But the technology also doesn’t have to be transformative either. I for one am not convinced.
The two things being true may as well be: LLM will not be transformative and the LLM business is mangled in terms of ROIC vs CoC.
In fact given the behavior of AI companies, I actually consider it more likely then not that the effects of the technology is severely over-hyped. I for one do not trust the words of the people who mangle their business in terms of ROIC and CoC. Why should I?
It is becoming increasingly hard not to see it as transformative. The man hours needed to do my work has been cut down by orders of magnitude. Wether this transformation is wholly a good thing is to be seen.
The world looks pretty different today. Even if the model maker companies go out of business (I’m skeptical), the model weights would stick around (many are open freeware already).
I know many software engineers who haven’t written code this year. AI chatbots are regularly used as alternatives to searching manually by many people. Agents are becoming a valuable new enterprise tool, and now consumers tech consumers are hopping on board.
The US spends a bit less than $400B/yr creating software and the worldwide spending is around $675B/yr.
Even if you eliminate 100% of the people involved in creating software, that's still not enough money (and of course, that's not going to happen because someone has to know what to build).
Everyone is adding agents to enterprise stuff, but an overwhelming majority of the general population now hate AI for most things -- especially the "AI support" these companies are using. I think most people would rather suffer through overseas call centers with absolutely terrible representatives than deal with AI support (studies seem to indicate 80+% prefer a human to AI for support in general).
AI as a search engine is useful, but a very different animal. Proficient users want a way to check the AI like Google's AI search does (though the links sometimes don't agree with the summary), but these run very small models (8b or so) with the RAG backend doing the real work.
How many competitors does this space need? Companies can build their own proprietary RAG search engines, but users would almost always prefer the company make that public data available to Google and just use one well-optimized search engine instead of dozens of bad copies.
What about profitability? Google enshittified their search to increase retention and ad time, but AI search should reduce retention/ad time AND costs a lot more money to run too meaning it should lower their bottom line. If that weren't enough, their RAG system is almost certainly more replaceable by users with alternatives than their traditional search system. This seems like all downside for Google.
> The US spends a bit less than $400B/yr creating software and the worldwide spending is around $675B/yr.
The currently well-served market for AI is not software engineering, it's approximately all of white-collar work, ranging from accounting and law, through medicine, general office work, to school administration, education, NGOs and governance.
Not everyone is going to just publicly brag about their AI use, but it's an open secret everyone is either using LLMs for half their work, or - if for some reason they're not busy enough to arrive at this idea on their own - under pressure to start using them.
If my ridiculous $100+/m subscription to the single dumb pipe that has a government-endorsed monopoly in my area is anything to go by, they're extracting plenty of value from this customer at least.
Which is northing compared to what google and other networking/application companies get from businesses to use their digital, no or little in comparison physical maintenance, products.
Which is still nothing compared to what users get out of this, which is why the technology continues to work and improve despite occasional bubble-popping events leaving some investors without their shirts.
Also computers. DEC and IBM and Gateway and whatever didn't extract the economic benefits of digitization. Even Apple extracts a vanishing fraction of the economic benefits the iPhone has created.
Well, you can extract all the value created using your tools and inventions, because then no one would have any incentive to use them to create value.
They try to get around this by charging you more after you've invested in your tools, but even then you can only go so far (though it's astonishing how far the big guys have managed to get regardless)
One thing I'm wonder about: previous busts left useful infrastructure behind (rail, fibre) that could still be used. If/When the AI/LLM companies go bust, what will be left over afterwards and how useful will it be?
> previous busts left useful infrastructure behind (rail, fibre) that could still be used. If/When the AI/LLM companies go bust, what will be left over afterwards and how useful will it be?
I’m pretty sure that when the railroads went bust that nobody pulled up the rails and put them somewhere else. And when the internet bust happened, they didn’t pull up all the dark fiber and lay it down somewhere else.
I would bet that if all those GPUs go idle, they stay housed and cooled in the data center they were originally installed in, for better or for worse.
Anyone who might be able to use the waste heat while remaining profitable via income from operations. It’s a shame that data centers aren’t being used for district heating, for example.
LLM technology will not go away. The premise here is that the foundation model companies will bust. So now there will be tons of idle GPU, where companies like AWS and Azure, or even Coreweave and Lambda, can pick them up for cheap and rent them out to people who want to run open source/open weight models.
LLMs as of whatever open models are SOTA when the proprietary bubble bursts, plus... whatever else you can do on GPUs - weather modeling, CFD, genetics research, and so on, and so on.
And that's at day 0. Because LLMs are not going away, thank to open weight models, and the race for capabilities has left so, so many low-hanging fruits unpicked, we'd have a decade or more of useful R&D to do even if LLM capabilities suddenly plateaued tomorrow and never improved.
I mean, look at Jev making round in the industry now - this is just a single example of a low-hanging fruit that took many years for someone to bother to pick up and market a bit. There's many, many more of these just laying around.
> Anthropic said nearly a quarter of its revenue came from two customers last year, and as part of its risk factors, warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending.
Is it possible those are cloud providers? Someone using Claude on google vertex might just switch to Claude on AWS or Anthropic. Also I don’t imagine cloud providers suddenly cutting the cord.
Why xAI? They have no market share and their own data centers are over 90% idle/rented out to the other AI shops, so it's more likely that the money flows in the opposite direction.
There won’t be a bailout. I doubt the obligations will even amount to anything substantial anyways.
Many of these obligations are with the big clouds who don’t have capacity to serve them anyways. Google has said they have hundreds of billions in purchase obligations they can’t fill because they don’t have enough TPUs and data centers. Google gets a bigger marketing and investor headline, and there’s a queue behind Anthropic if they bail, so of course everyone would sign the commitments knowing the risks.
I’m sure AWS and Azure are on the same page.
In the worst case, many of their biggest customers have tons of GPUs (Meta, AWS), so they could always license out their raw models at steep discount to help absolve themselves of the obligations.
Finally, in the worst case scenario, even with margin compression, they pre-purchased most of the compute so there’s not enough for the open models to run on. If Opus was the same price as Kimi/GLM, I doubt many people would pick the open models.
Yeah not sure about that. OpenAI and oracle are already losing money because of their obligations. They are already going to the worst case scenario and it's worse than you describe.
The numbers that Reuters describe in this article were entirely for 2025.
It's been well documented that Anthropic's revenue growth in 2026 has been enormous. This was the year of coding agents, and tokenmaxxing, and companies blowing enormous amounts of money on AI thanks to coding agent users spending hundreds (or thousands) of dollars a day.
Given that, I don't understand why the article and the headline are based exclusively on those 2025 numbers, with not so much as a hint to the reader that there are figures from the past 9 months that aren't covered by the documents Reuters saw.
Losing $8bn in 2025 isn't particularly notable if you've made ~10x that amount of revenue in 2026.
(How much did they lose in 2026? Wouldn't we love to know that!)
Is this just a thing with leaked IPO prospectuses and coverage of them?
Reuters probably saving 2026 for another writeup. To be honest, this is nothing surprising for 2025. I think we all expected unsurmountable costs. AI has always been and still is a huge bet into the future. The future is not written.
>It's been well documented that Anthropic's revenue growth in 2026 has been enormous.
Has it? Don't they just self-leak/press release 'ARR' revenue claims? Has there been any audited and formal filings? Including all the off-balance sheet and other creative stuff
We've also heard plenty of stories from companies like Uber about a need to set a limit on their token spending because they were spending so much on Anthropic. That's a solid sign that the revenue numbers are real.
What Uber and other companies said is that the money they spent didn't generate the expected value they were promised. If it had generated $2 for every $1 spent, then Uber would be spending with no limits.
The lack of demonstrable return on investment means there is a ceiling to the market size, and profit is capped as cheaper models and self hosted open source models limit the amount flagship models can charge.
I didn't mean 'no ROI', clearly AI is a productivity boost.
For the AI companies to IPO at the money they want, they have to promise that they are going to be worth many multiples more of that value.
But currently they are spending hundreds of billions and despite this their models are not light years ahead of each other, or of the companies with much lower budgets and compute power.
What they claim they can offer in future profits is subject to considerable debate.
It's in the interests of VC's and their underlying rich investors to hype it ip, IPO and bank the cold hard cash.
If Uber caps employee AI spend at $1500, then they've just admitted per employee AI spend on par with premium health care or commercial rent both in the 1000-1500 range, and this limited to just technical employees, i.e. subsect so presumably AI as % of business cost even lower. Health insurance 1T industry... commercial RE like 100 billion...
Anthropic thinks it's worth more than insuring all US workers AND putting roof over their heads. AI sector thinks its worth magnitude more in aggregate. If they fully displace people, i.e. untied to headcount eventually then sure, but if not, based on what Uber pays, AI is basically worth as much as insurance.
Simon you're really going to need to float the numbers for us if you're going to make claims about enormous revenue. And don't forget to include things like cost of revenue as well !
I assembled some of those numbers in May, when they claimed they had grown from $9bn in annualized revenue in December 2025 to $47bn in early May. https://simonwillison.net/2026/May/29/anthropic/
And sure, they might be lying about those figures - but if they are, that's investor fraud, and they'll be in hot water with the SEC when they try to IPO. I don't think they are lying about the figures.
If I had numbers on their cost of revenue I would share those. As it stands I'm going to have to wait for either more leaks or their S-1.
This isn't a standardized metric like GAAP revenue, and the company chooses how to compute it. Nothing stops them from taking a strong month (or week, or day, or hell even a second of revenue) and multiplying it out. It's also a different thing from booked revenue: the prospectus shows about $4.6B recognized in 2025, versus a ~$9B run rate at year-end.
Even taken at face value, it tells us nothing about costs. In 2025 they lost about $8B operating on $4.6B revenue, with compute alone at $7.3B. The $42B net loss is inflated by a ~$34B non-cash charge, so I'm ignoring that. The company says it hit positive adjusted operating income in Q2, but that's their own number, "adjusted" excludes whatever they choose such as training costs or the cost of staff, and we won't know the real picture until the full S-1 is public. Meanwhile they've disclosed $518B in future compute and infrastructure commitments against ~$20B in cash.
If they had a trillion dollars worth of revenue, it wouldn't mean jack shit if they had a trillion + 1 in losses.
> If they had a trillion dollars worth of revenue, it wouldn't mean jack shit if they had a trillion + 1 in losses.
I don't understand that argument.
If a company has a trillion dollars in revenue, even if they are losing money hand-over-fist, that still means they have convinced other companies to cough up a trillion dollars for what they are selling. That's a big deal!
The only case that isn't impressive is if they are literally selling dollar bills for 90 cents.
You can argue that Anthropic are subsidizing their tokens all you want, but since as a customer you can't just turn around and sell a token yourself for more than you paid for it that's still not a good argument for dismissing the amount people are willing to spend.
The "selling a dollar for ninety cents" thing only works for liquid products that can be resold.
If you make a trillion dollars by allowing other people to make a profit through arbitrage, that trillion dollars is meaningless.
If you subsidize a product such that you spend more money producing it than your customers pay you, but that product is NOT something they can sell on directly and pocket the difference, then selling a trillion dollars of it is still impressive because it demonstrates that you have a product people assign a trillion dollars of value to.
That was my question too. 2025 is really old in AI-company time. But I'm suspecting the "leak" was on purpose so they can now come and show how much better (but still not great) the first half of 2026 was.
You know the losses increased even more massively or Anthropic itself would have leaked the number already...
Most interesting: "Anthropic said nearly a quarter of its revenue came from two customers last year, and as part of its risk factors, warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending."
But that was in 2025. It's not surprising to hear that in 2025 they had two customers responsible for a quarter of their revenue - they hadn't yet sold Claude Code packages to vast numbers of large companies.
Gaap net loss almost 42B on less than 5B of revenue, I cant even imagine. Where does all the money go? If the name attached to this was anything other than Anthropic these would be some seriously horrible financials, and that's before getting into all the colorful accounting they're using.
This is the part that really surprised me: "Anthropic said nearly a quarter of its revenue came from two customers last year".
Who are these two customers? A quick Google search seems to indicate that Cognizant is their largest "enterprise" customer by number of seats, Microsoft is their largest API customer, and the $10B contract they recently signed with Meta is their largest contract ever. But those metrics don't necessarily translate directly into annual revenue.
> the gross margins right now are very positive. [...] So the company loses money. Each model makes money, but the company loses money.
Amodei mentioned the reason for the big losses in an interview released on Feb 2026[1]. Assuming he was telling the truth, that's not a bad position to be in. From what I can tell from the replies to simonw, a lot of people are cynical about Anthropic's growth potential. I'd like to encourage the cynics to bet against Anthropic wish them luck. I will be doing the opposite.
87 comments
[ 0.29 ms ] story [ 62.9 ms ] thread-LLMs profoundly change society -The LLM business is mangled in terms of ROIC vs CoC
Such a business already exists - airlines.
This nuance is what many here refuse/find difficult to understand.
The two things being true may as well be: LLM will not be transformative and the LLM business is mangled in terms of ROIC vs CoC.
In fact given the behavior of AI companies, I actually consider it more likely then not that the effects of the technology is severely over-hyped. I for one do not trust the words of the people who mangle their business in terms of ROIC and CoC. Why should I?
The world looks pretty different today. Even if the model maker companies go out of business (I’m skeptical), the model weights would stick around (many are open freeware already).
I know many software engineers who haven’t written code this year. AI chatbots are regularly used as alternatives to searching manually by many people. Agents are becoming a valuable new enterprise tool, and now consumers tech consumers are hopping on board.
Even if you eliminate 100% of the people involved in creating software, that's still not enough money (and of course, that's not going to happen because someone has to know what to build).
Everyone is adding agents to enterprise stuff, but an overwhelming majority of the general population now hate AI for most things -- especially the "AI support" these companies are using. I think most people would rather suffer through overseas call centers with absolutely terrible representatives than deal with AI support (studies seem to indicate 80+% prefer a human to AI for support in general).
AI as a search engine is useful, but a very different animal. Proficient users want a way to check the AI like Google's AI search does (though the links sometimes don't agree with the summary), but these run very small models (8b or so) with the RAG backend doing the real work.
How many competitors does this space need? Companies can build their own proprietary RAG search engines, but users would almost always prefer the company make that public data available to Google and just use one well-optimized search engine instead of dozens of bad copies.
What about profitability? Google enshittified their search to increase retention and ad time, but AI search should reduce retention/ad time AND costs a lot more money to run too meaning it should lower their bottom line. If that weren't enough, their RAG system is almost certainly more replaceable by users with alternatives than their traditional search system. This seems like all downside for Google.
The currently well-served market for AI is not software engineering, it's approximately all of white-collar work, ranging from accounting and law, through medicine, general office work, to school administration, education, NGOs and governance.
Not everyone is going to just publicly brag about their AI use, but it's an open secret everyone is either using LLMs for half their work, or - if for some reason they're not busy enough to arrive at this idea on their own - under pressure to start using them.
Also computers. DEC and IBM and Gateway and whatever didn't extract the economic benefits of digitization. Even Apple extracts a vanishing fraction of the economic benefits the iPhone has created.
They try to get around this by charging you more after you've invested in your tools, but even then you can only go so far (though it's astonishing how far the big guys have managed to get regardless)
People seem to have forgotten the bust in railroads and Internet (etc), even though those technologies were transformative:
* https://en.wikipedia.org/wiki/Technological_Revolutions_and_...
One thing I'm wonder about: previous busts left useful infrastructure behind (rail, fibre) that could still be used. If/When the AI/LLM companies go bust, what will be left over afterwards and how useful will it be?
Many many idle GPUs.
I would bet that if all those GPUs go idle, they stay housed and cooled in the data center they were originally installed in, for better or for worse.
Yes, but what use will they have? Rail lines and dark fibre could be used post-bust (and don't depreciate all that quickly).
And that's at day 0. Because LLMs are not going away, thank to open weight models, and the race for capabilities has left so, so many low-hanging fruits unpicked, we'd have a decade or more of useful R&D to do even if LLM capabilities suddenly plateaued tomorrow and never improved.
I mean, look at Jev making round in the industry now - this is just a single example of a low-hanging fruit that took many years for someone to bother to pick up and market a bit. There's many, many more of these just laying around.
Don't focus on the hardware. The models themselves are tremendously valuable. These tools would've been considered alien technology just 15 years ago.
- $42 billion net loss in 2025
- $518 billion in infrastructure obligations in the next year
- 1/4 of revenue coming from 2 customers
That spending obligation sounds extremely unlikely to be fulfilled without a bailout. Anthropic's IPO is them passing the hot potato forward.
Many of these obligations are with the big clouds who don’t have capacity to serve them anyways. Google has said they have hundreds of billions in purchase obligations they can’t fill because they don’t have enough TPUs and data centers. Google gets a bigger marketing and investor headline, and there’s a queue behind Anthropic if they bail, so of course everyone would sign the commitments knowing the risks.
I’m sure AWS and Azure are on the same page.
In the worst case, many of their biggest customers have tons of GPUs (Meta, AWS), so they could always license out their raw models at steep discount to help absolve themselves of the obligations.
Finally, in the worst case scenario, even with margin compression, they pre-purchased most of the compute so there’s not enough for the open models to run on. If Opus was the same price as Kimi/GLM, I doubt many people would pick the open models.
The numbers that Reuters describe in this article were entirely for 2025.
It's been well documented that Anthropic's revenue growth in 2026 has been enormous. This was the year of coding agents, and tokenmaxxing, and companies blowing enormous amounts of money on AI thanks to coding agent users spending hundreds (or thousands) of dollars a day.
Given that, I don't understand why the article and the headline are based exclusively on those 2025 numbers, with not so much as a hint to the reader that there are figures from the past 9 months that aren't covered by the documents Reuters saw.
Losing $8bn in 2025 isn't particularly notable if you've made ~10x that amount of revenue in 2026.
(How much did they lose in 2026? Wouldn't we love to know that!)
Is this just a thing with leaked IPO prospectuses and coverage of them?
What Uber and other companies said is that the money they spent didn't generate the expected value they were promised. If it had generated $2 for every $1 spent, then Uber would be spending with no limits.
The lack of demonstrable return on investment means there is a ceiling to the market size, and profit is capped as cheaper models and self hosted open source models limit the amount flagship models can charge.
Shouldn't they have set that per-employee budget to zero instead?
(I dug up the original source for that Uber doubts the ROI story a few months ago, it's a lot weaker than the headlines about it suggested: https://simonwillison.net/2026/May/27/product-market-fit/#th...)
For the AI companies to IPO at the money they want, they have to promise that they are going to be worth many multiples more of that value.
But currently they are spending hundreds of billions and despite this their models are not light years ahead of each other, or of the companies with much lower budgets and compute power.
What they claim they can offer in future profits is subject to considerable debate.
It's in the interests of VC's and their underlying rich investors to hype it ip, IPO and bank the cold hard cash.
Anthropic thinks it's worth more than insuring all US workers AND putting roof over their heads. AI sector thinks its worth magnitude more in aggregate. If they fully displace people, i.e. untied to headcount eventually then sure, but if not, based on what Uber pays, AI is basically worth as much as insurance.
Since then they've reported $65bn in annualized revenue by July: https://simonwillison.net/2026/Aug/23/anthropics-best-ai-mod...
And sure, they might be lying about those figures - but if they are, that's investor fraud, and they'll be in hot water with the SEC when they try to IPO. I don't think they are lying about the figures.
If I had numbers on their cost of revenue I would share those. As it stands I'm going to have to wait for either more leaks or their S-1.
By whom, exactly?
https://www.reuters.com/technology/anthropic-revenue-run-rat...
This isn't a standardized metric like GAAP revenue, and the company chooses how to compute it. Nothing stops them from taking a strong month (or week, or day, or hell even a second of revenue) and multiplying it out. It's also a different thing from booked revenue: the prospectus shows about $4.6B recognized in 2025, versus a ~$9B run rate at year-end.
Even taken at face value, it tells us nothing about costs. In 2025 they lost about $8B operating on $4.6B revenue, with compute alone at $7.3B. The $42B net loss is inflated by a ~$34B non-cash charge, so I'm ignoring that. The company says it hit positive adjusted operating income in Q2, but that's their own number, "adjusted" excludes whatever they choose such as training costs or the cost of staff, and we won't know the real picture until the full S-1 is public. Meanwhile they've disclosed $518B in future compute and infrastructure commitments against ~$20B in cash.
If they had a trillion dollars worth of revenue, it wouldn't mean jack shit if they had a trillion + 1 in losses.
I don't understand that argument.
If a company has a trillion dollars in revenue, even if they are losing money hand-over-fist, that still means they have convinced other companies to cough up a trillion dollars for what they are selling. That's a big deal!
The only case that isn't impressive is if they are literally selling dollar bills for 90 cents.
You can argue that Anthropic are subsidizing their tokens all you want, but since as a customer you can't just turn around and sell a token yourself for more than you paid for it that's still not a good argument for dismissing the amount people are willing to spend.
> but since as a customer you can't just turn around and sell a token yourself for more than you paid
You can't purchase a dollar for 90 cents and sell it for 95 if the original seller keeps selling for 90.
If you make a trillion dollars by allowing other people to make a profit through arbitrage, that trillion dollars is meaningless.
If you subsidize a product such that you spend more money producing it than your customers pay you, but that product is NOT something they can sell on directly and pocket the difference, then selling a trillion dollars of it is still impressive because it demonstrates that you have a product people assign a trillion dollars of value to.
Most interesting: "Anthropic said nearly a quarter of its revenue came from two customers last year, and as part of its risk factors, warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending."
FYI. Not sure why everyone is fixating on the 2025 numbers without even bothering to look up the 2026 numbers…
Who are these two customers? A quick Google search seems to indicate that Cognizant is their largest "enterprise" customer by number of seats, Microsoft is their largest API customer, and the $10B contract they recently signed with Meta is their largest contract ever. But those metrics don't necessarily translate directly into annual revenue.
Amodei mentioned the reason for the big losses in an interview released on Feb 2026[1]. Assuming he was telling the truth, that's not a bad position to be in. From what I can tell from the replies to simonw, a lot of people are cynical about Anthropic's growth potential. I'd like to encourage the cynics to bet against Anthropic wish them luck. I will be doing the opposite.
[1] https://www.dwarkesh.com/p/dario-amodei-2
What's in it for them?