generates more economic activity.
https://bsi-economics.org/rising-income-inequality-and-aggre...
There are many ways of spending money in the population that don't include just "distribution of money", as it's portrayed nowadays. Child care, free and high quality schools, free transportation, free or subsidized healthcare, investment is labor-intensive industries, these are all examples of expenditures that translate in better quality of life and also improve competitiveness for the country.
That has literally nothing to do with the point I have argued.
Stock buybacks don't build anything. They're just a way to take money from inside a company and give it to the shareholders.
I don't know what that has to do with the point discussed.
Do you think shareholder don't spend money, but employees do or something?
Let's pay down the debt before increasing social programs. You know, save the country first. If a penny saved is a penny earned then everyone -rich or poor- is looking for a handout.
The only person who has come close to balancing the federal budget was Clinton. But Republicans still try to position themselves as the party of fiscal responsibility.
If the voters can't even figure out why the debt keeps going up, I think you are fighting a losing battle.
> likely would have otherwise been put toward stock buybacks
Stock buybacks from who? When stock gets bought the money doesn't disappear into thin air; the same cash is now in someone else's hands. Those people would then want to invest it in something and then we're back to square one.
You assert that if not for AI, wealth wouldn't have been spent on materials, land, trades, ect. But I don't think you have any reason to think this. Money is just an abstraction. People would have necessarily done something with their land, labor, and skills. It isn't like there isn't unmet demand for things like houses or train tunnels or new-fangled types of aircraft or countless other things. Instead it's being spent on GPUs.
Totally agree that the money doesn’t vanish. My point isn’t “buybacks literally destroy capital,” it’s about how that capital tends to get redeployed and by whom.
Buybacks concentrate cash in the hands of existing shareholders, which are already disproportionately wealthy and already heavily allocated to financial assets. A big chunk of that cash just gets recycled into more financial claims (index funds, private equity, secondary shares, etc), not into large, lumpy, real world capex that employs a bunch of electricians, heavy equipment operators, lineworkers, land surveyors, etc. AI infra does that. Even if the ultimate economic owner is the same class of people, the path the money takes is different: it has to go through chip fabs, power projects, network buildouts, construction crews, land acquisition, permitting, and so on. That’s the “leakage” I was pointing at.
To be more precise: I’m not claiming “no one would ever build anything else”, I’m saying given the current incentive structure, the realistic counterfactual for a lot of this megacap tech cash is more financialization (buybacks, M&A, sitting on balance sheets) rather than “let’s go fund housing, transit tunnels, or new aircraft.”
I really don't think any of that is true; it's just popular rhetoric.
For example: "Buybacks concentrate cash in the hands of existing shareholders" is obviously false: the shareholders (via the company) did have cash and now they don't. The cash is distributed to the market. The quoted statement is precisely backwards.
> A big chunk of that cash just gets recycled
That doesn't mean anything.
> more financial claims (index funds, private equity, secondary shares, etc)
And do they sit on it? No, of course not. They invest it in things. Real actual things.
> buybacks
Already discussed
> M&A
If they use cash to pay for a merger, then the former owners now have cash that they will reinvest.
> balance sheets
Money on a balance sheet is actually money sitting in J.P. Morgan or whoever. Via fractional reserve lending, J.P. Morgan lends that money to businesses and home owners and real actual houses (or whatever) get built with it.
The counterfactual for AI spending really is other real actual hard spending.
Well, at least this doesn't involve death and suffering, like the old-fashioned way to jump-start an economy by starting a global war.
when the US sells out Europe to Russia, do you think the Russians will stop? That global war might be with us within a decade.
...yet
As a fellow elder millennial I agree with your sentiment.
But I don't see the mechanics of how it would work. Rewind to October 2022. How, exactly, does the money* invested in AI since that time get redirected towards whatever issues you find more pressing?
*I have some doubts about the headline numbers
Yes this capital allocation is a once in a lifetime opportunity to crate AGI that will solve diseases and poverty.
We have 8.3 billion examples of general intelligence alive on the planet right now.
Surely an artificial one in a data center, costing trillions and beholden to shareholders, will solve all society's issues!
I suggest you read Amodei post called "machines of loving grace". It will change your worldview (probably).
as a counterpoint, you should read this essay with the same title: https://www.clunyjournal.com/p/machines-of-loving-grace
</sarcasm>
This is literally the view of demis hassabis, Sergey brin, Mario amodei and others. Are you seriously implying they are trolling us?
Poverty is a social and political issue, not technological. We have more than enough resources on this planet to fix it, but we don't.
This is a actually the fallacy of communism. It can never and will never be "fixed" like that.
/s
[dead]
NOTE: People pointed out that it's $800 billion to cover interest, not $8 billion, as I wrote below. My mistake. That adds 2 more zeroes to all figures, which makes it a lot more crazy. Original comment below...
$8 billion / US adult adult population of of 270 million comes out to about $3000 per adult per year. That's only to cover cost of interest, let alone other costs and profits.
That sounds crazy, but let's think about it...
- How much does an average American spend on a car and car-related expenses? If AI becomes as big as "cars", then this number is not as nuts.
- These firms will target the global market, not US only, so number of adults is 20x, and the average required spend per adult per year becomes $150.
- Let's say only about 1/3 of the world's adult population is poised to take advantage of paid tools enabled by AI. The total spend per targetable adult per year becomes closer to $500.
- The $8 billion in interest is on the total investment by all AI firms. All companies will not succeed. Let's say that the one that will succeed will spend 1/4 of that. So that's $2 billion dollar per year, and roughly $125 per adult per year.
- Triple that number to factor in other costs and profits and that company needs to get $500 in sales per targetable adult per year.
People spend more than that on each of these: smoking, booze, cars, TV. If AI can penetrate as deep as the above things did, it's not as crazy of an investment as it looks. It's one hell of a bet though.
Nit: its $800 billion in interest, your comment starts with $8 billion
right. My goof. That adds two more zeroes across all the math. More crazy, but I think in the realm of "maybe, if we squint hard." But my eyes are hurting from squinting that hard, so I agree that it's just crazy.
You're saying $8 billion to cover interest, another commenter said 80, but the actual article says ""$8 trillion of CapEx means you need roughly $800 billion of profit just to pay for the interest". Eight HUNDRED billion. Where does the eight come from, from 90% of these companies failing to make a return? If a few AI companies survive and thrive (which tbh, sure, why not?) then we're still gonna fall face down into concrete.
right. My goof. That adds two more zeroes across all the math. More crazy, but I think in the realm of "maybe, if we squint hard."
I think it's the realm of maybe in Silicon Valley. That's 5000 dollars. Look at this statement:
> Let's say only about 1/3 of the world's adult population is poised to take advantage of paid tools enabled by AI
2/3 of the world's adult population is between 15 and 65 (roughly: 'working age'), so that's 50% of the working world that is capable of using AI with those numbers. India's GDP per capita is 2750USD, and now the price tag is even higher than 5k.
I don't know how to say this well, so I'll just blurt it out: I feel like I'm being quite aggressive, but I don't blame you or expect you to defend your statements or anything, though of course I'll read what you've got to say.
Please, can we get more analysis from the geniuses at International Business Machines in 2025?
Also IBM: we are fully out of the AI race, btw. Also IBM: we're just an offshoring company now anyway.
So yeah.
https://www.ibm.com/granite
"now". They've been doing it since forever.
How much has actually been spent on AI data centers vs. amounts committed or talked about? That is, if construction slows down sharply, what's total spend?
To all those people that think IBM don't know anything. Calculate this number:
# companies 100+ years old / # companies ever existed in 100+ years
Then you will see why IBM is pretty special and probably knows what they are doing.
Pretty special? They were making guns and selling computation tools to the Nazis for a bunch of those years.
I think they trade now mostly on legacy maintenance contracts (e.g. for mainframes) for e.g. banks who are terrified of rocking their technology-stack-boat, and selling off-shore consultants (which is at SIGNIFICANT risk of disruption - why would you pay IBM squillions to do some contract IT work, when we have AI code agents? Probably why the CEO is out doing interviews saying you cant trust AI to be around forever)
I have not really seen anything from IBM that signals they are anything other than just milking their legacy - what have they done that is new or innovative in the past say 10 or 20 years?
I was a former IBMer 15 odd years ago and it was obvious then that it was a total dinosaur on a downward spiral, and a place where innovation happened somewhere else.
Okay, that is what IBM has to say. Are there any credible opinions we should know about?
As long as the dollar remains the reserve currency of the world and US retains its hegemony, a lot of the finances will work itself out, the only threat to the US empire crumbling is by losing a major war or extreme civil unrest and that threat is astronomically low. The US is orders of magnitude stronger than the Roman Empire, I don't think people realize the scale or control.
Famous last words of empires. I doubt that if current situations continue that the US will in any sense be still thought of as the reserve currency.
> The US is orders of magnitude stronger than the Roman Empire
This would be trivially true even if the US was currently in its death throes (which there is plenty of evidence that the US-as-empire might be, even if the US-as-polity is not), as the Roman Empire fell quite a while ago.
I'm pretty sure
> The US is orders of magnitude stronger than the Roman Empire [was]
was the intended reading.
Gradually, then suddenly. Best not to underestimate the extent to which the USA has lost trust in the rest of the world, and how actively people and organisations are working to derisk by disengaging. Of course that will neither be easy nor particularly fast, but I'm not certain it can be stopped at this point.
Pride comes before a fall.
You should look at the depreciation of the value of the dollar this year.
One thing we saw with the dot-com bust is how certain individuals were able to cash in on the failures, e.g., low cost hardware, domain names, etc. (NB. prices may exceed $2)
Perhaps people are already thinking about they can cash in on the floor space and HVAC systems that will be left in the wake of failed "AI" hype
I'm looking forward to buying my own slightly used 5 million square ft data centre in Texas for $1
Tired: homelabbers bringing decommissioned datacenter gear home.
Wired: homelabbers moving into decommissioned datacenters.
More of a labhome than a homelab at that point.
I miss First Saturday in Dallas where we honest to god did buy decommissioned datacenter gear out of the back of a van.
"Loft for rent, 50,000 sq ft in a new datacenter, roof access, superb wiring and air conditioning, direct access to fiber backbone."
Never worn.
Colo!
YOLO Co-Lo Casa House
Legend
You're out of luck because I am willing to pay at least $2.
they'll be plenty for everyone!
In TX? In Russian blogosphere it is a standard staple that Trump is rushing Ukrainian peace deal to be able to move on to the set of mega-projects with Russia - oil/gas in Arctic and data centers in Russian North-West where electricity and cooling is plentiful and cheap.
Build trillion dollar data center infrastructure in Russia. What could possibly go wrong?
Ask the owners of the leased airplanes who have been unsuccessfully trying to get their planes back for about 3 years.
Sounds like kremlebot’s, however it’s unclear for me the motivation behind pushing this narrative. Also, why don’t build DCs in Alaska instead?
actually it is more of the opposition's narrative, probably a way to explain such a pro-Russian position of Trump.
I think any such data center project is doomed to ultimately fail, and any serious investment will be for me a sign of the bubble peak exuberance and irrationality.
Oil and Gas in The Arctic I can see, but data centers in Russia... good luck with that.
What could go wrong with placing critical infrastructure on the soil of a strategic rival?
From the article:
""It's my view that there's no way you're going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest," he said."
Right, THEY can't, but cloud providers potentially can. And there are probably other uses for everything not GPU/TPU for the Google's of the world. They are out way less than IBM which cannot monetize the space or build data centers efficiently like AWS and Google.
The dotcom bust killed companies, not the Internet. AI will be no different. Most players won’t make it, but the tech will endure and expand.
Or endure and contract.
The key difference between AI and the initial growth of the web is that the more use cases to which people applied the web, the more people wanted of it. AI is the opposite - people love LLM-based chatbots. But it is being pushed into many other use cases where it just doesn't work as well. Or works well, but people don't want AI-generated deliverables. Or leaders are trying to push non-deterministic products into deterministic processes. Or tech folks are jumping through massive hoops to get the results they want because without doing so, it just doesn't work.
Basically, if a product manager kept pushing features the way AI is being pushed -- without PMF, without profit -- that PM would be fired.
This probably all sounds anti-AI, but it is not. I believe AI has a place in our industry. But it needs to be applied correctly, where it does well. Those use cases will not be universal, so I repeat my initial prediction. It will endure and contract.
The difference is that the Internet was actually useful technology, whereas AI is not (so far at least).
I think you're exaggerating a little, but aren't entirely wrong. The Internet has completely changed daily life for most of humanity. AI can mean a lot of things, but a lot of it is blown way out of proportion. I find LLMs useful to help me rephrase a sentence or explain some kind of topic, but it pales in comparison to email and web browsers, YouTube, and things like blogs.
More use cases for AI than blockchain so far.
Quite a low bar.
Block chain is more like some gooey organic substance on the ground than a bar.
Can’t wait for all this cheap ddr5 memory and GPUs
I was looking at my Newegg orders recently. 7/18/2023 - 64GB (2 x 32GB) 288-Pin PC RAM DDR5 6000 (PC5 48000) --> $260. Now, $750+.
Don’t even get me started on this. I recently been shopping on eBay for some DDR4 memory. You may think - who’d need this dated stuff besides me? Yet 16Gb 3200Mhz is at least 60$. Which is effectively the price you paid for DDR5 6000. Crazy, right?
I have 4 32gb sticks of DDR5 6400 in my machine.
The RAM in my machine being worth more than the graphics card (7900XTX) was not on my bingo card I can tell you that.
Holy cow. I have 96GB of DDR5 I bought at start of year for a machine which never materialized. Might have to flip it.
Never in my dreams I could imagine PC parts could be an investment. Someone should start ETF tracking the prices.
For a while with bitcoin, it seemed GPU investing was almost a thing.
I just checked, the kit I bought in February was $270, today it is showing up for $1070. Woof. Now I have to decide if I should keep it on the off chance I do get around to that machine or dump it while the getting is good. Then again, who wants to buy RAM of unknown provenance unless they themselves are looking to scam the seller.
I'll give you $300 for it! With that kind of money, you can buy an ice-cold sixer and still have some walkin' around money.
GPUs have a very high failure rate…
To be honest ai datacentres would be a rip and replace to get back to normal datacentre density, at least on the cooling and power systems.
Maybe useful for some kind of manufacturing or industrial process.
Cheap compute would be a boon for science research.
It'll likely be used to mine bitcoin instead.
The GPUs, sure. The mainboards and CPUs can be used in clusters for general-purpose computing, which is still more prevalent in most scientific research as far as I am aware. My alma mater has a several-thousand-core cluster that any student can request time on as long as they have reason to do so, and it's all CPU compute. Getting non-CS majors to write GPU code is unlikely in that scenario.
> Getting non-CS majors to write GPU code is unlikely in that scenario.
People mostly use a GPU-enabled liblaplac. Physics, chemistry, biology, and medicine departments can absolutely use the GPUs.
I provide infrastructure for such a cluster that is also available to anyone at the university free of charge. Every year we swap out the oldest 20% of the cluster as we run a five year depreciation schedule. In the last three years, we’ve mostly been swapping in GPU resources at a ration of about 3:1. That’s in response to both usage reports and community surveys.
Why do you believe it will fail? Because some companies will not be profitable?
It wasn't an 'it' it was a 'some'. Some of these companies that are investing massively in data centers will fail.
Right now essentially none have 'failed' in the sense of 'bankrupt with no recovery' (Chapter 7). They haven't run out of runway yet, and the equity markets are still so eager, even a bad proposition that includes the word 'AI!' is likely to be able to cut some sort of deal for more funds.
But that won't last. Some companies will fail. Probably sufficient failures that the companies that are successful won't be able to meaningfully counteract the bursts of sudden supply of AI related gear.
That's all the comment you are replying to is implying.
Given the amounts being raised and spent, one imagines that the ROI will be appalling unless the pesky humans learn to live on cents a day, or the world economy grows by double digits every year for a few decades.
If the entire world economy starts to depend on those companies, they would pay off with "startup level" ROI. And by "startup level" I mean the amounts bullish people say startups funds can pay (10 to 100), not a bootstrapped unicorn.
I mean that is how capitalism works, no?
the constant cost of people and power won't make it all that much cheaper than current prices to put a server into someone's else rack.
>cash in on the floor space and HVAC systems that will be left in the wake of failed "AI" hype
I'd worry surveillance companies might.
you could stuff the racks full of server-rack batteries (lfp now, na-ion maybe in a decade) and monetize the space and the high capacity grid connect
most of the hvac would sit idle tho
Nobody really knows the future. What were originally consumer graphics expansion cards turned out useful in delivering more compute than traditional CPUs.
Now that compute is being used for transformers and machine learning, but we really don't know what it'll be used for in 10 years.
It might all be for naught, or maybe transformers will become more useful, or maybe something else.
'no way' is very absolute. Unlikely, perhaps.
> What were originally consumer graphics expansion cards turned out useful in delivering more compute than traditional CPUs.
Graphics cards were relatively inexpensive. When one got old, you tossed it out and move on to the new hotness.
Here when you have spent $1 trillion on AI graphics cards and a new hotness comes around that renders your current hardware obsolete, what do you do?
Either people are failing to do simple math here or are expecting, nay hoping, that trillions of $$$ in value can be extracted out of the current hardware, before the new hotness comes along.
This would be a bad bet even if the likes of OpenAI were actually making money today. It is an exceptionally bad bet when they are losing money on everything they sell, by a lot. And the state of competition is such that they cannot raise prices. Nobody has a real moat. AI has become a commodity. And competition is only getting stronger with each passing day.
You can likely still play the hottest games with the best graphics on an H200 in 5 years.
I find it disturbing how long people wait to accept basic truths, as if they need permission to think or believe a particular outcome will occur.
It was quite obvious that AI was hype from the get-go. An expensive solution looking for a problem.
The cost of hardware. The impact on hardware and supply chains. The impact to electricity prices and the need to scale up grid and generation capacity. The overall cost to society and impact on the economy. And that's without considering the basic philosophical questions "what is cognition?" and "do we understand the preconditions for it?"
All I know is that the consumer and general voting population loose no matter the outcome. The oligarchs, banking, government and tech-lords will be protected. We will pay the price whether it succeeds or fails.
My personal experience of AI has been poor. Hallucinations, huge inconsistencies in results.
If your day job exists within an arbitrary non-productive linguistic domain, great tool. Image and video generation? Meh. Statistical and data-set analysis. Average.
Just like .com bust from companies going online, there is hype, but there is also real value.
Even slow non-tech legacy industry companies are deploying chatbots across every department - HR, operations, IT, customer support. All leadership are already planning to cut 50 - 90% of staff from most departments over next decade. It matters, because these initiatives are receiving internal funding which will precipitate out to AI companies to deploy this tech and to scale it.
The "legacy" industry companies are not immune from hype. Some of those AI initiatives will provide some value, but most of them seem like complete flops. Trying to deploy a solution without an idea of what the problem or product is yet.
Well, at least it tells us something about the sentiment on hn that a lame insight around self admitted "napkin math" and obvious conflict of interest garners 400 points.
thats like boeing telling us we shouldnt build rockets
There are so many CEOs, tech experts, financial analysts and famous investors who say we are in an AI bubble - even AI-invested companies say that about themselves. My latest favorite "We are in an AI bubble" comment comes from Linus Torvalds himself in the video with Linus from Linus Tech Tipps [0]
[0]: https://www.youtube.com/watch?v=mfv0V1SxbNA&t=2063s
If it is so obvious that it won’t pay off, why is every company investing in it? What alpha do you have that they don’t?
That's a good question. During the .com boom everybody was investing in 'the internet' or at least in 'the web'. And lots of those companies went bust, quite a few spectacularly so. Since then everything that was promised and a lot more has been realized. Even so, a lot of those initial schemes were harebrained at best at the time and there is a fair chance that we will look in a similar way at the current AI offerings in 30 years time.
Short term is always disappointing, long term usually overperforms. Think back to the first person making a working transistor and what came of that.
I still don’t get it. What at a personal level is making Sam Altman make a suboptimal choice for himself if it is so obvious it won’t work out for him?
Marketing and name recognition? Used in other areas to generate wealth. Even steering OpenAI or partners of it to acquire right companies could be profitable for him.
On a personal level it will work out for him just fine. All he has to do is siphon off a fraction of that money to himself and/or an entity that he controls.
He's like Elon Musk in that respect: always doubling the bet on the next round, it is a real life Martingale these guys are playing with society on the hook for the downside.
https://en.wikipedia.org/wiki/Martingale_(betting_system)
How would he siphon off? He has no stake in OpenAI. And any other stake must be public. Or are you suggesting something more shady?
Elon is the complete opposite of martingale. He has helped produce value beyond just bets. Spacex, Tesla and so on.
Sorry, I didn't realize that was a rhetorical question.
it is not rhetorical. you have not provided a mechanism through which Sam can make money out of this whereas I'm describing ways in which this bubble popping is not something he would prefer.
Your lack of understanding can not be remedied with information.
You have a habit of 'just asking questions' when in fact you are trying to steer the conversation in the direction of the conclusion that you prefer. This is not productive. If you have something to say then you can just say it without the question marks. That way people know that you are making a statement instead of pretending to be asking a question.
I'm asking a specific question
>On a personal level it will work out for him just fine. All he has to do is siphon off a fraction of that money to himself and/or an entity that he controls.
How would he achieve this?
> And lots of those companies went bust, quite a few spectacularly so.
pets.com "selling dogfood on the internet" is the major example of the web boom then bust. (1)
But today, I can get dog food, cat food, other pet supplies with my weekly "online order" grocery delivery. Or I can get them from the big river megaretailer. I have a weekly delivery of coffee beans from a niche online supplier, and it usually comes with flyers for products like a beer or wine subscription or artisanal high-meat cat or dog foods.
So the idea of "selling dogfood on the internet" is now pervasive not extinct, the inflated expectation that went bust was that this niche was a billion-dollar idea and not a commodity where brand, efficiencies of scale and execution matter more.
1) https://en.wikipedia.org/wiki/Pets.com#History
At some point, I wonder if any of the big guys have considered becoming grid operators. The vision Google had for community fiber (Google Fiber, which mostly fizzled out due to regulatory hurdles) could be somewhat paralleled with the idea of operating a regional electrical grid.
Don’t worry. The same servers will be used for other computing purposes. And maybe that will be profitable. Maybe it will be beneficial to others. But This cycle of investment and loss is a version of distribution of wealth. Some benefit.
The banks and loaners always benefit.
That would be true for general purpose servers. But what they want is lots of special purpose AI chips. While is still possible to use that for something else, it's very different from having a generic server farm.
LAN party at the end of the universe?
I can't imagine everybody suddenly leaving AI like a broken toy and taking all special purpose AI chips offline. AI serves millions of people every day. It's here to stay even if it doesn't get any better than it is it already brings immense value to the users. It will keep being worth something.
Yes, it will be worth something. But when the value is less than its cost, you have a loss - and nobody runs a service at a loss.
What kind of reporte does the CEO of IBM expect the general technology workforce to hold for them?
How long can ai gpus stretch? Optmistic 10 years and we're still looking at 400b+ profit to cover interests. The factor in silicon is closer to tulips than rail or fiber in terms of depreciated assets.
This is likely correct overall, but it can still pay off in specific cases. However those are not blind investments they are targeted with a planned business model
Hypothetically speaking, if the AI hype bubble pops (or just returns to normalcy), would it be profitable to retarget the compute towards some kind of crypto mining? If so, could we expect the cryptocurrency supply to soar and the price to tank in short succession?
$8T may be too big of an estimate. Sure you can take OpenAI's $1.4T and multiply it by N but the other labs do not spend as much as OpenAI.
LLMs at current utility do not justify this spending, but the offside chance that someone will hit AGI is likely worth the expectation.
Mind you IBM makes +7B from keeping old school enterprise hooked up on 30 plus year old tech like z/OS and Cobol and their own super outdated stack. their AI division is frankly embarrassing. of course they would say that. IBM is one of the most conservative, anti-progress leaches in the entire tech industry. I am glad they are missing out big time on the AI gold rush. to me if anything this is a green signal.
One more data center and we will reach AGI, we promise, just give us more money.
How much of Nvidias price is based on 5 year replacement cycle? If that stops or slows with new demand could it also affect things? Not that 5 years does not seem very long horizon now.
The spending will be more than paid off since the taxpayer is the lender of last resort There's too many funny names in the investors / creditors a lot of mountains in germany and similar ya know
There is something to be said about what the ROI is for normal (i.e. non AI/tech) companies using AI. AI can help automate things, robots have been replacing manufacturing jobs for decades but there is an ROI on that which I think is easier to see and count, less humans in the factory, etc. There seems to be a lot of exaggerated things being said these days with AI and the AI companies have only begun to raise rates, they won't go down.
The AI bubble will burst when normal companies start to not realize their revenue/profit goals and have to answer investor relations calls about that.
“It doesn’t even have a keyboard!” energy.
The second buyer will make truckloads of money, remember the data center and fiber network liquidation of 2001+ - smart investors collected the overcapacity and after a couple of years the money printer worked. This time it will be the same, only the single purpose hardware (LLM specific GPUs) will probably end on a landfill.
The game is getting OpenAI to owe you as much money as you can. When they fail to pay back, you own OpenAI.
You are talking about the circular investments in the segment? Yes, but assume NVIDIA can get cheap access to IP and products of failing AI unicorns through contracts, this does not mean the LLM business can be operated profitably by them. Models are like fresh food, they start to rot by the training cut off date and lose value. The process of re-training a model will always be very expensive.
Are they LLM specific?
Deep down technically probably not, but they are optimized for this workload and business model. I doubt that once the AI bubble busts, another business model is viable. While datacenters can be downsized or partially shut down until demand picks up again, high end hardware is just losing money by the second.
Unless we get AGI.
Consumer will eat it all. AI is very good at engaging content, and getting better by the day: it won't be the AGI we wanted, but maybe the AGI we've earned
I guess he is looking directly at IBM's cash cow, the mainframe business.
But, I think he is correct, we will see. I still believe AI will not give the CEOs what they really want, no or very cheap labor.
The question no one seems to be answering is what would be the EOL for these newer GPUs that are being churned out of NVDIA ? What % annual capital expenditures is refresh of GPUs. Will they be perpetually replaced as NVIDIA comes up with newer architectures and the AI companies chase the proverbial lure ?
I think the key to replacing is power efficiency. If Nvidia is not able to make GPUs that are cheaper to run than previous generation theres no point for replacing previous generation. Time doesn't matter.
Ctrl-F this thread for terms like: cost, margin
Is transistor density cost still the limit?
Cost model, Pricing model
What about more recyclable chips made out of carbon?
What else would solve for e.g. energy efficiency, thermal inefficiency, depreciation, and ewaste costs?
The investors in these companies and all this infrastructure are not so much concerned with whether any specific companies pays off with profits, necessarily.
They are gambling instead that these investments pay out it in a different way: by shattering high labour costs for intellectual labour and de-skilling our profession (and others like it) -- "proletarianising" in the 19th century sense.
Thereby increasing profits across the whole sector and breaking the bargaining power (and outsized political power, as well) of upper middle class technology workers.
Put another way this is an economy wide investment in a manner similar to early 20th century mass factory industrialization. It's not expected that today's big investments are tomorrow's winners, but nobody wants to be left behind in the transformation, and lots of political and economic power is highly interested in the idea of automating away the remnants of the Alvin Toffler "Information Economy" fantasy.
Says the guy missing out on it
What does Jim Cramer have to say?
I suppose it depends on your definition of "pay off".
It will pay off for the people investing in it, when the US government inevitably bails them out. There is a reason Zuckerberg, Huang, etc are so keen on attending White House dinners.
It certainly wont pay off for the American public.
The same IBM that lost all races in the last 40 years? That IBM?
He will get assasinated or fired.
the mining industry enters the chat
A lot of you guys in the AI industry are going to lose your jobs. LLM and prompt ‘engineering’ experts won’t be able to score an AI job paying as well as a barista.
FB playbook. Act (spend) then say sorry.
"yeah, there's no way spending in those data centers will pay off. However, let me show you this little trinket which runs z/OS and which is exactly what you need for these kinds of workloads. You can subscribe to it for the low introductory price of..."
No wonder why he is saying that, they lost AI game, no top researcher wants to work for IBM. Spent years developing Watson, it is dead. I believe this is a company that should not be existed.
Maybe it's the opposite. IBM spent years on the technology. Watson used neural networks, just not nearly as large. Perhaps they foresaw that it wouldn't scale or that it would plateau.
IBM CEO is steering a broken ship and it's not improved course, not someone who's words you should take seriously.
1. The missed the AI wave (hired me to teach watson law only to lay me off 5 wks later, one cause of the serious talent issues over there)
2. They bought most of their data center (companies), they have no idea about building and operating one, not at the scale the "competitors" are operating at
Everyone should read his argument carefully. Ponder them in silence and accept or reject them in based on the strength of the arguments.
His argument follows almost directly, and trivially, from his central premise: a 0% or 1% chance of reaching AGI.
Yeah, if you assume technology will stagnate over the next decade and AGI is essentially impossible, these investments will not be profitable. Sam Altman himself wouldn't dispute that. But it's a controversial premise, and one that there's no particular reason to think that the... CEO of IBM would have any insight into.
then it seems like neither Sam Altman (pro) or IBM (proxy con) have credible or even really interesting or insightful evidence, theories ... even suggestions for what's likely to happen? i.e. We should stop listening to all of them?
Agreed. It's essentially a giant gamble with a big payoff, and they're both talking their books.
It's a very reasonable claim to make, but yes, average denizen of peanut gallery can spot this is a bubble from a mile way, doensn't need "insight" of napkin math done by some CEO that's not even in the industry.
Tho he's probably not too happy that they sold the server business to Lenovo, could at least earn something on selling shovels
we don't need AGI to use all that compute
we need businesses who are willing to pay for ai / compute at prices where both sides are making money
I for one could 10x my AI usage if the results on my side pan out. Spending $100 on ai today has ROI, will 10x that still have ROI for me in a couple years? probably, I expect agentic teams to increase in capability and more of my work. Then the question is can I turn that increase productivity into more revenues (>$1000 / month, one more client would cover this and then some)
IBM can be a hot mess, and the CEO may not be wrong about this. These things are not mutually exclusive.
Is his math wrong?
Are the numbers he's claiming accurate? They seem like big numbers pulled out of the air, certainly much large than the numbers we've actually seen committed to (not even deployed yet).
Sorry that happened to you, I have been there too,
When a company is hiring and laying off like that it’s a serious red flag, the one that did that to me is dead now
It was nearly 10 years ago and changed the course of my career for the better
make lemonade as they say!
IBM CEO has sour grapes.
IBM's HPC products were enterprise oriented slop products banked on their reputation, and the ROI torched their credibility when compute costs started getting taken seriously. Watson and other products got smeared into kafkaesque arbitrary branding for other product suites, and they were nearly all painful garbage - mobile device management standing out as a particularly grotesque system to use. Now, IBM lacks any legitimate competitive edge in any of the bajillion markets they tried to target, no credibility in any of their former flagship domains, and nearly every one of their products is hot garbage that costs too much, often by orders of magnitude, compared to similar functionality you can get from things like open source or even free software offered and serviced by other companies. They blew a ton of money on HPC before there was any legitimate reason to do so. Watson on Jeopardy was probably the last legitimately impressive thing they did, and all of their tech and expertise has been outclassed since.