46 条评论
- 0xbadcafebee 13小时前AMD, Alibaba should be on there too. AMD is making good money on AI, with R&D at less than half the AI revenue. Whereas Alibaba's weird financials show it's kinda-sorta-protifable?
I just wanna know how the OpenAI/Anthropic shell game works long-term. So both companies made equity deals with infrastructure providers; OpenAI on Azure, Anthropic on AWS, GCloud, and Colossus. They get a loan of compute credits and then pay for the compute with the credits. So the PaaS are effectively giving them free compute, then book it as revenue; and the AI provider lets them do inference and books that as revenue. So, it's like both types of company have a buffet, and let each other eat there for free. But somebody has to actually buy the pasta salad, with real dollars. Afaict, those real dollars are.... the cash reserves of the PaaS.
How long are they going to eat into that cash? Microsoft and AWS don't really have their own models, whereas Google and SpaceX do. And while Google has tons of cash, SpaceX is perpetually looking for cash. So the only player here that can actually afford to keep doing this, or leave the game entirely, is Google.
- 7thpower 12小时前With this line of thinking, nobody would have ever built refineries, or fabs, or clouds.
The frontier labs have fantastic margin on inference. You do not understand how fantastic. And they have license to change inputs at will based on profitability.
They are not only innovating on models and tooling, they are innovating on cogs (I wrote this btw, and I’m not going to stop writing this way because Claude discovered it’s brilliant).
Speaking of models, the cost of training is not scaling nearly as fast as demand for inference. Training used to be the biggest cost by far, now it’s not.
So margin is increasing, and guess what else is happening? Customers are finding value. And the customers that are finding value are also the ones who happen to have huge enterprise budgets.
And while this is happening, so is implicit collusion (and lock in, and hype, and all that). And so prices are going up.
They’re going to be just fine man, there is no inference bubble.
They can modulate supply. It’s all going to be fine. You should invest.
- sfifs 10小时前> The frontier labs have fantastic margin on inference. You do not understand how fantastic. And they have license to change inputs at will based on profitability.
This. The gross margin on inference is at least 95% if not higher - several open weight models on my tiny consumer DGX Spark easily replace the 15 dollars a day I was paying in tokens for Claw usage with a dollar a day electricity. You add data centre overhead and depreciation, the theoretical net margin will trend lower but depreciation is always far more aggressive than actual product degradation. The old NVIDIA GPU on a 9 year old second hand gaming PC I bought still serves up a small Gemma 4 variant quite reasonably.
- spwa4 3小时前To say nothing of the fact that they can just add "figure out how to change the answer to this question to benefit X" at the top of their system query.
It is baffling that any government lets either themselves or their local companies use these tools. Utterly baffling. The potential for total security compromise through these models is ... essentially 100%.
But ... it's slightly cheaper.
- gizzlon 9小时前> The frontier labs have fantastic margin on inference.
Source?
The OpenAi filing will be very interesting indeed.
("trust me bro" statements from sama et al does not count, since I don't trust them)
Edit:
The best argument I have seen look at the price of inference from smaller companies running open models. And assuming they are profitable-ish. Their prices are lower than the OpenAi and Anthropics best models, so maybe they do make money on inference (ignoring all other costs)
- harrall 12小时前Household names like Uber, Amazon, Blue Bottle Coffee and Fedex used the same playbook of burning investors’ cash for years and look where they’re now.
Everyone’s long term plan is hoping that they build out and survive long enough that, in the end, the market accepts them.
Even your local still-unprofitable restaurant is burning their grandparents’ inheritance money hoping that it works out.
But on the other hand, that’s what Theranos, WeWork, and Pets.com tried too.
- 0xbadcafebee 10小时前The old model works at tens of millions, not hundreds of billions. All the private capital is pretty tapped out, and banks aren't loaning (thank god). So they don't have investor money to burn (and when they do, they immediately burn it on new datacenters, which usually take years to build and aren't a certainty). That's why they made equity deals with hardware companies... it was the only way they could "afford" hardware. But someone has to pay for that hardware. And the person paying is... the hardware companies. They have a lot of cash, but not hundreds of billions of cash. Hence why Oracle pulled out, Nvidia scaled back its investment. Claude only doesn't suck right now because SpaceX literally loaned them a datacenter. So I'm saying... these companies will run out of cash, if they can't get paid back, sooner than later.
When OpenAI goes public it will initially get a tsunami of cash, but it'll also be open to new risk due to the different operating model and transparency. Anthropic might not make it to an S-1 (this year). Even if they got a $30B infusion of cash each, based on their current spending projections, it doesn't cover half of what they need just to break even. In the meantime the PaaS's are holding the bag (and shedding cash).
So where's it going to end? To me, all of this (combined with inflation, degrading of reserve currency, war in middle east) is spookily similar to the railroad panic of 1873. Over-investment in new technologies leveraging too much from the largest financial institutions resulting in prolonged economic crisis. Our only saving grace now are laws ensuring banks have to cover their end; if your money's FDIC/SIPC insured you're safe. But all the businesses and individuals who aren't safe are gonna take a bath, which'll have systemic ripples. Afaict, Google is the only player who can survive all that and come out with profitable AI. (But I'm sure I've missed something because it seems too obvious)
- renticulous 26分钟前> The old model works at tens of millions, not hundreds of billions.
The first computers were the size of buildings, now look where we are. I think same thing will happen to AI models. We will have a reasoning core installed on our phone connected to Google's Knowledge Graph or Ontology project via API. These companies just need to survive long enough to make themselves irreplaceable in the new ecosystem.
- zozbot234 9小时前They just have to incrementally raise the price of inference tokens and limit subscriptions to curtail existing demand (with much of it likely moving to slower and cheaper local models). Which, come to think of it, is exactly what seems to be happening right now.
> So they don't have investor money to burn (and when they do, they immediately burn it on new datacenters, which usually take years to build and aren't a certainty).
If AI models can get smarter and more practically useful via some combination of increased scale and more fine-tuned post-training on specific workloads (which is compute-heavy, even more than the usual kind of pre-training) these new datacenters are a fantastic investment.
- 0xbadcafebee 59分钟前They would have to raise the price of inference (and not just inference but actual contracts) 4x, within a year or so, for PaaS not to run out of cash. But nobody wants to pay that much money, and there are a swath of companies all over the world who will do it for much, much less. People will stick with them out of irrational fears (fear of the unknown, fear of missing out) until it gets too painful. And when people start leaving, what then? Anthropic's hope is that they can make a moat so high everyone is trapped. But it's actually not hard to replace Claude with a competitor.
They can get more efficient, but inference efficiency doesn't map linearly to cost efficiency. Firstly because software is a gas; if you give people more compute (for the same price), they immediately use it all up. But second, if you spend $50BN, you still have to make $50BN to break even. They could make inference cost $0.00000001, but that isn't going to cover their costs. That's what's driving their cost right now - they're trying to collect enough cash from people at the table to pay the bill, without the price scaring everyone out of the restaurant.
So they can't raise the price without scaring people off, and they can't lower the price and pay the bill.
- agentpenguin 37分钟前The subscriptions will drop if they increase the price and if they don't increase the price, they will run out of cash to operate.
It's simply not feasible.
- fatata123 5小时前[dead]
- wmf 13小时前There's a lot of revenue and outside investment coming in but the haters pretend it's all circular financing.
- nextaccountic 13小时前Outside investment is not a revenue, it's looking more like a pyramid scheme. It's yet more people putting their money in the scheme expecting a return
Either the actual revenue of paying customers ramp up or the bubble will pop at some point
I expect the paying customers will actually be companies buying ad, not people buying AI subscriptions
- crystal_revenge 13小时前It’s weird to me that people here suddenly seem to care about profitability for relatively early stage companies just because they’re “AI”.
I know a traditional SaaS company I worked for that IPO’d years ago and still has no signs that they can be profitable (and many others like it) and nobody seems particularly concerned.
- xbmcuser 13小时前These companies are spending more money than budgets of many countries enough to add 2+% to the US GDP so the amount of loss for if it comes all crashing down will be huge.
- fsckboy 13小时前if these companies go bankrupt, they will have spent (not lost) all their money, the large amounts of money that they got from investors. That money generated profits for other companies they bought stuff from, and income for their employees, and capital gains for other people if AIco acquired other companies.
the market cap of a company is computed by the current price of a company's shares, the last price paid; not all the shares of the company were bought at that price, the ones who got shares cheaper are showing paper profits, unrealized. Those who have already cashed out have money in their bank accounts that was transferred from people who wanted to get in. If the company goes bankrupt, their shares will be worthless, but the money they paid for them still remains in the accounts of people who sold their shares: the money was not lost even if some people lost money.
I'm not going to keep going through it but the reason it works to value things the way we do is that the values are comparable and they frequently work out, so snapshots of the economy and the participants are comparable. But "losses" are not like taking gold and feeding it into some deep fold in the earth where it will disappear into the molten middle of earth.
Stock valuations are "expectations for the future". Those expectations weren't money, they were lottery tickes where the lottery consisted of human creativity and human effort. People buying and selling share are moving real money around to trade the expectations. The money didn't go anywhere, it's still there, it's just that expectations for the future have been reduced. It all boils down to humans trading some of their time and potential on a bet that things work out. Some people's effort gets more rewarded than others. Not every team wins the world cup, but people like to play and like to watch.
- Retric 13小时前That’s an overly simplified model. AI companies spending results in infrastructure beyond the company such as manufacturing capacity, power lines, software systems, and even individual expertise.
If they fail then the negative impact ripples through the economy due to misallocation of resources.
- fsckboy 11小时前>infrastructure beyond the company
consider all the companies in a market and those that feed that market to be one virtual mega company, add up all the valuations and revenue streams, costs, etc and aggregate all the investors into one. Nothing changes about the picture I drew. We simplify models to make the real world understandable.
>negative impact ripples through the economy due to misallocation of resources
free or relatively free financial markets are the only way, the best way, the ne plus ultra of ways we know to allocate capital, we have no better way than for the owner of the capital and the reapers of the loss or reward to make a considered opinion that is risk "impedance" matched. By definition, the market does not "misallocate" capital, it optimally allocates it.
your theory is that we could somehow know the future, but that's a fallacy.
- Retric 11小时前> one virtual mega company
Free market efficiency is inherently tied to having multiple companies. Treating the entire economy as a single company gives nonsensical results because it fundamentally differs from what actually occurs. You might as well compare the economy to a game of tick tack toe, inherent complexity isn’t something you can simplify it has meaningful consequences.
Your ideas like many other ideas are simply wrong.
> could somehow know the future
Perfect accuracy isn’t the only possibility here, there’s levels of error.
Our system involves intermediaries between the actual owners of capital and the allocation of that capital who have very different incentives. When the worst possibility is missing a bonus there’s little difference between losing 10% of an investors money and 100%. That results in inefficiency through the misalignment of incentives.
That is actually true, and thus there’s no way to gloss over that truth without simply being wrong.
- dwb 9小时前Keep peddling that capitalist realism. “There is no alternative!” The market may not misallocate capital, by definition, but it very clearly and routinely misallocates resources. Let me guess: you’re doing relatively well for yourself?
- conradkay 13小时前That's one way to look at it, though it feels like you could say the same about the dot-com crash or 2008 which isn't too helpful. At the very least (extremely high-paying) jobs can be actually lost
- idiotsecant 13小时前This is way, way more neat and tidy than reality. When these stocks start to sink there is going to be an enormous evaporation of value from the overall market because people in riskier investments will get scared that other people will get scared. This will scare people with slightly safer investments, on up the line. Capital will dry up and velocity of money will drop. The market is not made by rational robots, it's run by barely sentient apes just minutes from reverting to crushing things with rocks. The markets run on vibes and fever dreams of hitting the next big thing.
- YetAnotherNick 13小时前Loss to who? Now all of a sudden, we are caring about investors and sovereign funds?
And I think we passed the threshold for crash down for AI, even if AI companies wont be that profitable. Nvidia/cloud providers will be profitable as long as there is demand for AI.
- dhosek 13小时前Their loss, big deal. Let them suffer. The problem is that when they crash they bring a lot of other stuff down along with them. The people who lost money in the 2008 crash were not the ones who suffered the aftermath.
- YetAnotherNick 13小时前Because in 2008 ordinary everyday people invested in overvalued things like house.
- Mistletoe 13小时前Almost every single person’s retirement has exposure to this unless they have some sort of Bitcoin/gold/small cap value type portfolio.
- rsoto2 13小时前Uhh I think a lot of people and their families likely have investment exposure to nvidia/hyperscalers. if places like Amazon spent unrealistically on ai or their stock goes down massively that could mean major job losses too.
If AI companies aren't that profitable...then they're going to stop spending so much money on GPUs to train AI models. A gigantic amount of Nvidia's profits would go bust overnight.
- YetAnotherNick 13小时前But inference is increasing dramatically. Google says they now do inference of 3.2 quadrillion tokens per month, 7x increase in a year.
Claude code and others are here to grow even if they don't do any further training.
- mgh2 13小时前The strategy of "scale for long term market dominance" or the idea of "build it and they will come" [1] were premised on the notion that adoption will be organic.
AI usage seems to have plateaued overall [2], except for niche use cases like coding, that is why companies are forcing it on their employees to justify ROI [3] or creating "products" w/ AI features [4] or embedded addiction.
[1] https://news.ycombinator.com/item?id=48241012
[2] https://news.ycombinator.com/item?id=48179021
- dcre 12小时前I don’t think “Usage has plateaued except for coding” is compatible with lab ARR at $80B and still growing exponentially.
https://www.theinformation.com/articles/anthropic-openais-sh...
- altcognito 13小时前> AI usage seems to have plateaued overall, except for niche use cases like coding,
I sure hope more people think like this, because it's going to leave a lot of money on the table (for me)
- altmanaltman 12小时前How? Like if AI usage skyrockets, I am sure the money on the table will be gobbled up by multi billion dollar companies before you, i would assume?
And if they are right then what? You won't get a lot of money?
Seems like a weird mix of inflated ego and lack of business understanding by you on this comment.
- scotty79 7小时前Most business is finding the river of money, attaching yourself to it skilfully and sucking a small fraction for yourself.
- Imustaskforhelp 13小时前(I don't quite understand your take?) but overall, companies like cloudflare are basically firing people for the costs associated with AI and layoffs are starting to being questioned with this take.
I don't know what your statement is but if you are an employee, then as your employer is forcing you to tokenmax and forcing you to use slop and creating leaderboards for these token spend which will all end up forcing the company to bleed money afterwards they might even lay off people.
If you are an employer then there are still long term issues associated. For example, cloudflare is a company which hasn't been in profit but it has burnt through 5 million dollars per month for AI as it first created an incentive (shrewd even) for employees to use it (for everything) only to please the investors but in the end, its still unclear how profitable all of it is for cloudflare.
Perhaps I have misunderstood you but I really don't understand how its going to leave a lot of money on the table, the only thing I see is a race to the bottom.
- sigmar 12小时前Plateaued? Lol. Based on what? Pg 18 and 45 on that link are not showing a plateau.
- mgh2 12小时前[dead]
- Aurornis 13小时前We go through this with every startup cycle. Startups are not expected to be profitable because they’re spending so much money on growth and R&D. The concept of running a business in an intentionally unprofitable state is confusing to those who don’t understand startup funding.
The weird thing is that so many people believe that inference is unprofitable. There are large open weights models that companies run at a profit while charging far less than what OpenAI and Anthropic charge. Deepseek V4 just made their 75% off deal permanent and it was already very cheap.
Yes, you have to consider costs of training the models, but as usage grows it’s going to become a smaller and smaller part of the business.
I think we will see some data center businesses and AI companies blow up, but I think the people expecting the entire AI scene to blow up because prices quadruple are going to be disappointed.
- Frannky 12小时前I wonder how much of this reasoning will make sense in the future. How much of the way you are thinking is based on the past curves reality worked before? Are you taking into account exponential acceleration? I guess abundance will flow in such a way that the idea of debt will be a thing of the past.
- sarchertech 13小时前> There are large open weights models that companies run at a profit while charging far less than what OpenAI and Anthropic charge.
You have no idea whether those companies are making a profit.
1. All it takes is one of them operating a loss to gain market share to force the other ones to lower prices to compete.
2. There’s not reason to expect that these relatively small companies are correctly pricing GPU depreciation.
- Aurornis 12小时前> You have no idea whether those companies are making a profit.
I doubt the various providers on OpenRouter are benevolently operating at a loss because they’re so generous.
You can also calculate the cost to run these models yourself. They are open weight and the hardware required to run them is not a secret. They can be modeled and many have done the business modeling.
I’m always surprised at how many Hacker News commenters are unaware that a lot of financial modeling and analysis has been done on these companies and models. It’s naive to think the the hottest topic in tech has not already been dissected and analyzed by the finance industry at every level.
- sarchertech 4小时前Selling a brands new project at a loss to gain market share or to compete with other companies doing it because you hope you can outlast them isn’t being benevolent or generous.
If you want to link to a specific cost analysis that was performed by someone without a vested interest in generating hype then do it and we’ll discuss that.
Because what you wrote sounds an awful lot like “let me tell you a lot of very smart people are saying it.”
- zozbot234 9小时前GPU depreciation cycles are slowing down a lot. A big chunk of frontier model inference is still being run on Hopper-era GPUs because anything more recent is heavily bottlenecked and it makes more sense to use the newer stuff for training,
- JimDabell 12小时前When I go to Amazon and pay them for DeepSeek inference, do you think that Amazon are subsidising that?
- sarchertech 4小时前It’s a brand new market that they want to claim a share of. I doubt they would be making much money of selling deepseek inference right now even if it were profitable, so why not throw sum subsidies at it for a little while in the hope that you are one of the big names left standing once everyone runs out of money.
- JimDabell 4小时前You didn’t answer my question: do you think they are doing this?
AWS already have a strategy in place for what you describe. They are very liberal in giving out credits. They don’t do it by subsidising prices.
- sarchertech 2小时前I don’t know enough to be certain either way. But I will say that I know that Amazon has operated certain product segments at a loss before. Whether that’s with direct price subsidies or credits is irrelevant in the face of a new product with hype unlike anything I’ve ever seen in over 20 years in the industry. It’s highly plausible in the face of this absolute mania and FOMO that Amazon is operating open source inference at a loss to gain market share. They might think that inference prices will drop in the future.
They might be panicking because they don’t have good models of their own. Or they might just be price matching other open source inference providers. They have cut prices to keep up with competition many times over the years.
Whether they are doing it or not, you don’t know they aren’t, and it’s plausible that they are. So the claim that starts with “we know that people are making a profit selling open source inference at X price therefore Y” is unfounded.
- vips7L 13小时前You have to be naive to believe that any pricing is permanent.
- DrewADesign 13小时前These companies are blowing through an incomparable amount of resources. If their bravado is misplaced, the economic impacts will be far more significant.
- choilive 13小时前How so? Most of these companies will take a hit but will be fine Alphabet, Amazon, Google, etc can write off their entire investments in AI and will be a-OK. The pure AI companies will obviously be dead.
- droidjj 13小时前This is what people said about the banks in 2007. Just because the big players’ balance sheets can take the hit doesn’t mean the wider economy is insulated.
- unsungNovelty 13小时前Exactly. The below reply to you also says the banks were bailed out. "So people were right".
How so? Big corps got home safe. Not the people. People committed suicides and lost their livelyhoods.
- fooker 13小时前And all these banks were bailed out by big brother. So the people were right.
- DrewADesign 13小时前A) they still screwed the economy and everyone in it except themselves. B) Nobody gives a shit about the banks as businesses. They got bailed out because they physically made much of the world’s economy function, like plumbing. That’s not going to happen here.
- Zetaphor 13小时前You're still ignoring their mention of the wider economy. The banks were bailed out, but everyone downstream of them still felt the brunt of the impact, atop paying for that bailout with tax dollars.
- dwb 9小时前Yeah and a lot of far less powerful people got fucked over from the crash. Is that what a successful, functioning economy looks like to you?
- onetokeoverthe 13小时前[dead]
- lenkite 5小时前Hah, none of the big companies are going to write off their entire investment. They will come begging for bailouts.
Privatize Profits and Socialize Losses is now Bog-Standard Operating Procedure.
- etempleton 13小时前All of those companies will be fine, but they are currently valued on the stock market for future earnings. Investors anticipate them making a lot more money in the future. So stocks will slide dramatically. Open AI and Anthropic might not survive. And suddenly you see a 20-50% pull back on stocks. That impacts retirement and pension funds. It may trigger a panic and sell off across sectors.
- DrewADesign 13小时前https://fortune.com/2026/05/18/is-ai-a-bubble-1997-or-1999-w...
The stock market. Stocks crash, companies go belly-up, tons of people get laid off, unemployment spikes, people die. I don’t give a shit about the companies themselves. I do give a shit about who they employ, both directly and downstream, and the job market that will result from many of them losing their jobs.
- truncate 13小时前dot-com bubble? It's less about black or white, and more about how much of it. Nothing weird to me about caring given how it all also impacts peoples lives and much wilder all these numbers are becoming.
- SXX 13小时前Difference is that Amazon, Microsoft, Google or Oracle are not going out of business if it all collapses. Neither chip or hardware manufacturers will be harmed.
- wmf 13小时前Oracle is on the edge; if they can't put their capex in SPVs they would get taken out by a crash.
- SXX 13小时前I'm in no way expert on corporate finance, but Oracle has always been known to be sleaziest of sleazy companies. And Larry Ellison is still 6th richest person in the world and is not known to throw money on crazy moonshots like Mark Zuckerberg.
Oracle likely structured everything the way that its gonna be everyone else problem before they go down. No?
- loeg 13小时前Oracle is a tiny fraction of the stock market.
- truncate 12小时前What are you saying then? Don't question or point out things that seem weird? Drink the kool-aid?
- SXX 6小时前This is just not dot com bubble. Its not like someone built x20 datacenter capacity than humanity needs or x10 chip manufacturing capacity or x5 power grid output.
So far capacity barely grown because its super slow to build, but prices skyrocketed x5 to x100.
If its blownup everyone will just return to selling hardware or capacity at 20% margin instead of 2000%.
Only major labs will collapse because they have nothing but models and losses. People working for them still gonna find a job just with $10,000 bunus instead of $1,000,000.
- squibonpig 13小时前The economy is currently kinda riding on them.
- etempleton 13小时前The difference is the sheer scale of the spend. I bet that SaaS company hasn’t spent the annual GDP of a small nation. If Chat GPT can’t pay the bills it is going to ripple through the economy likely causing at minimum a large correction. If the SAAS company goes under hardly anyone noticed.
- dnnddidiej 12小时前SaaS or web in general was disrupting X making it eventually the leader with some moat. I am not so sure about AI. I feel like there is a rush to make a commodity that will be nebulous to extract value from. Except for TMSC and NVidia.
- bertylicious 12小时前What's the company's name? And why the unnecessary secrecy in the first place? It's a publicly traded company so this information is public by definition.
- wmedrano 13小时前We're talking about ~1 trillion $$$ valuations here tho
- nozzlegear 13小时前What do you mean suddenly? People have been talking about it for as long as relatively early stage LLM companies have been noteworthy.
- quantummagic 13小时前You misunderstand. He's saying there is a double standard, one for pre-LLM companies, and another for LLM companies.
- jchw 12小时前For the past 3 decades, it really has been normal for companies to remain very unprofitable even up until their IPO, but I don't think it's actually normal in general. In fact, if AI investment really is a bubble and it pops, I reckon it could very well mark the end of this era!
(Is there a more extreme example so far of this than AI companies, just in terms of raw losses? As far as I know, Netscape's lifetime losses as an independent company "only" total a bit over $100 million dollars, which is a lot, it just doesn't look like all that much when put into perspective...)
- dev1ycan 13小时前Maybe because losing 700b so far is not "safe" for the economy?
- guidedlight 13小时前The US “loses” $1T every ~150 days on delivering basic government services, and every US citizen is on the hook for that, not just investors.
- somat 13小时前It does not have to be bad, it depends on who they lost it to. Nvidea probably wins, the data center construction companies, electric companies etc. The tricky thing about an economy is that big picture "losing" means money is not moving and "winning" means that it is.
- shimman 13小时前Well seeing how they've all collectively spent over a trillion dollars and American citizens still don't have medicare for all, universal childcare, free school lunch, a publics job program, or universal education; it's quite easy to see why the American public has soundly rejected this technology where some are even trying to inflict violence to stop it.
- cindyllm 13小时前[dead]
- onetokeoverthe 11小时前[dead]
- toomuchtodo 13小时前The AI Bubble – No One's Happy - https://news.ycombinator.com/item?id=48230753 - May 2026
- dangus 13小时前It’s weird to me that profitability is so thoroughly dismissed by the software tech industry because of an assumption that the tech industry will always be “early stage” and “high growth.”
We can look at a “success story” like Uber and it is still net negative over its entire existence. This is a business that’s in a literal monopoly/duopoly status in most markets it operates in with vastly reduced regulatory burden compared to the industry disrupted. Literally the ideal scenario for printing money and yet it hasn’t made any. It’s the poster child for the unicorn exit that founders dream of.
The end result is that Uber and companies like it are a financial instruments that transfer dollars away from one set of investors to another set of investors.
If Uber hasn’t yet made its investment back, I struggle to wonder how some of these AI ventures will ever make that money back when their expenditures make Uber look like a small little side project.
Meta has spent almost 4 years worth of its net income for FY2025 on AI going by this website’s data, and counting.
We are decades since Web 2.0 took off, almost 20 years since the iPhone launched, 50 years of Apple Computer. Software isn’t some new industry anymore. There isn’t an industry left that hasn’t completed its digital transformation. These spray and pray economies would have died off years ago if it wasn’t for the fact that software companies have uniquely low cost structures where they don’t need to build factories or distribution networks to get their products to their customers. These low cost structures might just be concealing the fact that it’s not going to be a growth industry forever.
- 48terry 13小时前And also: AI is basically the only thing anyone is talking about. Yeah, Uber existed and it's known about and was advertised and such. It has not overwhelmed every topic ever like the current LLM mandate has been. People are getting sit down and told they MUST engage with this stuff.
How has the sheer saturation of LLMs not resulted in profit? It has dominated the conversation, center stage, of every news outlet for like 4 years now. It is the most known-about thing currently out there.
And we haven't been able to convert that much captured attention into profitability yet? That seems... bad?
- bix6 13小时前But why would you make it profitable now? We are still in the early innings and its growth at all costs. Growing from sustainable cash flow isn’t fast enough for investors, they want HYPERGROWTH (now with RAWBERRY)
- dangus 13小时前Right! I think the only example that comes to mind for me as far as “bled money for years and eventually became a cash cow” is YouTube. Most other ventures that bled money that long ended up dying.
Maybe Reddit is an example? But my impression is that they ran a modest operation before going public.
ChatGPT is the 5th most visited website in the world. Gemini.Google.com is ranked above amazon.com. Where is the profit?
- abathologist 13小时前It's not weird if you consider the details and the many ways that the situations are very different. But also, people cared about that other kind of BS too, e.g., https://news.ycombinator.com/item?id=39438372 or https://www.currentaffairs.org/news/2017/10/undercover-at-th...
- rdl 14小时前For a rapidly growing new line of business, this isn't bad at all.
- emacdona 14小时前Yeah, my first impression when I saw this was: if this is accurate, the situation is not nearly as bad as I thought.
I do wonder why Nvida is included, though. If you include the company that all of the frontier models are pouring money into, of course the net (expenditure - profits) of the collective is going to be closer to zero :-)
- emacdona 13小时前Additionally:
If Nvidia is included, does that mean that the money Amazon, Microsoft, and Oracle get for selling compute to the frontier models are included in their revenue?
Because for Amazon in particular, the situation this pages shows is actually much WORSE than I expected. I thought they were making a killing selling compute for model training.
- SXX 13小时前Problem is not profitability as is. Nvidia's net of circular funding is the problem though:
- orblivion 13小时前Yeah they should also include the power companies for that matter.
- bandrami 14小时前Unfortunately the green bars are not just EBITDA, they're before discounts.
- diziet 13小时前Right, especially given that majority of this investment is into GPUs and data centers that are amortized over a longer period of time. This is actually very hopeful.
Given how the curves look like in terms of ramping of spend, these are very healthy numbers.
- abathologist 13小时前'Long', as in 1-3 years? (https://www.tomshardware.com/pc-components/gpus/datacenter-g...)
- rsanek 12小时前TPU v2s that were released ~10 years ago are still being sold via GCloud.
https://cloud.google.com/tpu/pricing https://en.wikipedia.org/wiki/Tensor_Processing_Unit#Second_...
- diziet 13小时前Even if it's 1-3 years, they are very likely to be ROI positive all in.
- delish 13小时前The critic I see most frequently on the unprofitability of AI is Ed Zitron. I am sincerely curious if he shorted Facebook's, Amazon's, or Google's stocks. Or if he's in index funds which have tech stocks like those.
For example: I have index funds which have some of these stocks. So I, by process of revealed-preference, don't think it's a bubble, or I think I will keep my money in through the bubble's pop. I don't have that much else to say!
For the record: I would respect the creator of this site equally or more if he/she said, "I'm shorting these stocks and this is why."
- locusofself 14小时前certainly will be interesting to find out..
- abathologist 13小时前Oh really? A 195% cost to revenue ratio isn't bad at all? I'm not a biz expert, but I spent a few minutes looking this up (e.g., what are usual cost-to-revenue ratios for new lines of business), and this sounds like BS to me.
- rdl 10小时前If "cost" is mostly capital investments, absolutely. Normally you'd use operating cost (which for capital equipment would be depreciation and interest), and here they are using the capital cost as full cost.
No one really knows how quickly AI hardware investments will become obsolete and thus how long it should be amortized, but 2-3 years would be extremely conservative, and in fact used H100 (discontinued/2 generations old) prices are higher today than they were when the equipment was new several years ago.
- ch_fr 4小时前But if it's fully being amortized, then it means they don't buy new Nvidia GPUs anymore for a while. The situation is either "your GPU AND the datacenter infrastructure it's running on is obsolete", or "Nvidia's profits tank because people are staying with current-level infrastructure".
- rdl 3小时前That would be true if everyone weren't supply constrained and buying literally everything they can find.
There are actual risks that this trend doesn't continue, but as long as the trend continues, it is pretty good for revenue. "AI shown to hit a wall/doesn't actually deliver/stops growing so fast", "massive improvement in hw efficiency or tech such that all the old stuff becomes obsolete", "bottleneck on power/regulations/etc such that no one wants anything but the most efficient cutting edge stuff" would be the ways it could end and then all these factors reverse. Right now, power is so constrained that old, inefficient power generation is actively being turned back on or set up at new sites (e.g. old aviation turbines which are very inefficient compared to combined cycle).
- ZetsuBouKyo 13小时前I am relatively pessimistic about the profitability of those panning for gold in the downstream AI market.
The core bottlenecks are power and computing capacity, and they actually trace back to the exact same issue. It all comes down to the physical energy it takes to flip or move a single bit inside the ram or disk storage. This concept is subject to fundamental physical barriers.
There are a few ways to tackle this, like improving power efficiency, reducing model size, or pushing hardware further. However, achieving orders-of-magnitude improvement in any of these areas will cost a massive amount of time and money. I wonder if governments, corporations, and investors have the patience to wait for these tech breakthroughs.
- hootz 14小时前So Nvidia is basically farming everyone else?
- SXX 14小时前Other hardware manufacturers also wastly more profitable - RAM, SSD, HDD and literally everyone in datacenter supply chain.
- dawnerd 13小时前It goes back to the whole, you don't make money mining gold, you make money selling shovels. Nvidia has been playing every tech hype cycle recently. Question is, what will be next.
- keyle 14小时前It's historically called: selling buckets and shovels during a gold rush.
The only way to get consistently rich in any bubble economy.
- frm88 8小时前Yes and shovelling the money around in yet another circle: https://www.youtube.com/watch?v=xUbJDrL6ZfM,
- bandrami 14小时前Them and Broadcom
- somat 14小时前It's the parable about how in a gold rush you want to be the guy selling shovels.
- DrewADesign 13小时前[dead]
- PinkaDunka 13小时前Oh wow, they already got 50% of investments back in roughly three years? This is going to be insane money making machine. Or is it not the point op was trying to make?
- 0xbadcafebee 45分钟前Keep in mind that some of the revenue booked is due to circular deals. AWS for example gives Anthropic "cpu credits" that Anthropic spends on AWS, and AWS books that as revenue. The reverse is true too: Anthropic gives AWS inference credits and Anthropic books that as revenue. When it's really just a pre-existing server sitting there and two companies claiming they made money off it. There are also equity deals going on the same way; they give each other stock, (or one gives the other stock in exchange for GPUs) and both benefit as the stocks get inflated despite there not really being more value.
- tim333 5小时前Technically, the site doesn't show profits, it shows something like cash flow excluding investment flows, or has more money been spent than has come back in from customers.
This will always be negative for any new business as you are effectively depreciating the assets straight away. Like if you build a hotel and deduct the cost of building it from room income - it would take years before you get the money back but may be quite profitable with GAAP accounting.
GAAP accounting (Generally Accepted Accounting Principles) is what's used for official reporting and tax returns but excludes any increases in IP value or goodwill unless there's a buyout. If you included those the likes of OpenAI or Anthropic would have done pretty well. I'm not sure there's a word for that but basically value of the business less the money that's gone in. It doesn't get reported because 'value of the business' is guesswork and can be prone to BS but is pretty important to real world outcomes. AI is probably doing well on that one. Maybe why
>Is AI Profitable Yet? NO. Everyone's Broke.
doesn't fit with the top companies on the list having many billions in the bank.
- casualscience 13小时前How are the Google numbers calculated? I've seen their net income increasing a lot as they've rolled out Gemini. This suggests that Gemini tokens are actually profitable, or at least not extremely unprofitable.
Yet this site suggests that tokens are very unprofitable
- missedthecue 13小时前The site doesn't suggest anything useful. It's more of a fun meme.
Building a datacenter that will produce hundreds of billions of dollars worth of tokens over a multi-decade life shouldn't surprise anyone that it's in the red in year 1 or 2. There's a lot of front loaded capex in this business. If someone built a tractor factory you wouldnt expect 1 year payback.
But the site sort of implies that these companies are selling tokens for less than it takes to inference them. As if this is some sort of COGS ledger. Especially by throwing Nvidia in there. Don't take it too seriously.
- girvo 13小时前> This suggests that Gemini tokens are actually profitable, or at least not extremely unprofitable.
Out of all the companies, considering their own silicon etc. I wouldn't be surprised. Though I do wonder in terms of total CapEx and R&D where it would be at...
- crowbahr 13小时前Google is making money on selling cloud compute. Their margins have gone from 9% to 32%.
They're soaking up the investor bonanza into AI - Gemini ain't making them money.
For context Cloud Compute made 20bn in Q1, Other services made 90bn.
- casualscience 11小时前Search, yt, etc.. are the real revenue sources for Google. But they are serving [3.2 quadrillion tokens per month](https://blog.google/innovation-and-ai/sundar-pichai-io-2026/...) and have ramped up a ton in 2026, yet profits continue to expand.
Comparing to ad revenue from a company like meta, the story that Gemini tokens are a strong cash drag on Google just doesn't add up. It seems at worst they are losing like 50 cents/1M tokens (including r&d spend, data centers, etc..), and very possible they are actually profitable per token.
Which is much better than anthropic and openai.
- nothercastle 13小时前I mean yes they are serving ads off websites they Plagiarized with AI. So if you use ai to serve up content you don’t own as your property then perhaps you can make money. The cost is that they are completely killing the creators
- c0rruptbytes 14小时前Deepseek is really killing it if that's their total spend
- conradkay 13小时前https://www.techinasia.com/news/chinas-deepseek-eyes-10b-fun...
Would be weird if they're raising $10 billion after spending only 0.3
https://newsletter.semianalysis.com/p/deepseek-debates
Probably more like 3-4 billion by now?
- hootz 14小时前Right? Their V4 model is too good for them to be spending less than 1% of what Anthropic is.
- keyle 14小时前Yeah. Meta on the other hand. Ouch.
- hootz 13小时前Apparently their strategy is to dump a fuckton of money in hopes that that will make them dominate the market, just like they did with the metaverse thing. It's like a hobbyist who buys the most expensive gear on the first day of trying out a new hobby.
- keyle 9小时前I thought it was to burn the planet so as long as one island remains untouched.
- turtleyacht 14小时前Didn't see Radeon, but they have an AI page: https://www.amd.com/en/products/graphics/radeon-ai.html
- TOMDM 13小时前AI startups taking unprofitable risky ventures in search of growth opportunity and future returns makes sense to me.
Maybe most of them or all of them lose on their bets, but there's potential for a future where revenue grows beyond the immense capex and research investments.
Oracle though... Immensely risky capex to service a startup industry with what will soon be a commodity...
- mhjkl 13小时前Reminds me of the “Has The Turing Test Been Passed” website. It says no, but if you read on they cite “The relatively minimal funding allocated to AI research” as one of the reasons AI hasn’t been achieved “yet”. Website stopped being updated before it became relevant, so you will never see it say “yes”, similarly to how the Loebner prize mysteriously vaporized when GPT-2 came out, just when winning it for real started becoming an interesting possibility
- mullingitover 10小时前Are we ignoring the 15 billion/yr deal with SpaceX/"Xai" and Anthropic?[1] Or is that not booked revenue yet?
[1] https://www.wired.com/story/spacex-ipo-anthropic-compute-fin...
- acoward113472 13小时前I don’t think this website is fair. It does not factor in productivity increase and ROI from other areas that utilize AI to complete what they were doing. For example, if a new operating system was built into AI, the profit for that would go towards increased sales of licenses but this site seems to only track return on AI businesses
- daveguy 13小时前Cool. Why don't you build that?
- samstokes 13小时前I don't have an MBA or anything but is it common practice to describe "revenue - capex" as "profit"?
- andai 13小时前That's pretty funny. For the "yet" part I would have expected a more recent cutoff, rather than the whole history of the companies. (Do they all have some kind of enormous debts they're going to need to pay off for decades once they do become profitable?)
- timonoko 13小时前Gemini now remembers you wholesale and makes good analogies and shortcuts knowing youres personal capabilities. You are already hooked and paying starts any day now. Or maybe it starts recommending some marvellous products somewhat related to your query.
- conradkay 13小时前The numbers for Meta are pretty misleading, I assume the $3 billion is something like direct generative AI revenue?
Sure they're torching money on building consumer LLMs, but they seem to be doing very well optimizing things like ad ranking
https://engineering.fb.com/2025/11/10/ml-applications/metas-...
https://engineering.fb.com/2026/03/31/ml-applications/meta-a...
- trimbo 12小时前Yes whoever made this website apparently doesn't understand how LLMs are being used for massively profitable consumer products like Instagram.
- throwaway85825 13小时前Isn't ad ranking just SORT price;
- OsrsNeedsf2P 13小时前Assuming you're not trolling, there's a few other things to consider -
1/ User targeting is complex - you can charge more for ads if the users you're showing the ads to click
2/ Ads impact user retention - you need to balance making money and keeping users around
3/ AI generated ads - this is a pretty big thing now, where instead of bringing your own media, you just describe your target audience and the AI will A/B test media + CTAs for you
4/ Integrity - you want to vet the ads against laws/site policies
Probably forgetting a few, but there's a reason the ad industry employs so many
- throwaway85825 2小时前I've heard of the ad auction and assumed the point of the auction was to maximize price.
As for integrity ad platforms don't have any. Most of the ads seem to be for scams. The first search result for OBS is usually malware. Scammers have a low cost to advertise because they use stolen credit cards. Advertisers don't mind because the charge back doesn't cost them, COGS is near 0.
- jwitthuhn 11小时前For immediate revenue yes, but if you want advertisers to keep coming back you need to give them a good conversion rate.
- sph 4小时前Why does NVIDIA, the most profitable AI company, not simply eat all the other non-profitable AI companies?
- gield 8小时前For some use cases, AI has been profitable for at least a decade.
- try-working 14小时前Possibly profitable for New-Gen Labs (DS, Qwen, Kimi, etc) and impossibly unprofitable for the Legacy Labs (OpenAI, Anthropic)
- gamander2 7小时前Never bet against Wall Street and jewish capital, they will make money off AI by hook and by crook. If necessary, they will push for escalation with China and enforce a global ban on Chinese AI. You may disagree, but after the experience of 200 years of Western capitalism, you will have to explain why this time it will be different.
- cindyllm 5小时前[dead]
- firecall 13小时前Is the "$ Spent on AI since page load" broadly indicative of spend at all, or just a fun animation?
- hn_throwaway_99 12小时前I know not the point, but is "PNL" a term people use now vs P&L?
- burnerRhodov3 13小时前Xai is making 1.25B a month off it's compute? Why is that not listed?
- alemseo 12小时前NVidia showing so much profit, even they have acceess to some models like QWEN for free. Talking about Anthropic and OpenAI, they charge so much, I dont understand this graph.
- beej71 13小时前Nvidia making out selling shovels, that's for sure.
- beloch 13小时前Remember the model:
1. Outspend and outlast your competition until you have market dominance. Win over and lock in your customers with sweetheart deals.
2. Enshittify and squeeze your customers to pay back your debt.
If you're using AI, you're not paying the true cost right now because we're in phase 1. Be ready for phase 2.
- sothatsit 13小时前Or, tokens are more like energy and prices will drop over time until they reach some equilibrium.
The big labs are actively moving into the application layer, where they’ll have more pricing power. Maybe that layer will end up with a Mac (Anthropic) vs Windows (OpenAI) vs Linux (open-source) dynamic as well if they can create a moat. But so far it’s pretty easy to move between providers.
- xigoi 11小时前> Or, tokens are more like energy and prices will drop over time until they reach some equilibrium.
In that case, AI companies will never get their money back, leading to a huge crash.
- fc417fc802 13小时前Given that the likes of openrouter exist I'm not sure how phase 2 is supposed to work.
- sunkeeh 13小时前This site is going to start doing the opposite of the author's intentions in a couple of years.
- dnnddidiej 13小时前Did someone say shovels?
- scotty79 7小时前Wow, the industry is only 50% in the hole during such a massive buildout that overburdens supply chains of such basic and general resources as energy and chips? This is going to be insanely profitable in less than a decade. I'd be delighted if a business I had put 100k in aleready had a 50k return, after first 3 years or so, while I'm buliding it in conditions where everybody else is doing pretty much the same.
- code_duck 13小时前I've received some decent benefits from it without paying anything.
- raincole 14小时前Now use common accounting standard and amortize the cost.
Oh it doesn't fit the narrative. Never mind then.
- BirAdam 14小时前The depreciation is also insane and thus to sustain operations and improve, the spend will keep going.
- locusofself 14小时前I assume you are saying it would look less ridiculous? By how much?
- bandrami 14小时前If OpenAI and Anthropic adopted GAAP nobody would be able to invest in them it would be so bad
- unmole 13小时前Yeah, institutional investors who plowed billions into them are unsophisticated rubes who got hoodwinked because they don't get GAAP. And it's not like both OpenAI and Anthropic are both going to IPO soon which would require disclosures far beyond GAAP numbers. /s
- bandrami 12小时前I take it you weren't alive in 2008 if you say that sarcastically
- unmole 11小时前You should do a Michael Burry and go short. Clearly you seem to think you are much better informed, put your money where your mouth is.
- bandrami 11小时前Shorting isn't the magic trick people seem to think it is, but yes I'm heavily hedged for OpenAI's IPO failing and taking Oracle down with it
- collingreen 14小时前In what ways do common accounting standards and amortizing the costs (this is tricky for ai and the current batch of gpus I hear!) change the data presented here? Does it detract from the point? Completely contradict it?
You can turn your drive-by dismissal into something really informative if you want to.
- raincole 13小时前First of all, the whole website is based on what the CEOs said they're going to spend. Not the actual money spent. So there is no real 'data' presented here or to contradict.
Second, even if you take CEOs' words at face value, they didn't distinguish the capex for hardware, electricity, software and salary. You can make up whatever the percentage for hardware and the depreciation rate you believe and fit an arbitrary narrative.
- qmr 12小时前Sell shovels.
- ekianjo 13小时前Most startups are deep in the red for years before they even have any revenue. Is that any different?
- ch_fr 4小时前The difference between serving 1 and 1 billion http queries is not the same as the difference between generating 1 and 1 billion tokens.
The startup blitz-scaling-market-capturing playbook makes makes sense when you spend to scale, not when you spend because you scale, yeah, I understand that step 2 is "and now you squeeze the users", but it will need to be by such a bigger factor...
- charcircuit 13小时前This ignores how much the stock has grown due to AI.
Also many of these companies like Amazon, Google, and Meta drive a lot of incremental value due to both AI powered content suggestion and AI powered ad suggestion. Personalized ads has driven a ton of revenue.
- abathologist 12小时前Good point. We mustn't forget the magical money box: that's where the real value is delivered!
- charcircuit 11小时前If you can spend $1 to raise the value of your company by $2 it makes sense to do regardless of if that $1 is directly earning a profit.
- blindriver 13小时前Anthropic is going to be profitable in the June quarter
https://www.cnbc.com/2026/05/20/anthropic-revenue-explosive-...
- MadxX79 12小时前But then go right back to being unprofitable again afterwards, which is a little weird.
- IAmGraydon 13小时前This is pretty funny. Now do it without Nvidia and including all costs, not just capex.
- bze12 13小时前Comparing fixed costs to revenue? Even if it’s cumulative this seems like a disingenuous framing
- cat_plus_plus 13小时前Yes, I spend my days writing lots of code using AI (I do rigorously review it, it's still much faster than hand typing) and I get paid enough for it to pay mortgage and send kids to college.
- clearstack 5小时前[dead]
- rohitsriram 14小时前[dead]
- xorgun 13小时前[dead]
- SXX 14小时前To be honest whatever author wanted to say there three categories of AI related companies: hardware manufacturers, cloud providers and purely AI companies.
Only the later have something to lose if AI bubble gone by tomorrow. Everyone else will just stay with grown capacity and reuse infrastructure for whatever.
Not listing other hardware companies is just dishinest. AI is not a crypto mining where resources are just burned.
- flexagoon 14小时前> Not listing other hardware companies is just dishinest. AI is not a crypto mining where resources are just burned.
AI is exactly like crypto mining in that Nvidia is the one who profited from both
- SXX 13小时前Crypto mining bubble was 99% of speculation plus scams and 1% of R&D for decentralized finance.
No matter what happen with the AI bubble text, image and video and other generative neural networks are here to stay.
Whatever you like it or not this tech already changed a lot of industries and there is no going back.
- flexagoon 6小时前I specifically said they're the same in that Nvidia profited from both of them, not that they are the same in all other aspects.
- nothercastle 13小时前Bitcoin is here to stay so is crypto but the impact was much more limited then initially predicted
- DrewADesign 13小时前And the money these companies are blowing on all of this shit is banking on them being ‘the’ dominant player in a completely world-changing commodity industry.
- DrewADesign 13小时前I haven’t heard any compelling use case, in the event of an industry implosion, for many many many billions of dollars of GPUs that were already proven too unprofitable to operate for the industry they were built for.
Dark fiber, for example, had a much more compelling use case.