So I'm reading the actual IMF reports itself [1] and I come across this on the first page:
"First, although core firms in the AI ecosystem have been profitable so far—easing pressure on AI-related equity valuations—weaker or less durable investment and profitability in AI could trigger a correction"
Is that correct - that the AI firms have been "profitable"?
They are talking about Meta, AWS, Microsoft, Google, Nvidia and so on.
Although Anthropic supposedly has a good margin on each token, so the inference is substantially less expensive than what they charge customers. Training is eating into that.
> Although Anthropic supposedly has a good margin on each token
I am still to see any evidence of this being true.
The actual data at hand suggests otherwise, with Anthropic's losses increasing with more usage, and they relying on shenanigans to give an appearance of being healthy. They also need to raise money like crazy and seem desperate to offload this hot potato through an IPO. Not really the sort of thing that screams "sustainable" to me.
Such as having a "profitable" quarter on one of their funny metrics when they were not paying for compute, or leaking completely obscure ARR, "EBTIT", etc.
What data? To my knowledge the only data point that exists is a 2025 document, which is completely irrelevant considering how massively they hiked the API token price.
To my knowledge, the subscriptions are the only part that may still be profiting from vc subventions. And if not, then their API margins are gigantic
- The suspiciously high amount of money they have to keep raising at a rapid pace
- The fact that they have to play around with silly metrics and leak bullshit to pretend they are healthy
- Their desperation for an IPO.
I feel kind of silly having to repeat myself, it was all there in my previous reply of you bothered to read. I point at that whenever people spread this bullshit of Anthropic being "profitable" especially when it comes to subscriptions. Everything points at it being ruinously unprofitable actually.
Profitability only matters if we consider that they are grifters willing to cash in as quickly as possible before this jenga tower falls apart. It matters less if the interpretation is that the fucking weirdos at Anthropic are death cultists that want to conjure their death god into existence.
but none of those are data points which actually meaningfully tell what we were discussing? They're not even data points, theyre merely observations from the outside, not at all objective.
Anthropic can be profitable and massively overvalued, AND deeply into debt from running too fast and loose with their money in previous years.
That situation would have them behave exactly the same: hurrying to get the IPO done while the stocks are still skyhigh to get over the debt burden and then buckle up for the profitability to weather them through coming correction...
Because it's kind of true, but also mostly wrong. People conflate two things: "Subsidized" subscriptions as a monetization model, and (unsustainable) subsidization through outside investors.
When you max out your Claude Max plan every billing period, are you consuming more tokens than you could realistically pay for with the same money spent on API inference? Definitely! This works because many Claude Max subscribers don't use it to the max (and you probably aren't using it to the max every single time either), so overall subscriptions might still be profitable. Technically, the subscribers subsidize each other, but this is entirely sustainable (see any other subscription model that is based on its consumers not maximizing their usage, e.g. gyms).
What people mean (or want to say) when they say inference is subsidized is that it is subsidized in an unsustainable way (e.g. with investments that will not last forever - if this is true at scale it should worry us!). What happens now is that they take the fact that an individual subscription used to the max by itself is a financial drain for the company offering it, attach the "subsidized" term to it, and infer from that a bunch of stuff that is usually associated with the term (such as unsustainability) but doesn't really apply here.
whether people want to use the service or not is irrelevant, the only thing that matters is whether or not (and how much) they actually do use it. Also there are subscription/flat rate based cinema services that have been in business for almost two decades now, so it doesn't seem like that wouldn't work.
Because the subscriptions grant you much more usage than what you get in tokens for the same money via the API. That being said, they probably don't have ruinous losses per subscription, if any. Don't forget that many users are casual and do not use Claude Code, so a casual user and a power user level out
Because they are, I pay anthropic 200 a month, if I were paying by the token like an enterprise it would be close to 11 grand a month. Yes I use 100 percent of my allocation on every window.
That’s only a subsidy if you argue about lost revenue (like the anti-piracy arguments). But somehow I doubt you’d be willing/able to pay even close to 11 grand a month.
The only real comparison point that matters for the argument of subsidization is how much money it cost costs Anthropic to serve you those tokens.
I haven't digged myself but many people have argued that Anthropic used accounting tricks to appear profitable to investors, but they removed some costs from their calculation.
I think the message to investors is that operating the model is profitable. Like the income they get from paying customers pays for the infrastructure of running the models.
Maybe - though they refer to hyperscalers explicitly in other parts of the report, and then there is this:
"Hyperscalers have strong earnings from their core businesses. Yet, free cash flows are rapidly declining as hyperscalers try to sustain large AI investment needs, leaving debt financing increasingly necessary"
Oh man I wish a publication would go with that future headline. "If AI agents were supposed to replace working people, will they be footing the bailout bill as well?" hmm maybe it's a bit too long for a headline.
The debt lasts longer than GPUs remain profitable, right? GPUs loose value most quicker than most investors believe. In best case they last like 2-3 years under heavy stress, so I think about all those data centers and what will it mean to replace those cards burned down - you cannot even replace one with newer one - all 8 GPUs should be the same type. Thinking about future costs with all the debt.
With an AI bubble implosion now all but inevitable, the shift of the key players behind the scenes has focused on making this seem too big to fail and thus needing a bailout.
The “it’s dangerous, regulate us please” angle as an out for explaining terrible financial performance doesn’t seem to be working. The final Hail Mary pass is the “if we fail you fail” play.
However all data indicates the risk here is fairly concentrated in private markets. It will be painful but it’s unlikely to take the whole economy out with it. Some VCs are in for a world of hurt and private credit markets will lose their shirts, but the average person would see limited direct impact.
With everything going on in the datacenter space a bailout for AI seems incredibly unlikely. The general sentiment amongst the masses seems to be “let it burn.”
The problem is that private credit agencies also asked for loans from banks and therefore they are exposed to this as well. I believe the current estimate is that 11% of outstanding loans from the big 4 US banks were issues to private credit firms which then very likely lent themselves to the AI industry complex.
Additionally, by the Fed's own admission, they will never let another "Great Depression" happen again. So they intervened in 2008 and they will intervene again by buying up those worthless assets. And the average person will see their savings real value plummet without even mentioning that a lot of private credits come from pension funds...
Finally, if private credit really goes belly up, private investors will likely lose a good portion of their purchasing power, meaning that the economy will at best correct and at worst crash only because of that. And this effect will be in addition to the AI economy grinding to a halt which will already hit the worldwide economy HARD (also IMHO).
A reckoning is coming. I believe the "let it burn" is not doable as it will take a few banks and the economy with it. A more measured slow bleed with the Fed printing money to buy these junk assets will be more likely.
Some banks will suffer losses but there isn’t data to say the banks will fail en mass. This is bad, but not 2008 bad in so far as the stability of the financial system.
The problem is that nobody knows how really bad this is. For instance, if I, a private individual, asked for a loan to put money in private credit there is no way to link these two actions.
The 11% I mentioned above is the verifiable exposure for banks. It could be a much much higher number and at a time where delinquencies on all sorts of debts are historically high that for sure doesn't bode well.
The money printing that the Fed would do to fix this would be gigantic (several Trillion dollars basically) and on the same scale as the covid intervention which sent inflation to 9+%. And we are already waiting for the oil shock to spread to everything else.
Yeah, the financial system might survive, but the pensions, savings, ... will be fucked up.
You underestimate the corruption of the White House. I've no doubt there will be a massive bailout of the guys who sat up there on the dais during his inauguration. Zuck, Musk, Pichar... He's hosted Altman and Amodei recently, too, right?
As many have commented, not really. Shorting is about exact timing as much as knowing the inevitable outcome. Get the timing just a bit wrong and you can be right and broke. The Big Short portrays that quite well.
Credit default swaps. Basically an insurance that pays if a stock/asset value plummets. The problem is that you have to pay for this insurance coverage, and currently the prices are really really high
[1], meaning that the market is expecting a failure soon, especially with interest rates/bonds yields going up.
> Some VCs are in for a world of hurt and private credit markets will lose their shirts,
They will walk out only rich rather then super rich. They will talk about it as if they were poor, but nope, they will still have toms of power and money.
If AI share crashes, that’s actually good for the public because all the value is now captured by them instead of companies. This is because a crash can only happen if prices go down (which is good) or Chinese models become as good as frontier (which is also good).
The question is, if this happens then how do you incentivise more R and D? Which investor would invest without the returns?
Edit: not surprisingly people don’t understand the point.
Would you rather NVIDIA and OpenAI get rich and capture the value or the public? Such a strange thing to see people rooting for the big companies.
In the short term things tend to freeze up as those with funds to invest lock down the hatches and weather the storm. But fairly quickly after yes funds start flowing elsewhere as money chases the next big thing.
AI is here to stay, but the market needs a big forest fire to wipe most things out. A few strong plants survive to build a better future.
It’s a cycle that keeps repeating, and we’re overdue for a big forest fire in tech.
> This is because a crash can only happen if prices go down (which is good) or Chinese models become as good as frontier (which is also good).
No... This is not accurate at all. AI valuations are based on high future growth. Prices could stay the same but growth slows and that could cause a crash by itself.
Right. AI companies have P/E ratios rarely seen for very large companies. Except for TSLA, which has a P/E ratio of 383, unheard of at that scale. No other large-cap company is much over 40.
TSLA should have crashed by now, but the hype just keeps on coming. Eventually reality will set in, but the market can remain irrational for longer than you can remain solvent. Big AI may be like that.
Investors like a story that lets their imagination run wild. TSLA offers that. AI / NVIDIA has that 10-fold. It’s “man’s last invention” - you couldn’t come up with a better hook to raise capital.
Short of an emergency rate hike to 10%, nothing stops this bubble.
It's just circular financing - we give you GPUs, you give us your shares of a promised future $ 2,000,000,000,000 valuation. Despite they are unprofitable today and tokens are getting cheaper and cheaper.
False. They are based on future revenues that companies can capture. If the companies can’t capture the value and there’s no moat, then the public stands to gain from it.
I think what you don't fully understand is that if these companies crashed and stopped getting in new investor money then lowering prices would not be the answer because they would not actually have enough revenue to even pay their own bills. Right now the AI costs are partly being subsidized just like when Uber first came out and rides only cost like $6 when it might now be $40. While what you are saying might come true it doesn't mean that it's directly related to what we are talking about, these companies crashing is not going to magically just result in lower prices. Lower prices are likely but would be due to other factors. Also, lowering prices is not going to cause a crash by itself. Just look how cheap and capable GPT-6 Luna is compared to GPT-3.5. That didn't cause OpenAI to crash, did it?
They can't properly meet the current demand. A crash will primarily hit NVIDIA, AMD and other AI chip makers as they have oversold already and the cards are sitting in storage waiting for the needed data centers and the power needed to run them.
Its like a gold rush where half of the picks are sitting in storage because there are not enough miners to wield them.
Every crisis is an opportunity. Maybe we could build something where we don't have ridiculous power and wealth concentration at the expense of the public - to the point where the world's richest man literally took over the US government to cut aid to the world's poorest.
Most people who are currently enjoying the benefits generative AI don't even have pensions, let alone ones that are heavy into AI stocks. Wtf are you even talking about?
GP jen729w was talking about that particular scenario, though. I have never claimed that would be an issue for everyone. But it will be an issue for a lot of people, if the crash is hard and recovery is slow.
This comment shows the exact problem with the incoherent ideology prevalent in some bubbles.
You fundamentally believe that AI is useless and that it wastes energy and hardware. Shouldn't you then want the bubble to burst even faster? If not, you are just delaying the bubble.. but to what end? So that you personally won't be affected? This is a morally bankrupt stance to take.
My ideology is at least more coherent in this context. I think AI is useful, and AI being cheaper for the masses is good for them rather than the value being captured by a few companies.
But it's objectively true that if the market crashes tomorrow, lots of normal people lose money. This hits extra hard – which happened back in 2007/8 – if that's the day your life-long savings are converted from stocks to cash.
Are you arguing that because those pensioners can now get a frontier model cheaper than they can today that this is a net benefit?
No, I disagree. This is not what happened in 2008 which was fundamentally different because it was the people that held the bad debts.
If market crashes tomorrow due to commoditised models, it is true that those people who bought these stocks would lose money. But its honestly not a big part of it - if you run the numbers, pensioners will lose 10% of their money.
This is a transfer of wealth from people who have exposure to these stocks vs everyone else. Do you disagree with this transfer of wealth? It comes down to that.
> This comment shows the exact problem with the incoherent ideology prevalent in some bubbles.
You're shadowboxing a strawman you've built, my friend.
> You fundamentally believe that AI is useless and that it wastes energy and hardware.
That is not what I believe. I believe that AI is useful at concentrating power in the oligarchy and that increasing energy and hardware cost is a nice side effect that further helps that goal.
> Shouldn't you then want the bubble to burst even faster? If not, you are just delaying the bubble.. but to what end? So that you personally won't be affected? This is a morally bankrupt stance to take.
That is in fact what I want. You're projecting your preference for the bubble to continue onto me, when I have never said such a thing. Only from the ashes of the old can we build the new.
> I think AI is useful, and AI being cheaper for the masses is good for them rather than the value being captured by a few companies.
I believe this is naive. If it's controlled by a few companies, they can end the cheapness at any point. I believe that is part of the plan, once they've made the masses sufficiently dependent.
You believe that AI is not useful and you want the bubble to burst.
I believe it is useful. I don't think its worth arguing at this point with anyone who thinks AI is a net bad to humanity, even when it is controlled by public. Its kinda extremist.
I literally said the opposite? You keep arguing with some phantom you've made up.
> thinks AI is a net bad to humanity, even when it is controlled by public
Also not what was said. My point is precisely around it being a tool for private power concentration. If it was publicly owned, that would not be the case, obviously.
> I don't think its worth arguing at this point
I don't think you were ever arguing, you have not addressed any of my points.
I'll go even further and tell you my read of what is going on: You cannot engage with the actual arguments, because doing so would force you to confront the inconsistency of your ideology. To step out of ideology hurts, so you instead address a strawman you've built and then escape the conversation when that becomes untenable.
You've deliberately been vague about your position on the utility of AI.
So I'll assume you think AI is useful for society and a net good if it were not controlled by capitalists. Again I wouldn't have to assume this if you had been straight about it.
If you think AI is useful if it were publicly owned, surely high energy costs and hardware costs are necessary if the public were to use it?
Whats the point of talking about hardware and energy costs when that's something the public needs to bear that cost in one way or another, whether by voting for it or not?
If the bubble bursts and the models become commoditised, the margins on hosting drop to near zero and people can access models at much cheaper costs and there's no central authority (like spooky oligarchy who control power) that you are scared of.
So why is hardware and energy costs even a problem if the public consents to using AI with their own revealed preference?
> You've deliberately been vague about your position on the utility of AI.
Because it's not the interesting question. The interesting question is what it does as long as it's controlled by a handful of people.
AI's usefulness depends on that question, it cannot be answered in isolation. A lighter is useful in the hands of somebody starting a camp fire to warm themselves and their community, but dangerous in the hands of an arsonist trying to burn down said community.
> If you think AI is useful if it were publicly owned, surely high energy costs and hardware costs are necessary if the public were to use it?
Public ownership and public usage are not the same thing. Again, ownership and power dynamics are the point (that you still haven't addressed!)
> Whats the point of talking about hardware and energy costs when that's something the public needs to bear that cost in one way or another, whether by voting for it or not?
How have we established that the public needs to bear it one way or another? If AI was public, we could have usage restrictions based on societal merit, as an example. Not every AI use is ... useful. All of them are costly to society.
> So why is hardware and energy costs even a problem if the public consents to using AI with their own revealed preference?
What are you talking about? The grandma's electric heating is getting more expensive so some jerkoff can create deepfake porn or so capitalists can fire workers. They're not the same people.
> If AI was public, we could have usage restrictions based on societal merit, as an example. Not every AI use is ... useful. All of them are costly to society.
Ok so you want people to vote for public AI usage. Like if I want to use AI for my own research, I need to first vote for it, get it approved for its "social merit".
It's the only way to avoid the catastrophic power concentration issues and resource allocation issues you still haven't addressed.
It's not exactly a new concept either. All sorts of things need approval (usually ones with high resource cost and externalities, see a pattern there?), be it buildings, flight plans, oil drilling, power lines or any number of other examples.
As with some other stuff (the de minimis exception for customs comes to mind), we can of course talk about exceptions for smaller-scale personal use, but large industrial use with heavy pollution and energy use implications should be treated like any other heavy industry.
You still haven't made any case against my structural argument. You still haven't provided any evidence in favor of your viewpoint. You keep digging into hypotheticals about alternatives (at least you've stopped with the straight up strawmanning, I guess), but you've never bothered to justify the implications of the status quo you support.
Kind of funny considering how you started out by accusing others of fallacies and being morally bankrupt, mindless extremists. Maybe the shoe's on the other foot?
Reading the comments here makes me feel this is such an interesting blip of time we're living through right now.
On the one hand it's treated like an open secret that we're nearing the end of an AI bubble that even the IMF is flagging as a risk as reality doesn't live up to the hype.
On the other hand the capability of systems in just the last two months has improved so much that it feels like the average hype undersells what's happening right now but that knowledge is extremely unequally distributed as the greatest capabilities are most familiar to software engineers that have fully embraced agentic coding, which even earlier this year was still somewhat questionable but now is too highly productive to avoid.
The question I have is if everyone is this bearish on AI stocks are they not more likely than not to move against consensus much higher?
> On the other hand the capability of systems in just the last two months has improved so much that it feels like the average hype undersells what's happening right now but that knowledge is extremely unequally distributed as the greatest capabilities are most familiar to software engineers that have fully embraced agentic coding, which even earlier this year was still somewhat questionable but now is too highly productive to avoid.
I've been hearing this for years now, and it's not my experience so my conclusion is that it's only true for a subset of software engineering that is overly represented on internet forums like here. I imagine if you work in the tech industry on a software product the latest capabilities are game changing. But for the majority like me who work in non tech industries where code was never the hard part of projects to begin with there has been little evolution in the job in one year. Sonnet 4.5 one year ago could already write the code. Most of the job is working out with other humans on the client/business side. To progress beyond that not only would AI need to have a human level cognitive/world model to understand all the dynamics, but also have access to all the information that only people have in their head, as what is written and in specs is only a tiny part of how a business works day to day in reality.
Not saying there is no disruption. Just giving the perspective that in software outside the tech industry there is no vertical acceleration happening, much less in day to day operations outside of software. The tech industry is a big part of the economy, but not all.
That makes sense it’s plausible models stay trapped in verifiable domains and even if software engineers can be 10x more productive today the overall economic impact of that is meaningful but perhaps too modest for the capital investment to be recouped in the time frame required. It does also seem plausible to me, though, that the state reached in software in the last two months gets hit in many more domains in the next couple of years and when zooming out the end consumer and business spend just keeps growing rapidly as more and more capabilities emerge and mature across sectors and professions.
Haha that is true and probably is some of the explanation for the gap, and there's still a long ways to go so who knows if the capturable value of the capability increase grows fast enough to justify the capital investment. I think it's important to take greater note of who is joining the choir this time, though, as Karpathy, Lütke, DHH, Sweeney, Persson, etc. are publicly communicating something different than a couple months ago [1][2][3]. While the people previously responsible were far more likely to be "talking their book" so to speak as producers of AI or act like religious converts, the new crop are consumers of AI for genuine productivity gains that view the technology with a more secular lens rooted in productivity growth.
If you're referring to DHH, nothing indicates he is a nazi. At most, he holds some fairly normal anti immigration and right wing views. Diluting the term nazi like that is bad.
You could probably say the same about most of the Nazi-supporting German public as well as most of the loyal party members, prior to/during WW2. Less extreme examples can still fit the description without diluting the term. Having views similar to Nazis is becoming normalized more than the term Nazi is being diluted.
People drink the koolaid incredibly hard about AI capabilities, yet none of it has materially resulted in anything beneficial for your average person. Software is at minimum still buggy and at worst you see larger and more frequent bugs (see: the recent gaffes by Microsoft). Anything involving a computer is now far more expensive, AI has become highly intrusive into our daily lives as it's foisted upon us no matter how many times we say no. There's so many ways it's made our lives a little bit worse that it's no wonder why people outside of the tech bubble despise it.
> On the one hand it's treated like an open secret that we're nearing the end of an AI bubble that even the IMF is flagging as a risk as reality doesn't live up to the hype.
As you pointed out, reality does live up to the hype.
Pessimists argue that the large investments into data centers are unreasonably high and that this debt is going to bankrupt AI labs.
However, as of yet the demand is there and the data centers are fully utilized.
We can see that currently OpenAI is struggling to serve their inference.
Inference speed on 6.1 Sol went to like 20-30 tokens per second and it frequently fails with the "model at capacity" error.
It's imo a bit too early to state with any confidence that there will be a crash. It seems dependent on the development of AI capabilities, resulting demand, and compute requirements.
Imagine you won an all expenses paid holiday to Bali. You get to stay at a resort for 6 months, with two rooms for you and your children. You are given a rental car for transport. You go to the beach every day. You eat restaurant food 3 times a day. Your rooms are made up by servants every day.
It feels so good, that you forget that this is just a holiday. You start thinking "this is the future of living".
But at the same time it is absolutely real, you're experiencing it, other people at the resort are experiencing it (though maybe not at a similar level as you).
You love it so much you start planning with your partner about which school the kids should go to. You start planning your whole life in this paradise.
threat that is only in the minds of the investors really, Kahneman at his best. Psychology driven 100%. Profits are there and humongous but narrative is decoupled with fear.
So I'm reading the actual IMF reports itself [1] and I come across this on the first page:
"First, although core firms in the AI ecosystem have been profitable so far—easing pressure on AI-related equity valuations—weaker or less durable investment and profitability in AI could trigger a correction"
Is that correct - that the AI firms have been "profitable"?
[1] https://www.imf.org/-/media/files/publications/gfsr/2026/oct...
They are talking about Meta, AWS, Microsoft, Google, Nvidia and so on.
Although Anthropic supposedly has a good margin on each token, so the inference is substantially less expensive than what they charge customers. Training is eating into that.
> Although Anthropic supposedly has a good margin on each token
I am still to see any evidence of this being true.
The actual data at hand suggests otherwise, with Anthropic's losses increasing with more usage, and they relying on shenanigans to give an appearance of being healthy. They also need to raise money like crazy and seem desperate to offload this hot potato through an IPO. Not really the sort of thing that screams "sustainable" to me.
Such as having a "profitable" quarter on one of their funny metrics when they were not paying for compute, or leaking completely obscure ARR, "EBTIT", etc.
> The actual data at hand suggests otherwise,
What data? To my knowledge the only data point that exists is a 2025 document, which is completely irrelevant considering how massively they hiked the API token price.
To my knowledge, the subscriptions are the only part that may still be profiting from vc subventions. And if not, then their API margins are gigantic
> What data?
- The suspiciously high amount of money they have to keep raising at a rapid pace
- The fact that they have to play around with silly metrics and leak bullshit to pretend they are healthy
- Their desperation for an IPO.
I feel kind of silly having to repeat myself, it was all there in my previous reply of you bothered to read. I point at that whenever people spread this bullshit of Anthropic being "profitable" especially when it comes to subscriptions. Everything points at it being ruinously unprofitable actually.
Profitability only matters if we consider that they are grifters willing to cash in as quickly as possible before this jenga tower falls apart. It matters less if the interpretation is that the fucking weirdos at Anthropic are death cultists that want to conjure their death god into existence.
but none of those are data points which actually meaningfully tell what we were discussing? They're not even data points, theyre merely observations from the outside, not at all objective.
Anthropic can be profitable and massively overvalued, AND deeply into debt from running too fast and loose with their money in previous years.
That situation would have them behave exactly the same: hurrying to get the IPO done while the stocks are still skyhigh to get over the debt burden and then buckle up for the profitability to weather them through coming correction...
> Anthropic supposedly has a good margin on each token
Genuine question: why are people still saying the plans are heavily subsidised?
Because it's kind of true, but also mostly wrong. People conflate two things: "Subsidized" subscriptions as a monetization model, and (unsustainable) subsidization through outside investors.
When you max out your Claude Max plan every billing period, are you consuming more tokens than you could realistically pay for with the same money spent on API inference? Definitely! This works because many Claude Max subscribers don't use it to the max (and you probably aren't using it to the max every single time either), so overall subscriptions might still be profitable. Technically, the subscribers subsidize each other, but this is entirely sustainable (see any other subscription model that is based on its consumers not maximizing their usage, e.g. gyms).
What people mean (or want to say) when they say inference is subsidized is that it is subsidized in an unsustainable way (e.g. with investments that will not last forever - if this is true at scale it should worry us!). What happens now is that they take the fact that an individual subscription used to the max by itself is a financial drain for the company offering it, attach the "subsidized" term to it, and infer from that a bunch of stuff that is usually associated with the term (such as unsustainability) but doesn't really apply here.
The gym model only works because people don't want to go to the gym. It doesn't work for cinemas, for example.
whether people want to use the service or not is irrelevant, the only thing that matters is whether or not (and how much) they actually do use it. Also there are subscription/flat rate based cinema services that have been in business for almost two decades now, so it doesn't seem like that wouldn't work.
Because the subscriptions grant you much more usage than what you get in tokens for the same money via the API. That being said, they probably don't have ruinous losses per subscription, if any. Don't forget that many users are casual and do not use Claude Code, so a casual user and a power user level out
Anthropic has been fairly strict on token limits for some plans, with laid overage.
OpenAI released an AI video app to the masses for free, made a bunch of promises, and then just shut it down.
So there’s a growing body of evidence that while neither of these are in great financial shape, Anthropic looks to be the more disciplined of the two.
When you say that something costs $10k worth of tokens at API pricing, it sounds impressive.
Some people then assume that this price means the subscriptions are deeply unprofitable at least for users that make heavy use of their usage limit.
If you look at estimates based on serving Kimi K3, margins at API pricing must be well north of 95%.
And that's assuming frontier labs don't serve inference more efficiently than Kimi K3.
I believe most people dramatically overestimate the inference costs.
It doesn't help that in many discussions about profitability, people mix many different metrics, including other costs like R&D.
Because they are, I pay anthropic 200 a month, if I were paying by the token like an enterprise it would be close to 11 grand a month. Yes I use 100 percent of my allocation on every window.
That’s only a subsidy if you argue about lost revenue (like the anti-piracy arguments). But somehow I doubt you’d be willing/able to pay even close to 11 grand a month.
The only real comparison point that matters for the argument of subsidization is how much money it cost costs Anthropic to serve you those tokens.
I guess they mean the hyperscalers rather than OpenAI/Anthropic.
Firms profit from others' investments.
I think they refer to that piece of news
https://news.ycombinator.com/item?id=49698936
I haven't digged myself but many people have argued that Anthropic used accounting tricks to appear profitable to investors, but they removed some costs from their calculation.
I think the message to investors is that operating the model is profitable. Like the income they get from paying customers pays for the infrastructure of running the models.
Might be referring to infrastructure.
Maybe - though they refer to hyperscalers explicitly in other parts of the report, and then there is this:
"Hyperscalers have strong earnings from their core businesses. Yet, free cash flows are rapidly declining as hyperscalers try to sustain large AI investment needs, leaving debt financing increasingly necessary"
yes totally profitable. It's all "omg it could happen" narrative.
surely the working man won't pay the price for this one too, like every other crash
Working man? You can replace them with agents!
Then you somehow still need to get rid of the working man, though...
Oh man I wish a publication would go with that future headline. "If AI agents were supposed to replace working people, will they be footing the bailout bill as well?" hmm maybe it's a bit too long for a headline.
Honestly a crash upward is more devastating for the working man
The debt lasts longer than GPUs remain profitable, right? GPUs loose value most quicker than most investors believe. In best case they last like 2-3 years under heavy stress, so I think about all those data centers and what will it mean to replace those cards burned down - you cannot even replace one with newer one - all 8 GPUs should be the same type. Thinking about future costs with all the debt.
but they don’t, the price of an h100 per hour has been going up across the year because you can squeeze more utility out of the computations
The only business model left: grow so big you need a bailout.
Who would have guessed ?
https://archive.ph/dAI6p
With an AI bubble implosion now all but inevitable, the shift of the key players behind the scenes has focused on making this seem too big to fail and thus needing a bailout.
The “it’s dangerous, regulate us please” angle as an out for explaining terrible financial performance doesn’t seem to be working. The final Hail Mary pass is the “if we fail you fail” play.
However all data indicates the risk here is fairly concentrated in private markets. It will be painful but it’s unlikely to take the whole economy out with it. Some VCs are in for a world of hurt and private credit markets will lose their shirts, but the average person would see limited direct impact.
With everything going on in the datacenter space a bailout for AI seems incredibly unlikely. The general sentiment amongst the masses seems to be “let it burn.”
Yes, but you are partially wrong IMHO.
The problem is that private credit agencies also asked for loans from banks and therefore they are exposed to this as well. I believe the current estimate is that 11% of outstanding loans from the big 4 US banks were issues to private credit firms which then very likely lent themselves to the AI industry complex.
Additionally, by the Fed's own admission, they will never let another "Great Depression" happen again. So they intervened in 2008 and they will intervene again by buying up those worthless assets. And the average person will see their savings real value plummet without even mentioning that a lot of private credits come from pension funds...
Finally, if private credit really goes belly up, private investors will likely lose a good portion of their purchasing power, meaning that the economy will at best correct and at worst crash only because of that. And this effect will be in addition to the AI economy grinding to a halt which will already hit the worldwide economy HARD (also IMHO).
A reckoning is coming. I believe the "let it burn" is not doable as it will take a few banks and the economy with it. A more measured slow bleed with the Fed printing money to buy these junk assets will be more likely.
You’re not wrong but this is not 2008 2.0 either.
Some banks will suffer losses but there isn’t data to say the banks will fail en mass. This is bad, but not 2008 bad in so far as the stability of the financial system.
The problem is that nobody knows how really bad this is. For instance, if I, a private individual, asked for a loan to put money in private credit there is no way to link these two actions.
The 11% I mentioned above is the verifiable exposure for banks. It could be a much much higher number and at a time where delinquencies on all sorts of debts are historically high that for sure doesn't bode well.
The money printing that the Fed would do to fix this would be gigantic (several Trillion dollars basically) and on the same scale as the covid intervention which sent inflation to 9+%. And we are already waiting for the oil shock to spread to everything else.
Yeah, the financial system might survive, but the pensions, savings, ... will be fucked up.
You underestimate the corruption of the White House. I've no doubt there will be a massive bailout of the guys who sat up there on the dais during his inauguration. Zuck, Musk, Pichar... He's hosted Altman and Amodei recently, too, right?
Only if Trump wins midterm.
Trump is a symptom of the disease.
You can make billions overnight if you think it is inevitable.
As many have commented, not really. Shorting is about exact timing as much as knowing the inevitable outcome. Get the timing just a bit wrong and you can be right and broke. The Big Short portrays that quite well.
Then you have to clarify what "inevitable" means.
If it goes up another 200% then down 30%, would you call that implosion?
How? I would like to! And don't say shorting because, well inevitable doesn't mean imminent.
Credit default swaps. Basically an insurance that pays if a stock/asset value plummets. The problem is that you have to pay for this insurance coverage, and currently the prices are really really high [1], meaning that the market is expecting a failure soon, especially with interest rates/bonds yields going up.
[1] https://www.tradingview.com/news/seekingalpha:bce1fc46a094b:...
“Markets can remain irrational longer than you can remain solvent."
Why do you think all of them want to be in the S&P 500?
> Some VCs are in for a world of hurt and private credit markets will lose their shirts,
They will walk out only rich rather then super rich. They will talk about it as if they were poor, but nope, they will still have toms of power and money.
Some yes. But it’s no secret that there are a lot of folks way in over their head at the moment with no clear way out.
Too big to fail, by intention
If AI share crashes, that’s actually good for the public because all the value is now captured by them instead of companies. This is because a crash can only happen if prices go down (which is good) or Chinese models become as good as frontier (which is also good).
The question is, if this happens then how do you incentivise more R and D? Which investor would invest without the returns?
Edit: not surprisingly people don’t understand the point.
Would you rather NVIDIA and OpenAI get rich and capture the value or the public? Such a strange thing to see people rooting for the big companies.
Would a correction unlock funds for other interesting and important ideas, even in AI space (not just LLM). I guess only time will tell.
In the short term things tend to freeze up as those with funds to invest lock down the hatches and weather the storm. But fairly quickly after yes funds start flowing elsewhere as money chases the next big thing.
AI is here to stay, but the market needs a big forest fire to wipe most things out. A few strong plants survive to build a better future.
It’s a cycle that keeps repeating, and we’re overdue for a big forest fire in tech.
The real question is if the money leaves AI stocks where does it go.
Well, a lot of that money is leveraged, so it will just disappear.
Cash because no AI bubble means the market is cooked for a decade or so
I agree. I've been telling people that AI is the last great bubble. Every other place where printed money can go has already been tried.
> This is because a crash can only happen if prices go down (which is good) or Chinese models become as good as frontier (which is also good).
No... This is not accurate at all. AI valuations are based on high future growth. Prices could stay the same but growth slows and that could cause a crash by itself.
Right. AI companies have P/E ratios rarely seen for very large companies. Except for TSLA, which has a P/E ratio of 383, unheard of at that scale. No other large-cap company is much over 40.
TSLA should have crashed by now, but the hype just keeps on coming. Eventually reality will set in, but the market can remain irrational for longer than you can remain solvent. Big AI may be like that.
Investors like a story that lets their imagination run wild. TSLA offers that. AI / NVIDIA has that 10-fold. It’s “man’s last invention” - you couldn’t come up with a better hook to raise capital.
Short of an emergency rate hike to 10%, nothing stops this bubble.
NVIDIA would have a market cap of 50 trillion if it matched the TSLA pe ...
Anthropic IPO will be a better example.
NVIDIA is still a widget business.
It's just circular financing - we give you GPUs, you give us your shares of a promised future $ 2,000,000,000,000 valuation. Despite they are unprofitable today and tokens are getting cheaper and cheaper.
False. They are based on future revenues that companies can capture. If the companies can’t capture the value and there’s no moat, then the public stands to gain from it.
Why is this controversial?
I think what you don't fully understand is that if these companies crashed and stopped getting in new investor money then lowering prices would not be the answer because they would not actually have enough revenue to even pay their own bills. Right now the AI costs are partly being subsidized just like when Uber first came out and rides only cost like $6 when it might now be $40. While what you are saying might come true it doesn't mean that it's directly related to what we are talking about, these companies crashing is not going to magically just result in lower prices. Lower prices are likely but would be due to other factors. Also, lowering prices is not going to cause a crash by itself. Just look how cheap and capable GPT-6 Luna is compared to GPT-3.5. That didn't cause OpenAI to crash, did it?
They can't properly meet the current demand. A crash will primarily hit NVIDIA, AMD and other AI chip makers as they have oversold already and the cards are sitting in storage waiting for the needed data centers and the power needed to run them.
Its like a gold rush where half of the picks are sitting in storage because there are not enough miners to wield them.
In what world is a global financial crash good for the public? At least short-term this will be painful for everyone.
At some point out becomes a case of least worst I think.
If a crash unlocks capital and eases the compute price squeeze, it might be worth the pain.
Every crisis is an opportunity. Maybe we could build something where we don't have ridiculous power and wealth concentration at the expense of the public - to the point where the world's richest man literally took over the US government to cut aid to the world's poorest.
It's already painful now.
Weird that you're downvoted. It would suck very hard indeed if your pension portfolio matured the day after a crash.
Why do you discount the immense gains due to AI that the public can use cheaply?
There has to be a name for this fallacy.
Because if you can no longer pay your rent from your pension portfolio, AI capabilities matter much less to you.
Most people who are currently enjoying the benefits generative AI don't even have pensions, let alone ones that are heavy into AI stocks. Wtf are you even talking about?
GP jen729w was talking about that particular scenario, though. I have never claimed that would be an issue for everyone. But it will be an issue for a lot of people, if the crash is hard and recovery is slow.
What does the public get from it, materially? More expensive energy and hardware.
This comment shows the exact problem with the incoherent ideology prevalent in some bubbles.
You fundamentally believe that AI is useless and that it wastes energy and hardware. Shouldn't you then want the bubble to burst even faster? If not, you are just delaying the bubble.. but to what end? So that you personally won't be affected? This is a morally bankrupt stance to take.
My ideology is at least more coherent in this context. I think AI is useful, and AI being cheaper for the masses is good for them rather than the value being captured by a few companies.
GP here. I use AI all day. I love Claude.
But it's objectively true that if the market crashes tomorrow, lots of normal people lose money. This hits extra hard – which happened back in 2007/8 – if that's the day your life-long savings are converted from stocks to cash.
Are you arguing that because those pensioners can now get a frontier model cheaper than they can today that this is a net benefit?
No, I disagree. This is not what happened in 2008 which was fundamentally different because it was the people that held the bad debts.
If market crashes tomorrow due to commoditised models, it is true that those people who bought these stocks would lose money. But its honestly not a big part of it - if you run the numbers, pensioners will lose 10% of their money.
This is a transfer of wealth from people who have exposure to these stocks vs everyone else. Do you disagree with this transfer of wealth? It comes down to that.
> This comment shows the exact problem with the incoherent ideology prevalent in some bubbles.
You're shadowboxing a strawman you've built, my friend.
> You fundamentally believe that AI is useless and that it wastes energy and hardware.
That is not what I believe. I believe that AI is useful at concentrating power in the oligarchy and that increasing energy and hardware cost is a nice side effect that further helps that goal.
> Shouldn't you then want the bubble to burst even faster? If not, you are just delaying the bubble.. but to what end? So that you personally won't be affected? This is a morally bankrupt stance to take.
That is in fact what I want. You're projecting your preference for the bubble to continue onto me, when I have never said such a thing. Only from the ashes of the old can we build the new.
> I think AI is useful, and AI being cheaper for the masses is good for them rather than the value being captured by a few companies.
I believe this is naive. If it's controlled by a few companies, they can end the cheapness at any point. I believe that is part of the plan, once they've made the masses sufficiently dependent.
Which is why the most important thing is to make sure tokens are a fungible commodity. That is the only thing that counts.
You believe that AI is not useful and you want the bubble to burst.
I believe it is useful. I don't think its worth arguing at this point with anyone who thinks AI is a net bad to humanity, even when it is controlled by public. Its kinda extremist.
> You believe that AI is not useful
I literally said the opposite? You keep arguing with some phantom you've made up.
> thinks AI is a net bad to humanity, even when it is controlled by public
Also not what was said. My point is precisely around it being a tool for private power concentration. If it was publicly owned, that would not be the case, obviously.
> I don't think its worth arguing at this point
I don't think you were ever arguing, you have not addressed any of my points.
I'll go even further and tell you my read of what is going on: You cannot engage with the actual arguments, because doing so would force you to confront the inconsistency of your ideology. To step out of ideology hurts, so you instead address a strawman you've built and then escape the conversation when that becomes untenable.
You've deliberately been vague about your position on the utility of AI.
So I'll assume you think AI is useful for society and a net good if it were not controlled by capitalists. Again I wouldn't have to assume this if you had been straight about it.
If you think AI is useful if it were publicly owned, surely high energy costs and hardware costs are necessary if the public were to use it?
Whats the point of talking about hardware and energy costs when that's something the public needs to bear that cost in one way or another, whether by voting for it or not?
If the bubble bursts and the models become commoditised, the margins on hosting drop to near zero and people can access models at much cheaper costs and there's no central authority (like spooky oligarchy who control power) that you are scared of.
So why is hardware and energy costs even a problem if the public consents to using AI with their own revealed preference?
> You've deliberately been vague about your position on the utility of AI.
Because it's not the interesting question. The interesting question is what it does as long as it's controlled by a handful of people.
AI's usefulness depends on that question, it cannot be answered in isolation. A lighter is useful in the hands of somebody starting a camp fire to warm themselves and their community, but dangerous in the hands of an arsonist trying to burn down said community.
> If you think AI is useful if it were publicly owned, surely high energy costs and hardware costs are necessary if the public were to use it?
Public ownership and public usage are not the same thing. Again, ownership and power dynamics are the point (that you still haven't addressed!)
> Whats the point of talking about hardware and energy costs when that's something the public needs to bear that cost in one way or another, whether by voting for it or not?
How have we established that the public needs to bear it one way or another? If AI was public, we could have usage restrictions based on societal merit, as an example. Not every AI use is ... useful. All of them are costly to society.
> So why is hardware and energy costs even a problem if the public consents to using AI with their own revealed preference?
What are you talking about? The grandma's electric heating is getting more expensive so some jerkoff can create deepfake porn or so capitalists can fire workers. They're not the same people.
> If AI was public, we could have usage restrictions based on societal merit, as an example. Not every AI use is ... useful. All of them are costly to society.
Ok so you want people to vote for public AI usage. Like if I want to use AI for my own research, I need to first vote for it, get it approved for its "social merit".
Is this the kind of utopia you are arguing for?
Indeed, now you're getting it.
It's the only way to avoid the catastrophic power concentration issues and resource allocation issues you still haven't addressed.
It's not exactly a new concept either. All sorts of things need approval (usually ones with high resource cost and externalities, see a pattern there?), be it buildings, flight plans, oil drilling, power lines or any number of other examples.
As with some other stuff (the de minimis exception for customs comes to mind), we can of course talk about exceptions for smaller-scale personal use, but large industrial use with heavy pollution and energy use implications should be treated like any other heavy industry.
You still haven't made any case against my structural argument. You still haven't provided any evidence in favor of your viewpoint. You keep digging into hypotheticals about alternatives (at least you've stopped with the straight up strawmanning, I guess), but you've never bothered to justify the implications of the status quo you support.
Kind of funny considering how you started out by accusing others of fallacies and being morally bankrupt, mindless extremists. Maybe the shoe's on the other foot?
> Kind of funny considering how you started out by accusing others of fallacies and being mindless extremists. Maybe the shoe's on the other foot?
Your utopia is me asking some one approval before I use my LLM for my own work.
I'll let others decide whether this is extreme or not.
You are still not defending your view or disproving any part of the structural argument. The only interpretation left is that you cannot.
Wouldn't be so bad if shitty companies were allowed to fail.
Reading the comments here makes me feel this is such an interesting blip of time we're living through right now.
On the one hand it's treated like an open secret that we're nearing the end of an AI bubble that even the IMF is flagging as a risk as reality doesn't live up to the hype.
On the other hand the capability of systems in just the last two months has improved so much that it feels like the average hype undersells what's happening right now but that knowledge is extremely unequally distributed as the greatest capabilities are most familiar to software engineers that have fully embraced agentic coding, which even earlier this year was still somewhat questionable but now is too highly productive to avoid.
The question I have is if everyone is this bearish on AI stocks are they not more likely than not to move against consensus much higher?
> On the other hand the capability of systems in just the last two months has improved so much that it feels like the average hype undersells what's happening right now but that knowledge is extremely unequally distributed as the greatest capabilities are most familiar to software engineers that have fully embraced agentic coding, which even earlier this year was still somewhat questionable but now is too highly productive to avoid.
I've been hearing this for years now, and it's not my experience so my conclusion is that it's only true for a subset of software engineering that is overly represented on internet forums like here. I imagine if you work in the tech industry on a software product the latest capabilities are game changing. But for the majority like me who work in non tech industries where code was never the hard part of projects to begin with there has been little evolution in the job in one year. Sonnet 4.5 one year ago could already write the code. Most of the job is working out with other humans on the client/business side. To progress beyond that not only would AI need to have a human level cognitive/world model to understand all the dynamics, but also have access to all the information that only people have in their head, as what is written and in specs is only a tiny part of how a business works day to day in reality.
Not saying there is no disruption. Just giving the perspective that in software outside the tech industry there is no vertical acceleration happening, much less in day to day operations outside of software. The tech industry is a big part of the economy, but not all.
That makes sense it’s plausible models stay trapped in verifiable domains and even if software engineers can be 10x more productive today the overall economic impact of that is meaningful but perhaps too modest for the capital investment to be recouped in the time frame required. It does also seem plausible to me, though, that the state reached in software in the last two months gets hit in many more domains in the next couple of years and when zooming out the end consumer and business spend just keeps growing rapidly as more and more capabilities emerge and mature across sectors and professions.
> the capability of systems in just the last two months has improved so much
People have been saying this every month since 2022 lol
Haha that is true and probably is some of the explanation for the gap, and there's still a long ways to go so who knows if the capturable value of the capability increase grows fast enough to justify the capital investment. I think it's important to take greater note of who is joining the choir this time, though, as Karpathy, Lütke, DHH, Sweeney, Persson, etc. are publicly communicating something different than a couple months ago [1][2][3]. While the people previously responsible were far more likely to be "talking their book" so to speak as producers of AI or act like religious converts, the new crop are consumers of AI for genuine productivity gains that view the technology with a more secular lens rooted in productivity growth.
[1] https://x.com/karpathy/status/2109361546505966046?s=20
[2] https://www.businessinsider.com/minecraft-creator-markus-per...
[3] https://x.com/dhh/status/2102936073642869121
Yeah. The worst people in the world are all on board. Add Trump, Musk, Hegseth etc to the list.
Who do you think are the best (most technically informed and productive) people in the technology industry?
> Who do you think are the best (most technically informed and productive)
That's quite the framing. I'd say the trajectory count way more than how productive you are to follow it in assessing someone is good.
I don't know but it's certainly not some washed up nazi Ruby developer?
If you're referring to DHH, nothing indicates he is a nazi. At most, he holds some fairly normal anti immigration and right wing views. Diluting the term nazi like that is bad.
You could probably say the same about most of the Nazi-supporting German public as well as most of the loyal party members, prior to/during WW2. Less extreme examples can still fit the description without diluting the term. Having views similar to Nazis is becoming normalized more than the term Nazi is being diluted.
People drink the koolaid incredibly hard about AI capabilities, yet none of it has materially resulted in anything beneficial for your average person. Software is at minimum still buggy and at worst you see larger and more frequent bugs (see: the recent gaffes by Microsoft). Anything involving a computer is now far more expensive, AI has become highly intrusive into our daily lives as it's foisted upon us no matter how many times we say no. There's so many ways it's made our lives a little bit worse that it's no wonder why people outside of the tech bubble despise it.
From the creators of: it’s quicker to do this online to it’s quicker to do this in person.
> On the one hand it's treated like an open secret that we're nearing the end of an AI bubble that even the IMF is flagging as a risk as reality doesn't live up to the hype.
As you pointed out, reality does live up to the hype.
Pessimists argue that the large investments into data centers are unreasonably high and that this debt is going to bankrupt AI labs.
However, as of yet the demand is there and the data centers are fully utilized.
We can see that currently OpenAI is struggling to serve their inference.
Inference speed on 6.1 Sol went to like 20-30 tokens per second and it frequently fails with the "model at capacity" error.
It's imo a bit too early to state with any confidence that there will be a crash. It seems dependent on the development of AI capabilities, resulting demand, and compute requirements.
Imagine you won an all expenses paid holiday to Bali. You get to stay at a resort for 6 months, with two rooms for you and your children. You are given a rental car for transport. You go to the beach every day. You eat restaurant food 3 times a day. Your rooms are made up by servants every day.
It feels so good, that you forget that this is just a holiday. You start thinking "this is the future of living".
But at the same time it is absolutely real, you're experiencing it, other people at the resort are experiencing it (though maybe not at a similar level as you).
You love it so much you start planning with your partner about which school the kids should go to. You start planning your whole life in this paradise.
Obligatory: "Markets can remain irrational longer than you can remain solvent".
[dead]
threat that is only in the minds of the investors really, Kahneman at his best. Psychology driven 100%. Profits are there and humongous but narrative is decoupled with fear.
Pretty much everyone knows this is going to crash and make life difficult for everyday people. It’s a cash grab by desperate liars.
As I predicted, it shall crash right after midterm election.
why "after" and not "before"?
Politicians tend to prop things up before elections.
aaaaahhahahahahahaha!!!
nice place you got here, be a shame if something happened to it....
[flagged]
AI build out goes on until the universe is filled with AI robots reporting back to us