Just last week I was on the call with a a techy architect from one of the major-major financial services companies, and he was discussing getting token capacity and his words on third best option was "we'd get it from Coreweave or whoever is still alive". So yeah, I am pretty sure that if that's the sentiment from software people at the major-major financial company, Coreweave is cooked.
There is no secret about the mismatch. Voters last week have elected people who are openly hostile to datacenters, and for some of them hostile to AI outright. The demand is there (at what price is a longer term debate). But if you can't power on new datacenters this doesn't do any good.
To give you an idea of the scale. US has currently about 50 GW of DC online. Depending on the source, you are looking at between 60 GW and 400 GW of new DC currently planned. For reference the total dependable capacity of the US grid is around 1000 GW.
I think there is a scenario where compute cost sky rockets while chip prices crash, as they end up unused in warehouse because of the lack of power and DC capacity.
Every civilizational advance “exacerbates inequality” because smart people can use it to become more productive, but dumb people can’t.
By that logic we should ban books. This is Boko Haram ideology and you are just a communist who thinks he should be protected from competition.
Instead of “making everyone well off“ here is what will actually happen, because that’s what happens every time communists seize power: everyone will get poorer, because they will be outcompeted by more sane jurisdictions, eg China.
That sounds like
“Threatening me with a good time”. It will be very cheap for me to buy my hardware. Unsurprisingly I have power at home, and I can use the free power I have during the day from solar to do all sorts of cool things on my own hardware. It will be like the 1980s and 90s again.
But a large reason for the price increase has been because companies have been hoarding them in warehouses as they wait for DC buildout. If you sell then you remove that pressure and prices decrease.
It's only RAM, mass storage and GPUs. Desktop CPUs, mainboards, PSUs, and cases have become cheaper because PC building demand has declined. Peripheral prices have remained largely unchanged.
This is how all capital-intensive utilities operate. The marginal cost of using a little more power or tokens is easy to justify because you can at any point stop using more.
But the cost to add more power capacity is huge and highly variable. Just because oil is expensive today doesn’t mean the capex to pump more is justified.
Really what AI needs is large customers that will commit to large spend over the next 5-10 years. Without this AI companies are basically offering duration arbitrage of electricity. Not a good place to be; the finances will always look bad.
Unfortunately, the amount of capital required by the entire AI industry is enough that counterparty risk is now a concern.
Wait, Anthropic has $112 billion in cash right now and generates about $70B in revenue per year and grew 10x YoY. If they succeed in raising $100B in cash in the IPO, which I believe they will, they'll have $212B. If their growth declines to 3x (down from 10x) in 2027, they'll still be at $230B in yearly revenue.
Further more, their compute commitments are over 10-20 years. Their average quarterly compute commitment is only about $11B - $15B which they have already surpassed with their revenue.
At $230b/year, they’d be at around $57b in quarterly revenue against compute of around $16b. This is not bleeding cash.
Why do you say they're running out of cash? Can we see your math?
The only reason Anthropic has cash is because VC keep giving them the money. They never had a positive quarter, with revenue always lower than expenses regardless of their growth, thus burning money from VC.
If VC money gets tight, they would have to issue debt, that pay interests. With a current negative EBITDA, their bond will be junk at IPO, paying higher interests and having a lot less market than AAA debt (e.g. Google has access to that market). Both things combined translate to "running out of cash".
It's not exactly laziness. It's that most people don't understand basic financial concepts like fixed versus operating costs.
Companies like Anthropic and OpenAI have invested huge sums in building/training models. Those are the headline numbers everyone talks about. What they don't understand is that the gross profit margins on AI model inference are very high.
The question is whether these companies ever get to the point where the profit from inference justifies (recoups) the investment made in training. If they have to keep investing vast sums in training new models, and/or competition causes the margins on inference to collapse, they're in trouble. But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
> What they don't understand is that the gross profit margins on AI model inference are very high.
Can we be sure of this? The second they tried to hike the prices to be profitable, they lost a lot of clients, or a lot of clients lowered they token expenditure. In any market margins tend to go to zero. I am using frontier models, but only because they are dirty cheap if not free. If they raise the price, I just switch to a cheaper model, start exploring the option of renting a GPU-VPS or buying my own card. Or reducing my AI usage, that currently includes stupid useless things like "translate this cooking recipe from french... for free".
> But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
They are being priced as the "ifs" are certainties, and not only a possibility. If anything goes a millimetre below perfect for them, their finances collapse. High margins attract competitors, and unless you have a huge moat (and their only moat is expensive training), your high margins vanish.
1. Model Build/Training costs are escalating if anything
2. They are in an arms race with competitors and can't afford to stop building new models - if they lose the perception of leading, they lose lots of customers overnight.
3. Inference margin is under attack from more specialised models and open weights models.*
4. They have no moat given how simple the interface to the products is - a bunch of text files and text prompts, and little prospect of building a moat.
So those all run counter the possibility of coasting on previous research. To me the most promising avenue for making money for these companies is to replace Google, not AGI or taking over the world's knowledge work. That would mean they are overvalued. Even coasting on previous research, how long would it take to pay off the money they have spent?
> It's that most people don't understand basic financial concepts like fixed versus operating costs.
No man. It’s that, like you’ve explained further in your comment, they haven’t become net profitable yet. Being profitable in 1-2 quarters is a good trend but it doesn’t matter if you are still behind on mountains of debt or investors wanting their money back and your market is one where people talk about shit that happened 4 months ago as obsolete.
There is no proof that the current profit they are making on inference is durable.
Gross profit margins include all the direct costs of providing a service. So if gross profit margins of Anthropic are positive, they are positive after they subtracted the cost of electricity from revenues. And not just electricity, but every cost that is necessary to run a large scale inference operation.
> It's not exactly laziness. It's that most people don't understand basic financial concepts like fixed versus operating costs.
> Companies like Anthropic and OpenAI have invested huge sums in building/training models. Those are the headline numbers everyone talks about. What they don't understand is that the gross profit margins on AI model inference are very high.
Only when you ignore that the models they provide inference for have an extremely high cost and pretend they don't depreciate. They keep having to train new models. I don't think they recovered the cost for a model when a new one releases.
It is kind of cute to say other people "don't understand basic financial concepts like fixed versus operating costs" when you seem to not understand basic financial concepts such as depreciation and amortization.
Also, and this is important, we don't really know their true compute cost. We don't know if their compute with the likes of Google and Amazon were initially subsidized, and for how long it might be. In Q2 they also got to not pay SpaceX for compute for a while.
What's cute is that you're falling into the same trap of confusing accounting concepts.
Gross margin is revenue minus cost of revenue (COGS). COGS is basically the compute that serves inference (rented capacity, or depreciation on owned GPUs used for inference, plus expenses like support, etc.).
Model training is normally treated as R&D. Under US GAAP, R&D is expensed as it occurs and is below gross profit as an operating expense. It is not COGS.
You mention "depreciation" but depreciation applies to physical assets like GPUs. The GPU depreciation that belongs in gross margin is already in COGS. A trained model would be an intangible asset, so that is amortization, not depreciation. Even that only applies if the training cost was capitalized, and under GAAP it generally wouldn't be. The full cost is expensed when it's incurred. Nobody is "pretending models don't depreciate." Their cost is counted all at once but not in gross margin because that's simply not what gross margin measures.
Obviously, models lose value fast in an economic sense. They get replaced within a year or two and competition limits what they can earn. But that's an argument about whether the R&D expense is sustainable. It doesn't mean that gross margins are being misstated.
"Inference has high gross margins" and "training is extremely expensive" can both be true, which oddly seems to be something many people have trouble comprehending. But this is the crux of the questions that will determine whether these businesses are viable long-term:
1. Will training costs fall over time, or at least grow more slowly than revenue?
2. Does the gross profit a model earns from inference before it's replaced exceed what it cost to train?
3. Will inference margins hold up as competition and open-weight models push prices down?
You have a legitimate point about compute costs. If the deals with Google, Amazon or anyone else are priced below market, that directly inflates gross margin, because it lowers COGS. But we don't know those terms, so that's just speculation, and it's still a different argument from depreciation.
To reiterate what I wrote:
> The question is whether these companies ever get to the point where the profit from inference justifies (recoups) the investment made in training. If they have to keep investing vast sums in training new models, and/or competition causes the margins on inference to collapse, they're in trouble. But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
> Model training is normally treated as R&D. Under US GAAP, R&D is expensed as it occurs and is below gross profit as an operating expense. It is not COGS.
Isn’t it plausible that this is conceptually wrong? What if a model’s revenue-generating potential only lasts for something like a year due to, loosely speaking, competition? Doesn’t this make model development more akin to, say, the costs of annual agricultural planting/harvesting/delivery cycles, or a tv network’s annual (back in the day at least…) production cycle for its main content?
They are when they say that inference is incredibly profitable. Its incredible profits exist only when you pretend that models have 0 cost to be created (since they are a digital asset that can't depreciate), and R&D is a imaginary thing that exists only as a fixed cost of doing business that doesn't need to be recovered.
Anthropic released Opus 4.6 in February 2026 and deprecated it in June 2026. Was it anywhere close to recover the "R&D" costs during those few months?
More importantly, how long a model needs to be current until it recovers "R&D"? Their economic viability hangs on this, and the compute costs both to train and servr models.
A lot of people like to vaguely gesture at the possibility of compute costs coming down, but it is in fact coming up with the rising cost of hardware (and possibly with sweetheart compute deals being replaced by contacts where providers actually have to make money).
> They are when they say that inference is incredibly profitable. It is incredibly profits exist only when you pretend that models have 0 cost to be created (since they are a digital asset that can't depreciate), and R&D is a imaginary thing that exists only as a fixed cost of doing business that doesn't need to be recovered.
Expensing the cost of training a model isn't treating it as free. The whole cost is charged to the income statement the year it's incurred, which is actually faster than depreciation would ever recognize it. Calling something R&D does not make the cost "imaginary". It means the cost is counted. It's just not in gross margin.
We are actually asking the same basic question (can the investment made ever be recouped?). What you're struggling with is the accounting concepts and terminology. The costs of creating the models are a different part of the income statement than the costs of serving inference.
Again, "inference has high gross margins" and "training is extremely expensive" can both be true at the same time.
In absolutely none of my comments did I state that R&D never needs to be recovered. In fact, I stated repeatedly that the entire crux of whether these businesses are viable long-term comes down to whether the costs of creating the models is eventually recouped.
> Anthropic released Opus 4.6 in February 2026 and deprecated it in June 2026.
First off, Opus 4.6 is still available, even though Anthropic recommends migrating to 5.5. Google Cloud says it will be on offer until February 2027. So revenue for this model didn't drop to $0 overnight.
More importantly, your comment implies that models cease to have value when the next version is released. This completely ignores the reality of model training, which is that the subsequent models almost certainly built on the research, data and techniques of 4.6. And 4.6 was built on the research, data and techniques of the models that came before it.
So the whole notion that point versions of these models are standalone investments that cost billions to train and then become worthless the minute a new point version is released is incredibly naive. It's just not how model training works.
> A lot of people like to vaguely gesture at the possibility of compute costs coming down, but it is in fact coming up with the rising cost of hardware (and possibly with sweetheart compute deals being replaced by contacts where providers actually have to make money).
You have absolutely no evidence of "sweetheart compute deals" so this is pure speculation.
Furthermore, there's a flaw in your logic. It's entirely possible for chips and memory to get more expensive while the cost per token served falls due to more efficient hardware and better inference software. What matters here is the cost per unit of output, not the price of a GPU.
Bottom line: inference has high gross margins and these businesses have had to invest massive sums. If inference profits grow sufficiently faster than costs, or costs decrease sufficiently over time, these businesses could be viable and perhaps incredibly successful. If that doesn't happen, their viability is questionable.
Lots of people have an opinion about this but as this discussion proves, lots of people also don't understand the basic accounting required to make a cogent argument one way or the other.
(units in Kilo-millions)
Revenue in first 2026 9-months 33.2
Compute and infra expenses 18.8
Personel 4.8
Sales 5.5
Administrative expenses 5.3
Operating earnings (loss) (1.2)
* In the case you are confused, () in finances means negative values.
Q-3 only, this is calculated substracting first half from 9 first months:
Revenue in Q3 16.9
Compute and infra expenses 9.9
Personel 1.7
Sales 2.3
Administrative expenses 3.6
Operating earnings (loss (0.6)
Cashflow might be confusing here, because the cashflow table include huge items like "Change in fair value of convertible notes, derivative liabilities, and SAFEs" (translates to "we have paper valued at $X, but now it is valued at $Y", but it is still paper) or "Proceeds from issuance of redeemable convertible preferred stock, net of issuance cost" (translates to "we sold paper to VCs for $X") that are not business or recurrent items, or at least I hope Anthropic don't plan on "earning" money by selling paper to VCs forever.
Small gripe but I don't like 'declines to 3x'. 3x growth would mean repeating this year's record growth twice over, which is smaller relatively but an increase in absolute growth rate.
I have seen most revenue projections are on the order of 2-3x, but I suspect about 4-5x.
IMHO profitability will be worse though. They will contain to take on more and more investments - training, R&D, and 'sales and marketing' for enterprise share. They will face more competition than ever before. The range and types of models will continue to diversify, with real-time models for robotics and computer use paving the way for new entrants.
The investor confidence to back this doubling down will be spurred by rapid progress, panic deregulation while old-guard corporate democrats still hold power in the US, and much stricter sanctions against Chinese competitors, their models, and their eventual domestic hardware. The tech world will look a lot more factional by that point, with pseudo mergers & acquisitions amongst the big companies beginning, unabated by a toothless FTC. Anthropic will probably get closer to Amazon, and we could see big moves amongst Meta, Microsoft, and/or Google too.
Ultimately, Anthropic and OpenAI will maintain their market leading positions through at least 2028/2029 but only after a protracted period of significant equity dilution due to the expensive acquisitions and large outside investments which will be necessary in these market conditions.
Chinese hardware and models will look like a serious threat by the end of 2027 but will be largely locked out of US/Western markets. At the point the equation will have changed drastically - credit issues in the West led by mortgage security collapses will shape the investment environment and 2028 election, and rapid developments in physical technology and manufacturing will take the spotlight away from intelligence itself.
People lack the accounting know-how to talk about it correctly. But compute for inference is just not the main expense right now.
If they take on more funding for training and/or build out, then they have to share the profits with more people. The growing revenue is shared by more people. As more options vest, they have to share with more people. If they need to acquire external companies for particular capabilities, profit is shared. If they have to pay hand over fist for hyperscaler compute in 2027 - more profit shared.
So as their revenue grows rapidly, the fraction of that revenue which they've promised to other people is growing too - case in point, they operated at an $80B loss on paper this year because the unrealized future-profits promised to early investors is so valuable relative to current revenues. If they have to make more such deals to continue investment (e.g a low IPO price which require shelling even more shares to stay competitive), then they will continue to create vast value but less of that value will be captured by the current investors (the 'burning cash' may go to compute providers, outside investors, as discounts to enterprise and consumers, etc instead of to Anthropic).
This is pure speculation or can you give some facts showing where that growth could come from? Also if you burn more cash than you have revenue, at some point you will run out of it.
Growing from 7b revenue to 70b revenue is crazy % growth, but with 2026 being the tokenmaxxing year AND it being the year where models are finally 'good enough' for everyday work it sort of makes sense.
Growing 70b revenue to 700b revenue is, ah, a smidge tougher. Even getting $70b to $230b would be an immense success, methinks. The initial hypecycle seems to have cooled down slightly, and people are no longer ranked by how many tokens they can burn through with no real deliverables expected. There's even reports that companies are starting to tighten their belts when it comes to token spend; a catastrophe for fat US labs that need those big-wallet companies to spend API prices which pays for everything. Subscriptions might make them some money but it's not where the real profit is.
Which isn't to say that the revenue won't grow, or even that it'll contract, but I certainly don't see Anth growing to eye-watering $700b next year, especially as long as open-weights models keep getting released and a BigCo can't get Opus deployed to their own hardware sitting in their basement with no chance of the secret sauce leaking out to anybody, including Anthropic.
> Wait, Anthropic has $112 billion in cash right now and generates about $70B
Anthropic operates at a loss. They lose money.
That is what they keep having to raise money and are desperate to IPO as soon as possible.
> Why do you say they're running out of cash? Can we see your math?
My brother in Christ, you can't tell the difference between revenue and profit, and think that just because their revenue (not profits) grew 10x that they can grow now a "measly" 3x.
They might have enough money (ot whatever they consider to be equivalent) in hand to keep being unprofitable for a bit, but they alao have something in between half a trillion and a trillion of contracts for compute in the next few years.
AI doomers have predicted 1095 of the last 0 AI bubble pops.
Everyone is using more and more tokens, everyone is complaining that providers lobotomize flag ship models because they don't have the capacity to serve everyone.
Then they turn around and compare it to the dark fiber bust or railways or whatever when in each of those the issue was the capacity being build that no one used.
It's been 2 years since I could buy a gpu for love or money. It's been one for ram.
My old rig is worth ~two~ three[0] times what I paid for it inflation adjusted.
If tokens are being given away below cost, of course you're gonna use more. It's not gonna last. How many tokens would you spend on random experiments if they were much more expensive? If tokens are so valuable, why do I (and many others) get to use them on economically worthless things until I'm bored?
We just don't know the true value of tokens until those subsidies stop, and the subsidies don't stop because the respective companies don't want to know, at least until everyone cashed out. The valuations of AI companies are very difficult to reconcile with current revenue (which doesn't cover cost) and with a sober mind it's hard to imagine where all this additional revenue is going to come from, down the line. And that's not to say the the technology isn't fundamentally transformative and revolutionary. So was the internet, but it still created an investment bubble.
The RAM prices are in yet another pork cycle, which will be exacerbated by the fact that many previous pork cycles left RAM manufacturers with excess capacity and rock-bottom prices, so they're much more careful investing and more assertive negotiating contracts now.
I'm regularly using thousands of dollars worth of tokens for a flat fee of $200. Their API margins may be good, but not that good. This is below cost. Therefore, if tokens are so scarce, so valuable, in such high demand, why are they still handing out such generous packages? Clearly there must be enough excess capacity to run this advertising campaign, and clearly the advertising campaign must be deemed necessary to maintain the belief in massive future demand growth.
I'm also not saying that this subsidy is irrational, it does buy mindshare, but it still distorts the true demand for tokens. I wouldn't be using anywhere near as many tokens if the price was as high as the API costs, if anything I would switch to a cheaper model, and likely not an Anthropic one.
If you're gonna point to gross margins on the API, those 85% really are meaningless in isolation. They can charge top dollar because they have one of the top models, but only because they keep spending top dollar on training it, as does OpenAI. It's a game of chicken where whoever stops heavy training investments first is likely the one to lose market share to whoever keeps spending. So the play is to try and keep the investment capital flowing and be the last one standing, which only works if you can keep selling a fantasy of future revenue, not the dire reality of current revenue.
No, this whole thread has been a back and forth of pro AI(or at least pro anthropic) people arguing that the company is profitable because if you constrain your view to the last 2 quarters and just inference it looks that way while the other side keeps pointing out that you don’t get to do inference or the last two quarters without the massive cash spend occurring on training due to hardware and buildout.
These models become obsolete in months. Until the labs prove that they can produce a model that has a positive return on investment in it, it’s insane to think that the labs are in a long term good position.
Usually the markets react to a supply imbalance by increasing prices which destroys demand. For some reason token prices from the major players have instead been reduced.
All the money is being funneled towards nVidia. CoreWeave touches it briefly, and sends it to Jensen. Then nVidia user some of this cash to finance AI labs or clouds again. But the biggest losers seem to be VC's, that are on the basis of this piramid and still have to see a cent in return.
Please lets not out micro anywhere near Newton's laws of motion. Micro is nowhere near that level of theoretical soundness (even with mkdern lens of Newton being incorrect).
You can use Newton for most stuff and get pretty good answers. Micro is questionable in all kinds of ways. Assumptions about actors, assumptions about responses, assumptions about how utility maps onto monetary value. It's a huge mess really.
Subsidies cause excess demand that can't be met with supply, leading to shortages. Whatever financial wizardry Anthropic is doing to give themselves an apparent operating cost that is orders of magnitude lower than any ML/AI application I've ever seen, is a different question.
Edit: And Anthropic was never a specific point of discussion until you brought it up.
The money is there but not at the risk levels these companies have pushed things to. They're spending billions like it's candy, and treating millions like change they found in their couch cushions.
Financial systems are usually in the hands of people that have some risk management. When these labs say things like this will change everything, there's no way to know how to evaluate any of this.
Trillions have been lent. What you're saying is like eating 55 burgers, 55 pizzas, 55 tater tots and then complaining that the fridge is empty and nobody will give you more food.
CRWV is about to have an earnings report in less than 20 hours. It's probably best to just shut up for a day, wait for the earnings and then talk after that comes out.
Aside from the price signal, the fact they were issuing these at a 9.625% coupon rate continues to run up their future debt servicing costs.
Coreweave doesnt seem to be in a winner-take-all sector, they have a narrow moat and will have high ongoing capex. I struggle to see how they dont go under eventually even with the ability to restructure debt
Yeah, that coupon rate is wild! It is more than 450bps more than a similar Oracle bond. The simplest explanation: When this bond was issued, CoreWeave had a speculative/junk grade corporate credit rating: BB-. Surprisingly, Oracle also barely has investment grade credit rating: BBB-.
"Exodus Communications Inc., which never made a profit since it went public in March 1998, filed for bankruptcy protection after losing customers in the collapsing Internet economy."
"Exodus, which runs 44 Internet data centers on four continents, listed $5.98 billion in assets and $4.44 billion in debts in Chapter 11 papers filed Wednesday in U.S. Bankruptcy Court in Delaware. The company operates so-called server farms that provide Web site services for businesses."
"The Santa Clara, Calif.-based company’s losses increased more than tenfold in the second quarter, and it fired workers as it failed to win new contracts fast enough to replace those it lost. Exodus counted Yahoo Inc. and AMR Corp.’s American Airlines among its customers."
"Trading in Exodus shares was halted before the markets opened Wednesday. The shares fell 33 cents to 17 cents Tuesday and have plunged 99% from a record $89.81 in March 2000."
"Analysts say Exodus, with eight different junk bond offerings totaling about $3.4 billion, ran up more debt than it could handle."
“We sacrificed profitability in exchange for growth and market share, over-expanding in some areas in advance of demand, not anticipating the decline as the dot-com bubble burst and the economy weakened,” Krause said.
"Exodus’ second-quarter loss grew on acquisition and restructuring costs to $583.4 million, or $1.05 a share, from $51.8 million, or 13 cents, a year earlier. Sales rose 79% to $318.7 million but missed analyst estimates. The company lost $256 million on sales of $818.4 million in 2000."
That’s edging up on credit card rates for strong borrowers wild. NerdWallet has Chase Freedom with a lower bound of 18.49% and Citi Diamond Preferred with 16.74%.
CoreWeave pays 9.625%. But someone expected them to not be able to meet that so they sold what debt they had at discount to someone who now effectively gets 16% on amount they paid for.
Still, this means that people would only be willing to loan new money to them at that 16% rate.
Hyperscalers and SPVs are still planning on accelerating the buildout in 2027. Also even if everything hit the fan tomorrow, it will take time for production to shift back into consumer products if that's even what companies decide to shift to.
I don't the envisioned scenario is that a data centre company fails and Nvidia starts making RTX 5090s for consumers. Rather, one of these companies fails, their stock of enterprise GPUs gets flooded on the market, and suddenly the people/companies buying RTX 5090s for local inference can buy full fat enterprise GPUs, shifting demand.
If by enterprise you mean datacenter (which is what CoreWeave is), the "full fat" cards aren't as plug-and-play in a desktop PC as the consumer cards. They don't typically come with fans, and are meant to be cooled using some kind of holistic whole-rack (or larger) setup. Hobbyists do sometimes buy datacenter GPUs secondhand (and I have considered it, for the price/performance ratio) but it's kind of a lot of work and they tend to run hot.
There's probably still plenty of demand for that. At least someone in the executive team was floating around the idea of buying up some Nvidia cards to run AI models for the entire company somewhere locally. I don't think they'd be opposed to buying up data centre GPUs without fans if they suddenly flooded onto the market, even if that meant getting a rack chassis and a bunch of noisy server fans.
I've also seen someone sell their old fanless Nvidia cards on the local marketplace. There are definitely consumers running these things in their basements.
If a neocloud fails, it makes more sense to recapitalize it and leave the GPUs in place. If you sell the GPUs nobody will have enough power to run them.
This is a preview of what AI companies will face moving forward trying to raise capital. 16% is cheap if AI lives to its promise of massive returns. Problem is the ROI isn’t materializing and 16% would destroy most companies.
Meanwhile the traditional VC path is also drying up with OpenAI struggling with its latest round.
It’s going to get very ugly for the AI bros as cash starts to run out.
If the Anthropic valuation over the past week and the only partial OpenAI raise is anything to go by, the ocean is already actively receding.
I think we're going to find who was swimming the most naked by February at the rate this is going. Particularly the dedicated Neoclouds who had to fund by pure debt without alternate revenue streams. The problem is not the tech, it's the money.
They started at 9.6%. So it was a junk bond when it was issued. Now it's junkier.
Where is this bond in the bankruptcy hierarchy?
Is it secured by anything? GPUs, maybe. Is it subordinated behind other debt?
That's probably going to matter.
The yield is bounded from below by the treasury yield, which is just a little lower - and with the treasury you have really big problems if they default! But realistically the US will inflate its currency rather than default, so the yield is really just reflecting expected future inflation. It’s not as great a deal as it seems. Bonds are probably the most fairly priced assets in the world.
Fair could be construed as "the layers of market participants receive a share of the profits" . As far as I can tell there is essentially none left, and thats because folks will bid it down to 0 profit because they're lending others' money. (they're only responsible for their opportunity cost, but even 1% on others money is a good deal for the bank)
That bond has 24 years to maturity. If you look at US Treasury bond yields with about 25 years to maturity, the current yield is about 5.29%. That means the spread is less than 100bps. That is pretty small. Google (Alphabet) still has an excellent credit rating: AA+.
Most recent earnings show continued growth in queries and ad revenue. Phillip made clear we’ve not yet begun to see the full potential for ad placement in ai results. I’m very bullish long term Google.
There are discussions here on HN where Google is juicing search growth by simply raising prices or doing shady things like getting rid of exact match. This can raise revenue temporarily but will lose long term.
Total human search volume has likely already started to decline.
Further more, we are at a point where searching in ChatGPT or AI apps is simply better and faster than Google.
In this new world, Google is not the leader nor a monopoly.
> Google can go under and literally nothing will flinch.
Come, come: most email is Google, most phones are Google, most web browsers are Google. If you measured the global impact of any single company going under (with nothing else inherently changing), I wonder if Google might be #1.
Maybe it is in your world (and I have no idea where you borrowed that statistics). Personally I don't use any google product and them going under won't affect my life or life people I know to that matter.
Compare this with HSBC going under or say JPMorgan. Or Adyen. Or amazon (screw their retail business - it would be much better for everyone if they die, but unfortunately a lot of companies still build their IT on AWS).
I hadn’t thought about AWS. If going under meant that everything just disappeared one day, yeah, that would probably be enough to tip it to bigger global impact than Google. If it was an orderly collapse, I don’t feel I can assess it so well.
But the others you name: not a chance, big businesses already diversify such services. They don’t partner with one bank or one payment provider only.
The treasury is having daily rates of 4.56% at 2 year bond and 4.88% at 10 years rate.
Yes, Google has a very small risk but still it is a non zero risk that google can bankrupt whereas US govt can physically not bankrupt itself given that it can print the money (there could only be some gridlock within raising the ceiling though but that's true)
Also I would argue that Google itself has changed with trillions of dollars spent into AI/Datacenters, it simply can't pivot as well. There is a reason as to why Google is having such high level of interest rate
Google isn't stupid and wants to give such high level of interests if not for the amount of risk involved. I dont believe that there's free lunch but also that perhaps this can be a good deal to some and this is gonna push people even more away from the more riskier neoclouds funding and cause their debts to rise in interests rate even more.
Matt Levine once noticed that Microsoft’s bonds were actually cheaper than Treasuries (during a time of Trump-instigated turbulence).
His point was that Microsoft has an interest in maximizing its credit worthiness which necessitated the corporate governance etc, while the administration had an interest in maximizing… other stuff (?) which did not necessitate the same things.
Just last week I was on the call with a a techy architect from one of the major-major financial services companies, and he was discussing getting token capacity and his words on third best option was "we'd get it from Coreweave or whoever is still alive". So yeah, I am pretty sure that if that's the sentiment from software people at the major-major financial company, Coreweave is cooked.
So there seems to be a huge mismatch: demand hasn't waned, but the narrative around AI has changed in the last month.
There is no secret about the mismatch. Voters last week have elected people who are openly hostile to datacenters, and for some of them hostile to AI outright. The demand is there (at what price is a longer term debate). But if you can't power on new datacenters this doesn't do any good.
To give you an idea of the scale. US has currently about 50 GW of DC online. Depending on the source, you are looking at between 60 GW and 400 GW of new DC currently planned. For reference the total dependable capacity of the US grid is around 1000 GW.
I think there is a scenario where compute cost sky rockets while chip prices crash, as they end up unused in warehouse because of the lack of power and DC capacity.
Data centers can be build in Mexico or in a different friendlier to AI jurisdiction.
AI has been passively hostile to society. VC cash has too. Non of that is irrational.
You want them to embrace a tech that exacerbates inequality and its founders promote that?
I don't see the previous poster saying they "want" anything.
Every civilizational advance “exacerbates inequality” because smart people can use it to become more productive, but dumb people can’t.
By that logic we should ban books. This is Boko Haram ideology and you are just a communist who thinks he should be protected from competition.
Instead of “making everyone well off“ here is what will actually happen, because that’s what happens every time communists seize power: everyone will get poorer, because they will be outcompeted by more sane jurisdictions, eg China.
That sounds like “Threatening me with a good time”. It will be very cheap for me to buy my hardware. Unsurprisingly I have power at home, and I can use the free power I have during the day from solar to do all sorts of cool things on my own hardware. It will be like the 1980s and 90s again.
Of course there is still demand for RAM and SIMD compute!
But also hoarding hardware in glorified warehouses is not the genius business idea everyone thinks it is.
I'm assuming his friend's finance company is planning to make use of the cloud compute they find in some finance workload.
I don't think they're buying GPUs outright to store in a warehouse.
Any sort of hardware produced in the last 5 years has at least doubled in price. Some, like RAM are 3-5x more expensive.
So whoever hoarded it in warehouses could sell it now immediately for at least 100% profit.
They seem like geniuses to me now, worst case they make 100% profit on the investment.
But a large reason for the price increase has been because companies have been hoarding them in warehouses as they wait for DC buildout. If you sell then you remove that pressure and prices decrease.
It's only RAM, mass storage and GPUs. Desktop CPUs, mainboards, PSUs, and cases have become cheaper because PC building demand has declined. Peripheral prices have remained largely unchanged.
Naturally; it should be hoarded in my glorified homelab, where I can more readily listen to the sweet purr of fans.
Which is such a wild thing about our industry. These can be simultaneously true
- No one has available tokens/gpus/datacenter for sale at any price
- No one will lend people money to build it
This is how all capital-intensive utilities operate. The marginal cost of using a little more power or tokens is easy to justify because you can at any point stop using more.
But the cost to add more power capacity is huge and highly variable. Just because oil is expensive today doesn’t mean the capex to pump more is justified.
Really what AI needs is large customers that will commit to large spend over the next 5-10 years. Without this AI companies are basically offering duration arbitrage of electricity. Not a good place to be; the finances will always look bad.
Unfortunately, the amount of capital required by the entire AI industry is enough that counterparty risk is now a concern.
Anthropic and OpenAI are these customers, willing to buy the infrastructure. But they are running out of cash, too.
Wait, Anthropic has $112 billion in cash right now and generates about $70B in revenue per year and grew 10x YoY. If they succeed in raising $100B in cash in the IPO, which I believe they will, they'll have $212B. If their growth declines to 3x (down from 10x) in 2027, they'll still be at $230B in yearly revenue.
Further more, their compute commitments are over 10-20 years. Their average quarterly compute commitment is only about $11B - $15B which they have already surpassed with their revenue.
At $230b/year, they’d be at around $57b in quarterly revenue against compute of around $16b. This is not bleeding cash.
Why do you say they're running out of cash? Can we see your math?
The only reason Anthropic has cash is because VC keep giving them the money. They never had a positive quarter, with revenue always lower than expenses regardless of their growth, thus burning money from VC.
If VC money gets tight, they would have to issue debt, that pay interests. With a current negative EBITDA, their bond will be junk at IPO, paying higher interests and having a lot less market than AAA debt (e.g. Google has access to that market). Both things combined translate to "running out of cash".
They were cash flow positive for 2 quarters in a row.
Their S1 is public. We have AI. Are people this lazy? Throw it into ChatGPT and ask it to confirm the numbers.
> Are people this lazy?
It's not exactly laziness. It's that most people don't understand basic financial concepts like fixed versus operating costs.
Companies like Anthropic and OpenAI have invested huge sums in building/training models. Those are the headline numbers everyone talks about. What they don't understand is that the gross profit margins on AI model inference are very high.
The question is whether these companies ever get to the point where the profit from inference justifies (recoups) the investment made in training. If they have to keep investing vast sums in training new models, and/or competition causes the margins on inference to collapse, they're in trouble. But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
> What they don't understand is that the gross profit margins on AI model inference are very high.
Can we be sure of this? The second they tried to hike the prices to be profitable, they lost a lot of clients, or a lot of clients lowered they token expenditure. In any market margins tend to go to zero. I am using frontier models, but only because they are dirty cheap if not free. If they raise the price, I just switch to a cheaper model, start exploring the option of renting a GPU-VPS or buying my own card. Or reducing my AI usage, that currently includes stupid useless things like "translate this cooking recipe from french... for free".
> But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
They are being priced as the "ifs" are certainties, and not only a possibility. If anything goes a millimetre below perfect for them, their finances collapse. High margins attract competitors, and unless you have a huge moat (and their only moat is expensive training), your high margins vanish.
A few problems for Anthropic or OpenAI:
1. Model Build/Training costs are escalating if anything
2. They are in an arms race with competitors and can't afford to stop building new models - if they lose the perception of leading, they lose lots of customers overnight.
3. Inference margin is under attack from more specialised models and open weights models.*
4. They have no moat given how simple the interface to the products is - a bunch of text files and text prompts, and little prospect of building a moat.
So those all run counter the possibility of coasting on previous research. To me the most promising avenue for making money for these companies is to replace Google, not AGI or taking over the world's knowledge work. That would mean they are overvalued. Even coasting on previous research, how long would it take to pay off the money they have spent?
* https://strandsagents.com/blog/amazon-saves-millions-replaci...
> It's that most people don't understand basic financial concepts like fixed versus operating costs.
No man. It’s that, like you’ve explained further in your comment, they haven’t become net profitable yet. Being profitable in 1-2 quarters is a good trend but it doesn’t matter if you are still behind on mountains of debt or investors wanting their money back and your market is one where people talk about shit that happened 4 months ago as obsolete.
There is no proof that the current profit they are making on inference is durable.
> gross profit margins on AI model inference are very high.
the gross margins on growing potatoes is very high, if you exclude all the other costs needed to get the potatoes to market.
Think about it, you take some potatoes you had left over from last yeah, dump them in the ground and wait 9 months. Boom pure profit.
This of course excludes the cost of the land, machinery needed to plant, till and harvest, storage, pesticide, fungicide and herbicide.
Then the transport, cleaning and packaging.
If you arbitrarily cut out all the bits of the business needed to make revenue, everything is a profit.
Gross profit margins include all the direct costs of providing a service. So if gross profit margins of Anthropic are positive, they are positive after they subtracted the cost of electricity from revenues. And not just electricity, but every cost that is necessary to run a large scale inference operation.
> It's not exactly laziness. It's that most people don't understand basic financial concepts like fixed versus operating costs.
> Companies like Anthropic and OpenAI have invested huge sums in building/training models. Those are the headline numbers everyone talks about. What they don't understand is that the gross profit margins on AI model inference are very high.
Only when you ignore that the models they provide inference for have an extremely high cost and pretend they don't depreciate. They keep having to train new models. I don't think they recovered the cost for a model when a new one releases.
It is kind of cute to say other people "don't understand basic financial concepts like fixed versus operating costs" when you seem to not understand basic financial concepts such as depreciation and amortization.
Also, and this is important, we don't really know their true compute cost. We don't know if their compute with the likes of Google and Amazon were initially subsidized, and for how long it might be. In Q2 they also got to not pay SpaceX for compute for a while.
What's cute is that you're falling into the same trap of confusing accounting concepts.
Gross margin is revenue minus cost of revenue (COGS). COGS is basically the compute that serves inference (rented capacity, or depreciation on owned GPUs used for inference, plus expenses like support, etc.).
Model training is normally treated as R&D. Under US GAAP, R&D is expensed as it occurs and is below gross profit as an operating expense. It is not COGS.
You mention "depreciation" but depreciation applies to physical assets like GPUs. The GPU depreciation that belongs in gross margin is already in COGS. A trained model would be an intangible asset, so that is amortization, not depreciation. Even that only applies if the training cost was capitalized, and under GAAP it generally wouldn't be. The full cost is expensed when it's incurred. Nobody is "pretending models don't depreciate." Their cost is counted all at once but not in gross margin because that's simply not what gross margin measures.
Obviously, models lose value fast in an economic sense. They get replaced within a year or two and competition limits what they can earn. But that's an argument about whether the R&D expense is sustainable. It doesn't mean that gross margins are being misstated.
"Inference has high gross margins" and "training is extremely expensive" can both be true, which oddly seems to be something many people have trouble comprehending. But this is the crux of the questions that will determine whether these businesses are viable long-term:
1. Will training costs fall over time, or at least grow more slowly than revenue?
2. Does the gross profit a model earns from inference before it's replaced exceed what it cost to train?
3. Will inference margins hold up as competition and open-weight models push prices down?
You have a legitimate point about compute costs. If the deals with Google, Amazon or anyone else are priced below market, that directly inflates gross margin, because it lowers COGS. But we don't know those terms, so that's just speculation, and it's still a different argument from depreciation.
To reiterate what I wrote:
> The question is whether these companies ever get to the point where the profit from inference justifies (recoups) the investment made in training. If they have to keep investing vast sums in training new models, and/or competition causes the margins on inference to collapse, they're in trouble. But if training becomes less costly and inference margins don't compress too much, it's a totally different story.
> Model training is normally treated as R&D. Under US GAAP, R&D is expensed as it occurs and is below gross profit as an operating expense. It is not COGS.
Isn’t it plausible that this is conceptually wrong? What if a model’s revenue-generating potential only lasts for something like a year due to, loosely speaking, competition? Doesn’t this make model development more akin to, say, the costs of annual agricultural planting/harvesting/delivery cycles, or a tv network’s annual (back in the day at least…) production cycle for its main content?
> Nobody is "pretending models don't depreciate."
They are when they say that inference is incredibly profitable. Its incredible profits exist only when you pretend that models have 0 cost to be created (since they are a digital asset that can't depreciate), and R&D is a imaginary thing that exists only as a fixed cost of doing business that doesn't need to be recovered.
Anthropic released Opus 4.6 in February 2026 and deprecated it in June 2026. Was it anywhere close to recover the "R&D" costs during those few months?
More importantly, how long a model needs to be current until it recovers "R&D"? Their economic viability hangs on this, and the compute costs both to train and servr models.
A lot of people like to vaguely gesture at the possibility of compute costs coming down, but it is in fact coming up with the rising cost of hardware (and possibly with sweetheart compute deals being replaced by contacts where providers actually have to make money).
> They are when they say that inference is incredibly profitable. It is incredibly profits exist only when you pretend that models have 0 cost to be created (since they are a digital asset that can't depreciate), and R&D is a imaginary thing that exists only as a fixed cost of doing business that doesn't need to be recovered.
Expensing the cost of training a model isn't treating it as free. The whole cost is charged to the income statement the year it's incurred, which is actually faster than depreciation would ever recognize it. Calling something R&D does not make the cost "imaginary". It means the cost is counted. It's just not in gross margin.
We are actually asking the same basic question (can the investment made ever be recouped?). What you're struggling with is the accounting concepts and terminology. The costs of creating the models are a different part of the income statement than the costs of serving inference.
Again, "inference has high gross margins" and "training is extremely expensive" can both be true at the same time.
In absolutely none of my comments did I state that R&D never needs to be recovered. In fact, I stated repeatedly that the entire crux of whether these businesses are viable long-term comes down to whether the costs of creating the models is eventually recouped.
> Anthropic released Opus 4.6 in February 2026 and deprecated it in June 2026.
First off, Opus 4.6 is still available, even though Anthropic recommends migrating to 5.5. Google Cloud says it will be on offer until February 2027. So revenue for this model didn't drop to $0 overnight.
https://platform.claude.com/docs/en/models/opus-4-6/overview
More importantly, your comment implies that models cease to have value when the next version is released. This completely ignores the reality of model training, which is that the subsequent models almost certainly built on the research, data and techniques of 4.6. And 4.6 was built on the research, data and techniques of the models that came before it.
So the whole notion that point versions of these models are standalone investments that cost billions to train and then become worthless the minute a new point version is released is incredibly naive. It's just not how model training works.
> A lot of people like to vaguely gesture at the possibility of compute costs coming down, but it is in fact coming up with the rising cost of hardware (and possibly with sweetheart compute deals being replaced by contacts where providers actually have to make money).
You have absolutely no evidence of "sweetheart compute deals" so this is pure speculation.
Furthermore, there's a flaw in your logic. It's entirely possible for chips and memory to get more expensive while the cost per token served falls due to more efficient hardware and better inference software. What matters here is the cost per unit of output, not the price of a GPU.
Bottom line: inference has high gross margins and these businesses have had to invest massive sums. If inference profits grow sufficiently faster than costs, or costs decrease sufficiently over time, these businesses could be viable and perhaps incredibly successful. If that doesn't happen, their viability is questionable.
Lots of people have an opinion about this but as this discussion proves, lots of people also don't understand the basic accounting required to make a cogent argument one way or the other.
I just did that. It says that’s not true.
It is true. Try Opus 5.5.
Ha! Calling people "lazy" for not using ChatGPT? I can read an income statement, is not that difficult, no need for ChatGPT. This is the S1, ammended:
https://www.sec.gov/Archives/edgar/data/2130357/000162828026...
Income at page F-7 (about 350 overall):
* In the case you are confused, () in finances means negative values.Q-3 only, this is calculated substracting first half from 9 first months:
Cashflow might be confusing here, because the cashflow table include huge items like "Change in fair value of convertible notes, derivative liabilities, and SAFEs" (translates to "we have paper valued at $X, but now it is valued at $Y", but it is still paper) or "Proceeds from issuance of redeemable convertible preferred stock, net of issuance cost" (translates to "we sold paper to VCs for $X") that are not business or recurrent items, or at least I hope Anthropic don't plan on "earning" money by selling paper to VCs forever.Operating income and cash flow are different numbers.
The problem with AI is that it's still dependent on the skill of the prompter.
> The problem with AI is that it's still dependent on the skill of the prompter.
Ah so when people don’t have the same outlook as you we’ve gone from being lazy to now it’s a skill issue.
No, but facts are facts. Outlooks are not facts.
Oh and the footnote about future liabilities, strongly adding to compute and infra expenses.
> Their S1 is public.
Their S1 was taken down. As of today (https://www.reuters.com/world/china/global-markets-view-euro...) This is the second time they changed it.
Apparently they file it with Claude.
Those figures are “earnings before bad things”.
What are the bad things?
All the debt they have accrued?
What debt?
Non-cancellable liabilities == debt. But it's just 900B$ so forget about it...
Small gripe but I don't like 'declines to 3x'. 3x growth would mean repeating this year's record growth twice over, which is smaller relatively but an increase in absolute growth rate.
So what do you think their growth YoY will be in 2027?
I have seen most revenue projections are on the order of 2-3x, but I suspect about 4-5x.
IMHO profitability will be worse though. They will contain to take on more and more investments - training, R&D, and 'sales and marketing' for enterprise share. They will face more competition than ever before. The range and types of models will continue to diversify, with real-time models for robotics and computer use paving the way for new entrants.
The investor confidence to back this doubling down will be spurred by rapid progress, panic deregulation while old-guard corporate democrats still hold power in the US, and much stricter sanctions against Chinese competitors, their models, and their eventual domestic hardware. The tech world will look a lot more factional by that point, with pseudo mergers & acquisitions amongst the big companies beginning, unabated by a toothless FTC. Anthropic will probably get closer to Amazon, and we could see big moves amongst Meta, Microsoft, and/or Google too.
Ultimately, Anthropic and OpenAI will maintain their market leading positions through at least 2028/2029 but only after a protracted period of significant equity dilution due to the expensive acquisitions and large outside investments which will be necessary in these market conditions.
Chinese hardware and models will look like a serious threat by the end of 2027 but will be largely locked out of US/Western markets. At the point the equation will have changed drastically - credit issues in the West led by mortgage security collapses will shape the investment environment and 2028 election, and rapid developments in physical technology and manufacturing will take the spotlight away from intelligence itself.
If 5x, then they'll end Q3 2027 with about $350B annualized revenue.
Looking at their 2027 compute commitments, it looks like around $72B total.
So I don't understand where this "burning cash" claim comes from by a few people here.
I'm going to lock in $325B.
People lack the accounting know-how to talk about it correctly. But compute for inference is just not the main expense right now.
If they take on more funding for training and/or build out, then they have to share the profits with more people. The growing revenue is shared by more people. As more options vest, they have to share with more people. If they need to acquire external companies for particular capabilities, profit is shared. If they have to pay hand over fist for hyperscaler compute in 2027 - more profit shared.
So as their revenue grows rapidly, the fraction of that revenue which they've promised to other people is growing too - case in point, they operated at an $80B loss on paper this year because the unrealized future-profits promised to early investors is so valuable relative to current revenues. If they have to make more such deals to continue investment (e.g a low IPO price which require shelling even more shares to stay competitive), then they will continue to create vast value but less of that value will be captured by the current investors (the 'burning cash' may go to compute providers, outside investors, as discounts to enterprise and consumers, etc instead of to Anthropic).
This is pure speculation or can you give some facts showing where that growth could come from? Also if you burn more cash than you have revenue, at some point you will run out of it.
they also have $1T in commitments
Yes, over 10-20 years and I'm guessing many of those contracts are just intents and not hard commitments.
Let's say it's 15 years. That's $16B/quarter in commitment. They had $17B in revenue in Q3 2026.
If anything, they're far too conservative.
Most of their compute commitments are non-cancellable.
Based on how their growth is going, I think they're far too conservative and they should sign more.
now do the math with profit instead of revenue
most recently (on both a quarterly and annual basis), it was negative
Why with profit instead of revenue?
Their revenue pays for the compute.
Their revenue pays for expenses. If profit is negative, revenue doesn't cover their expenses.
They can't just turn off all other expenses and just pay for compute.
Their revenue does cover their expenses - at least for last 2 quarters.
Growing from 7b revenue to 70b revenue is crazy % growth, but with 2026 being the tokenmaxxing year AND it being the year where models are finally 'good enough' for everyday work it sort of makes sense.
Growing 70b revenue to 700b revenue is, ah, a smidge tougher. Even getting $70b to $230b would be an immense success, methinks. The initial hypecycle seems to have cooled down slightly, and people are no longer ranked by how many tokens they can burn through with no real deliverables expected. There's even reports that companies are starting to tighten their belts when it comes to token spend; a catastrophe for fat US labs that need those big-wallet companies to spend API prices which pays for everything. Subscriptions might make them some money but it's not where the real profit is.
Which isn't to say that the revenue won't grow, or even that it'll contract, but I certainly don't see Anth growing to eye-watering $700b next year, especially as long as open-weights models keep getting released and a BigCo can't get Opus deployed to their own hardware sitting in their basement with no chance of the secret sauce leaking out to anybody, including Anthropic.
> Wait, Anthropic has $112 billion in cash right now and generates about $70B
Anthropic operates at a loss. They lose money.
That is what they keep having to raise money and are desperate to IPO as soon as possible.
> Why do you say they're running out of cash? Can we see your math?
My brother in Christ, you can't tell the difference between revenue and profit, and think that just because their revenue (not profits) grew 10x that they can grow now a "measly" 3x.
They might have enough money (ot whatever they consider to be equivalent) in hand to keep being unprofitable for a bit, but they alao have something in between half a trillion and a trillion of contracts for compute in the next few years.
AI doomers have predicted 1095 of the last 0 AI bubble pops.
Everyone is using more and more tokens, everyone is complaining that providers lobotomize flag ship models because they don't have the capacity to serve everyone.
Then they turn around and compare it to the dark fiber bust or railways or whatever when in each of those the issue was the capacity being build that no one used.
It's been 2 years since I could buy a gpu for love or money. It's been one for ram.
My old rig is worth ~two~ three[0] times what I paid for it inflation adjusted.
[0] fuck me ram is rising again.
AI is a bubble that will pop. No one really can say when Those both things can be true and second one doesn't invalidate the first one.
If tokens are being given away below cost, of course you're gonna use more. It's not gonna last. How many tokens would you spend on random experiments if they were much more expensive? If tokens are so valuable, why do I (and many others) get to use them on economically worthless things until I'm bored?
We just don't know the true value of tokens until those subsidies stop, and the subsidies don't stop because the respective companies don't want to know, at least until everyone cashed out. The valuations of AI companies are very difficult to reconcile with current revenue (which doesn't cover cost) and with a sober mind it's hard to imagine where all this additional revenue is going to come from, down the line. And that's not to say the the technology isn't fundamentally transformative and revolutionary. So was the internet, but it still created an investment bubble.
The RAM prices are in yet another pork cycle, which will be exacerbated by the fact that many previous pork cycles left RAM manufacturers with excess capacity and rock-bottom prices, so they're much more careful investing and more assertive negotiating contracts now.
Tokens are not given below cost: https://www.indmoney.com/blog/us-stocks/anthropic-gross-marg...
I'm regularly using thousands of dollars worth of tokens for a flat fee of $200. Their API margins may be good, but not that good. This is below cost. Therefore, if tokens are so scarce, so valuable, in such high demand, why are they still handing out such generous packages? Clearly there must be enough excess capacity to run this advertising campaign, and clearly the advertising campaign must be deemed necessary to maintain the belief in massive future demand growth.
85% gross margins.
does that take into account model development costs?
That is rather a big move of the goal posts. The previous post was that inference was unprofitable which is simply not true.
I didn't say inference through the API was unprofitable, I said token generation is heavily subsidized, artificially inflating demand.
Semianalysis estimates that subscriptions make up 40% of tokens generated for only the 10% of the revenue:
https://newsletter.semianalysis.com/p/anthropic-subscription...
I'm also not saying that this subsidy is irrational, it does buy mindshare, but it still distorts the true demand for tokens. I wouldn't be using anywhere near as many tokens if the price was as high as the API costs, if anything I would switch to a cheaper model, and likely not an Anthropic one.
If you're gonna point to gross margins on the API, those 85% really are meaningless in isolation. They can charge top dollar because they have one of the top models, but only because they keep spending top dollar on training it, as does OpenAI. It's a game of chicken where whoever stops heavy training investments first is likely the one to lose market share to whoever keeps spending. So the play is to try and keep the investment capital flowing and be the last one standing, which only works if you can keep selling a fantasy of future revenue, not the dire reality of current revenue.
No, this whole thread has been a back and forth of pro AI(or at least pro anthropic) people arguing that the company is profitable because if you constrain your view to the last 2 quarters and just inference it looks that way while the other side keeps pointing out that you don’t get to do inference or the last two quarters without the massive cash spend occurring on training due to hardware and buildout.
These models become obsolete in months. Until the labs prove that they can produce a model that has a positive return on investment in it, it’s insane to think that the labs are in a long term good position.
Usually the markets react to a supply imbalance by increasing prices which destroys demand. For some reason token prices from the major players have instead been reduced.
That is precisely what price subsidies look like. These companies are hiding the the true cost of offering these models.
They're not subsidizing enough for CoreWeave to make money.
All the money is being funneled towards nVidia. CoreWeave touches it briefly, and sends it to Jensen. Then nVidia user some of this cash to finance AI labs or clouds again. But the biggest losers seem to be VC's, that are on the basis of this piramid and still have to see a cent in return.
Source?
Literally any microeconomics class. That's like asking for a source when someone discusses Newton's laws of motion.
Please lets not out micro anywhere near Newton's laws of motion. Micro is nowhere near that level of theoretical soundness (even with mkdern lens of Newton being incorrect).
You can use Newton for most stuff and get pretty good answers. Micro is questionable in all kinds of ways. Assumptions about actors, assumptions about responses, assumptions about how utility maps onto monetary value. It's a huge mess really.
How does taking a microeconomics class prove that Anthropic is subsidizing the costs?
Subsidies cause excess demand that can't be met with supply, leading to shortages. Whatever financial wizardry Anthropic is doing to give themselves an apparent operating cost that is orders of magnitude lower than any ML/AI application I've ever seen, is a different question.
Edit: And Anthropic was never a specific point of discussion until you brought it up.
> No one will lend people money to build it
The money is there but not at the risk levels these companies have pushed things to. They're spending billions like it's candy, and treating millions like change they found in their couch cushions.
Financial systems are usually in the hands of people that have some risk management. When these labs say things like this will change everything, there's no way to know how to evaluate any of this.
Trillions have been lent. What you're saying is like eating 55 burgers, 55 pizzas, 55 tater tots and then complaining that the fridge is empty and nobody will give you more food.
CRWV is about to have an earnings report in less than 20 hours. It's probably best to just shut up for a day, wait for the earnings and then talk after that comes out.
The circular matrioshka financing is so massive the earnings reports of these cos are just a potemkin village
Aside from the price signal, the fact they were issuing these at a 9.625% coupon rate continues to run up their future debt servicing costs.
Coreweave doesnt seem to be in a winner-take-all sector, they have a narrow moat and will have high ongoing capex. I struggle to see how they dont go under eventually even with the ability to restructure debt
Yeah, that coupon rate is wild! It is more than 450bps more than a similar Oracle bond. The simplest explanation: When this bond was issued, CoreWeave had a speculative/junk grade corporate credit rating: BB-. Surprisingly, Oracle also barely has investment grade credit rating: BBB-.
You can see precise details about the CoreWeave bond issue here: https://bondblox.com/bond-market/Coreweave-Inc-USU2069EAE06
CoreWeave is the Exodus Communications of this bubble party.
[Sept 2001] https://www.latimes.com/archives/la-xpm-2001-sep-27-fi-50296...
"Exodus Communications Inc., which never made a profit since it went public in March 1998, filed for bankruptcy protection after losing customers in the collapsing Internet economy."
"Exodus, which runs 44 Internet data centers on four continents, listed $5.98 billion in assets and $4.44 billion in debts in Chapter 11 papers filed Wednesday in U.S. Bankruptcy Court in Delaware. The company operates so-called server farms that provide Web site services for businesses."
"The Santa Clara, Calif.-based company’s losses increased more than tenfold in the second quarter, and it fired workers as it failed to win new contracts fast enough to replace those it lost. Exodus counted Yahoo Inc. and AMR Corp.’s American Airlines among its customers."
"Trading in Exodus shares was halted before the markets opened Wednesday. The shares fell 33 cents to 17 cents Tuesday and have plunged 99% from a record $89.81 in March 2000."
"Analysts say Exodus, with eight different junk bond offerings totaling about $3.4 billion, ran up more debt than it could handle."
“We sacrificed profitability in exchange for growth and market share, over-expanding in some areas in advance of demand, not anticipating the decline as the dot-com bubble burst and the economy weakened,” Krause said.
"Exodus’ second-quarter loss grew on acquisition and restructuring costs to $583.4 million, or $1.05 a share, from $51.8 million, or 13 cents, a year earlier. Sales rose 79% to $318.7 million but missed analyst estimates. The company lost $256 million on sales of $818.4 million in 2000."
That’s edging up on credit card rates for strong borrowers wild. NerdWallet has Chase Freedom with a lower bound of 18.49% and Citi Diamond Preferred with 16.74%.
https://www.nerdwallet.com/credit-cards/best
it's pretty well accepted that if you ever actually pay the credit card rate, you're doing it wrong
Indeed, that's what's crazy about CoreWeave's bond rate being so high. They're actually paying it?!
CoreWeave pays 9.625%. But someone expected them to not be able to meet that so they sold what debt they had at discount to someone who now effectively gets 16% on amount they paid for.
Still, this means that people would only be willing to loan new money to them at that 16% rate.
Unless they issued the bonds with a call option, yes.
So there’s a chance I can pickup some discounted GPUs next year?
Zero chance. Secondary market for enterprise grade GPUs is white hot.
Next year is a long way away.
Less than two months?
There are weeks when decades happen.
Hyperscalers and SPVs are still planning on accelerating the buildout in 2027. Also even if everything hit the fan tomorrow, it will take time for production to shift back into consumer products if that's even what companies decide to shift to.
I don't the envisioned scenario is that a data centre company fails and Nvidia starts making RTX 5090s for consumers. Rather, one of these companies fails, their stock of enterprise GPUs gets flooded on the market, and suddenly the people/companies buying RTX 5090s for local inference can buy full fat enterprise GPUs, shifting demand.
If by enterprise you mean datacenter (which is what CoreWeave is), the "full fat" cards aren't as plug-and-play in a desktop PC as the consumer cards. They don't typically come with fans, and are meant to be cooled using some kind of holistic whole-rack (or larger) setup. Hobbyists do sometimes buy datacenter GPUs secondhand (and I have considered it, for the price/performance ratio) but it's kind of a lot of work and they tend to run hot.
> [...] but it's kind of a lot of work and they tend to run hot.
Anyone that can open their computer to install an SSD, or a consumer-grade GPU, can install a data center GPU with a fan-kit; it's not rocket science.
There's probably still plenty of demand for that. At least someone in the executive team was floating around the idea of buying up some Nvidia cards to run AI models for the entire company somewhere locally. I don't think they'd be opposed to buying up data centre GPUs without fans if they suddenly flooded onto the market, even if that meant getting a rack chassis and a bunch of noisy server fans.
I've also seen someone sell their old fanless Nvidia cards on the local marketplace. There are definitely consumers running these things in their basements.
If a neocloud fails, it makes more sense to recapitalize it and leave the GPUs in place. If you sell the GPUs nobody will have enough power to run them.
If you sell the GPUs, they will get dispersed all over the world.
GPUs allow you to buy becoming more powerful. They are going to continue to be sought after.
No, the other neoclouds will scoop up the remains. The H100 was released 4 years ago and its still near impossible to find at any discounted price.
Nop. The hyperscalers will gobble everything available, including any neo cloud that faces financial troubles
If you can out-bid the Chinese.
One of the cras published a report on the potential upcoming AI stress and it did highlight a bunch of vulnerabilities.
Just checked and coreweave rated at bb-
This is a preview of what AI companies will face moving forward trying to raise capital. 16% is cheap if AI lives to its promise of massive returns. Problem is the ROI isn’t materializing and 16% would destroy most companies.
Meanwhile the traditional VC path is also drying up with OpenAI struggling with its latest round.
It’s going to get very ugly for the AI bros as cash starts to run out.
I don't think any of this matters. The only thing that matters is demand and supply.
If the Anthropic valuation over the past week and the only partial OpenAI raise is anything to go by, the ocean is already actively receding.
I think we're going to find who was swimming the most naked by February at the rate this is going. Particularly the dedicated Neoclouds who had to fund by pure debt without alternate revenue streams. The problem is not the tech, it's the money.
They started at 9.6%. So it was a junk bond when it was issued. Now it's junkier.
Where is this bond in the bankruptcy hierarchy? Is it secured by anything? GPUs, maybe. Is it subordinated behind other debt? That's probably going to matter.
Fraudulent circular accounting is so very entertaining.
No Chinese honey traps for the Coreweave CEO for the foreseeable future
It almost reads like a semiconductor process report.
totally crazy that Google has bonds yielding 5.68%
That's a great return for practically zero risk.
(If google defaults on bonds there's much bigger problems in the world than the loss of your capital)
The yield is bounded from below by the treasury yield, which is just a little lower - and with the treasury you have really big problems if they default! But realistically the US will inflate its currency rather than default, so the yield is really just reflecting expected future inflation. It’s not as great a deal as it seems. Bonds are probably the most fairly priced assets in the world.
I'm old enough to remember Argentina issue a 100 year bond with an 8% yield.
NEVER buy an Argentinian bond at full price. always wait until we are burning stuff up. don't believe any argentinian president, ever.
Was it denominated in USD or ARS?
$2.75 billion in U.S. dollar-denominated sovereign debt.
https://www.economist.com/finance-and-economics/2017/06/24/i...
I’m old enough to remember that they defaulted on those bonds three years later.
If by fairly priced you mean so competitively priced that essentially all the real profit has been taken out.
What else would fairly priced mean?
Fair could be construed as "the layers of market participants receive a share of the profits" . As far as I can tell there is essentially none left, and thats because folks will bid it down to 0 profit because they're lending others' money. (they're only responsible for their opportunity cost, but even 1% on others money is a good deal for the bank)
I cannot find the exact bond that you mention. However, I found this one: https://www.bondsupermart.com/bsm/bond-factsheet/US02079KAF4...
GOOGL 2.050% 15Aug2050 Corp
Current Bid/Ask yield: 6.134/6.045
That bond has 24 years to maturity. If you look at US Treasury bond yields with about 25 years to maturity, the current yield is about 5.29%. That means the spread is less than 100bps. That is pretty small. Google (Alphabet) still has an excellent credit rating: AA+.
So much low-risk debt has been issued by the likes of Google, Amazon, etc. that it’s hard for other debt issuers to compete.
US 10y bonds are at 4.871, which is lower but fits your description even better.
That being said, inflation is at 3.7% YoY and it's the real rate that maters, not the nominal one.
I disagree. I think Google Search is in huge trouble. Most of their profits come from it.
Most recent earnings show continued growth in queries and ad revenue. Phillip made clear we’ve not yet begun to see the full potential for ad placement in ai results. I’m very bullish long term Google.
There are discussions here on HN where Google is juicing search growth by simply raising prices or doing shady things like getting rid of exact match. This can raise revenue temporarily but will lose long term.
Total human search volume has likely already started to decline.
Further more, we are at a point where searching in ChatGPT or AI apps is simply better and faster than Google.
In this new world, Google is not the leader nor a monopoly.
> Total human search volume has likely already started to decline.
source?
Here’s one: https://ppc.land/ziff-davis-loses-37-of-its-google-search-tr...
Because Google doesn’t break down search volume, we don’t know for sure.
ZIRP days are over
Google can go under and literally nothing will flinch. I don't think it will cause any trouble on my side, let alone bigger on then loss of capital.
> Google can go under and literally nothing will flinch.
Come, come: most email is Google, most phones are Google, most web browsers are Google. If you measured the global impact of any single company going under (with nothing else inherently changing), I wonder if Google might be #1.
Maybe it is in your world (and I have no idea where you borrowed that statistics). Personally I don't use any google product and them going under won't affect my life or life people I know to that matter.
Compare this with HSBC going under or say JPMorgan. Or Adyen. Or amazon (screw their retail business - it would be much better for everyone if they die, but unfortunately a lot of companies still build their IT on AWS).
I hadn’t thought about AWS. If going under meant that everything just disappeared one day, yeah, that would probably be enough to tip it to bigger global impact than Google. If it was an orderly collapse, I don’t feel I can assess it so well.
But the others you name: not a chance, big businesses already diversify such services. They don’t partner with one bank or one payment provider only.
13 million businesses use their business offering, 3 billion users losing email/productivity apps would be a noticeable dent in the world (https://sqmagazine.co.uk/google-workspace-statistics/)
The treasury is having daily rates of 4.56% at 2 year bond and 4.88% at 10 years rate.
Yes, Google has a very small risk but still it is a non zero risk that google can bankrupt whereas US govt can physically not bankrupt itself given that it can print the money (there could only be some gridlock within raising the ceiling though but that's true)
Also I would argue that Google itself has changed with trillions of dollars spent into AI/Datacenters, it simply can't pivot as well. There is a reason as to why Google is having such high level of interest rate
Google isn't stupid and wants to give such high level of interests if not for the amount of risk involved. I dont believe that there's free lunch but also that perhaps this can be a good deal to some and this is gonna push people even more away from the more riskier neoclouds funding and cause their debts to rise in interests rate even more.
Matt Levine once noticed that Microsoft’s bonds were actually cheaper than Treasuries (during a time of Trump-instigated turbulence).
His point was that Microsoft has an interest in maximizing its credit worthiness which necessitated the corporate governance etc, while the administration had an interest in maximizing… other stuff (?) which did not necessitate the same things.
[1] https://www.wsj.com/finance/investing/why-microsoft-has-lowe...
Now let's see Anthropic sell enough of its junk bonds to cover its half a billion in non-cancelable obligations.