IBM CEO says there is 'no way' spending on AI data centers will pay off
(businessinsider.com)561 points by nabla9 17 hours ago
561 points by nabla9 17 hours ago
> 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.
For the fact that they invented Deep Blue, they are really struggling with AI
> 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.
IBM was making "calculating cheese cutters" back in the day [0].
I'm sure they can pivot to something else if the need arises.
> 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.
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
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.
>> 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.
> 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.
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.
[0] As another commenter points out: https://news.ycombinator.com/item?id=46131245
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.
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.
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.
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.
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.
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.
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.
What does that have to do with the current CEO's assessment of the situation?
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.
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?
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".
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.
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
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.
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.
> 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.
You could try addressing the actual topic of discussion vs this inflammatory and lazy "dunk" format that frankly, doesn't reflect favorably on you.
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 ?
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...
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.
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.
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?
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.
Sure but these are likely just variations of existing things. And yet the quality is still behind the original
> 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.
> 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.
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"?
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.
> 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.
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.
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.
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.
> normal servers have 7 year lifespans in many cases fwiw
Eight years if you use Hetzner servers!
It's just the same dynamic as old servers. They still work fine but power costs make them uneconomical compared to latest tech.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
> 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?
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.
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.
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)
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.
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-...
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.
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?
> 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...
> 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.
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.
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.
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"
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.
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.
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?
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.
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.
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.
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.
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.
A lot of it is propped by the fact with GPU and modern server CPUs the die area just got bigger
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)
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.
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.
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
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 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