There is just no way all these data center investments in the trillions pay off. That has to be paid out of cashflow, like, real profit. The price to do useful things keeps falling, the payroll economy will crash long before there's actual trillions of dollars of cashflow for tokens.
It is the entire scam and many of those funding the data center build-out know this. Otherwise why are they hiding the trillions of debt under the rug?
There's a reason why a company like Stripe can stay private far longer than Anthropic or OpenAI can.
These AI companies have taken in all the capital from private investors and are still losing hundreds of billions and have no choice but to hype up the IPO and dump some of the stock at a purposefully inflated valuation to retail investors.
They are worth (on paper) so much that there are not enough retail money to buy them anymore. All these companies can do is put Uncle Sam on the hook to print money for them. There is no other way.
They just need 5% of the worlds population to get $50-100 value per month out of them.
Even in my non-SWE job, paying $100/mo for my current $20/mo plan would still be a no-brainer.
I don't think there is much concern about open models either. Compute is constrained for the foreseeable future, and money is what will determine who gets it. Nevermind that the US will likely block Chinese model imports or China will block exports at some point. The cold war has already begun here.
I highly doubt 415 million people will find enough reason to purchase $100 worth of Anthropic, especially when the price of intelligence keeps going down and smaller models become more and more capable to meet the average person's needs like drafting emails, customer support, basic RAG.
The cost of intelligence doesn't matter, it will just make margins wider.
Just look at software over the last 20 years. People pay based on the value they receive, not the cost the produce or serve it. That fact is literally is the backbone of tech, and why it has been an absolute money machine.
I think the worst case scenario for the labs is current (or next gen) SoTA models reaching a point where cheap consumer hardware can fully run them. But the labs practically have a monopsony on compute, and getting the kind of long context current models thrive on out of 16GB GDDR6 is gonna be a trick.
Tech has been a money machine because it has a marginal cost if approximately 0, so tech companies could literally give there products away and live off of the pennies they get from serving adds. Software is one of the hardest product classes to get people to pay for because the cost is anchored at 0.
Why that many people per month? What timeframe were you considering for them to pay off their expenditure? For that matter what are you estimating their total expenditure to be?
Enterprises will pay tens of millions a month, millions of individuals will pay $100 a month and there will be a long tail as they offer cheaper pricing and perhaps ad-supported pricing.
The average American family won’t be willing to pay more than a Netflix subscription.
I think in five years, it will only be power users that use a model in its raw form - everyone else will mostly consume using wrapper apps.
Token cost is falling rapidly for a given quality. We don't think about this too much as newer models have made legacy apps obsolete - but its an interesting question what the cheapest text to sql or similar model would be. I still go by 10x cost decrease per 6 months for any given model quality.
There's 5B working age people world wide and only 50% of those make more than $500/mo where it's even conceivable to spend $50/mo. I did the napkin math, it can still sorta kinda work out.
I think more realistically we'll have something like the Google/social media US-vs-world profit split of 40-50% US vs rest of world combined. Even those numbers can work out but then I don't see tremendous growth.
But the sector valuation is already priced for wholesale workforce replacement or massively expanded productivity and AI platform providers taking a lot of that pie for themselves.
With corporate profits already near all-time highs with respect to GDP, who is going to buy all those new products (from expanded productivity ) if all the gains only go to OIA and Anthropic employees?
I would be curious to see if they ever publish detailed statistics on this. I'm sure as others have said the average family will not be paying much if anything for AI. Just within the HN bubble I have been paying a bit for it just for my hobbies and it's been fun, enlightening, incredibly useful for rewriting other peoples code and asking it all the dumb questions that I would get entirely roasted for here. Curious to know how many others are using it that way for hobbies, silly questions, rewriting other peoples code, finding and fixing vulnerabilities, debugging performance bottlenecks, etc... rather than strictly professional use cases.
betting on a compute bottleneck sounds like a recipe to get thrashed when the bottleneck relieves itself.
At the investment scales being discussed, CUDA/architecture and other advantages do not matter - you could spend 1 billion on building a new chip architecture. The ram/fab inputs have been a commodity market for years. Heck, even the model bottleneck doesn't seem real when it's only 1-4 billion or less to get a state of the art model.
At some point the compute bottleneck will be relieved, you can see NVidia hedging their strategy with both open models and on-device chips targeted for local inference. The 200 dollar a month plan will absolutely be taken over by local hardware in the future.
Dario Amodei has apparently recently suggested that Anthropic might become only only private AI company in the entire world, which obviously it won't.
There is competition everywhere, and it is intensifying and catching up, not fading away. Open weight models are becoming more common, both within the US as well as elsewhere. Treasury secretary Scott Bessent just praised Meta's open weight models.
There is demand for AI at all different price points, and as all models at all price points become more capable, it seems that increasingly developers are seeing the most expensive ones as specialized tools, not daily drivers.
Compute/memory may be constrained for a few years until production capacity catches up, but this does not mean that demand for cheaper and open weight models will go away, else it would already be happening. Anthropic would like to sell an expensive Ferrari to everyone on the planet, but 99.99% of those people have no need for anything more than a Yugo.
No - but we already have cheap LLMs priced way below frontier models. This is not the housing market. There will always be someone willing to take a lower profit margin for a slice of the pie, and of course smaller models are cheaper to serve so can afford to be cheaper.
DeepSeek recently said that their super-low pricing let's them recoup the cost of the hardware it runs on in 10 months, so there is evidentially plenty of profit to be had over a projected 3+ year lifespan of a "GPU".
Some in the AI industry, or breathing the same air (Dwarkesh) project that limited GPUs will only be used to serve the most expensive models with the highest profit margins, but it is just not what we are seeing. If the only LLMs available were ones at Opus/Fable price points then the GPU scarcity would disappear since the demand at that price is just not there. It's remarkably like trying to fill all the seats on a plane - you can fill a few at 1st class prices, but most of the plane better be coach if you want to sell all the seats.
For a GPU, "selling all the seats", keeping it busy 24x7, is critical to profitability since the primary cost to serving is the GPU which has a limited lifespan.
There's a huge range of cost-per-task variation across models, and the capability of the smaller cheaper models keeps increasing.
For example, here we have Fable 5 at $3.14/task vs Kimi K3 at $0.84/task, with very little difference between them in coding capability (and this isn't even a coding/agentic fine tune of Kimi).
We now have models like Qwen 3.8 27B, small enough to run locally, with coding capability similar to Opus 4.5 based on challenging tasks like the Anthropic Kernel challenge.
I think we are rapidly getting to the "good enough" stage of LLMs, just like we did long ago with PCs. A cheap PC/LLM is all you need for 99.9% of normal use cases. Maybe nothing can touch whatever latest greatest models Anthropic and OpenAI have when it comes to solving Erdos problems, but most developers are working on problems more like the Anthropic Kernel challenge in complexity (or in fact typically way simpler ones).
Sure, they "just" need to exceed Netflix's global subscriber count, at a significant multiple of the monthly cost of Netflix in their most expensive markets, all in time for an IPO that is rumored to be happening later this year. No problem.
So thinking this through to come up with your numbers, total investment through 2027 is ~$2T and interest on debt is over 7%, which makes servicing this ~$140B/yr. However, failure rates on H100-B300 installed HW have been over 12%/yr even as the power and cost efficiency per token of the later builds has risen ~5x. So depreciation on the data centers is conservatively $240B/yr ignoring power costs (likely only $30B/yr at $0.05/kWh). Conveniently, this is $20B/month and if 5% of the population uses it, that's only 400million people so it's $50/mo or $600/yr, only if the AI-vendors make zero profit and $100/mo if they make 50% margins. Realistically, the OpenAI and Anthropic go to zero or it's $600/year.
I don't know where the 5% of world population came from, because that's clearly not just professionals or people making a lot of money. That's Uber drivers, and retirees in the developed world or tech workers in Asia making <$10000/year. Those don't look like great markets. This needs to be 2x higher value than their cell phone and internet that they might spend $300/year on today (that's a new iPhone every 3 years on an ATT plan). It's not like it can replace their plan, because they need that connectivity to use it!
Who's getting this value other than SWEs? There aren't 40 million SWEs and I don't see them spending over $6000/year. If their business does, it still has to pass on the cost to consumers and/or fire SWEs.
I did some napkin math in a comment a little while ago, that if the whole shebang comes to a screeching hard stop where all these investments are written down to 0 and all AI revenue disappears completely, these trillions of debt could be repaid, with interest, by the hyperscalers with their pre-AI firehoses of cash flow in 6 - 8 years. It’s never going to be that simplistic, of course, but that doesn’t seem like a very dire situation.
The optimistic view is AI improves and generates significant value. You probably need it to generate of the order of 1% of world GDP for the investments to make sense.
As someone who does not understand how IPO works. So they do not need to show whole “big picture” of their revenue vs costs before going public? Wouldn’t build it trust to show they are healthy company worth investing besides speculation? (It must be naive question from)
Yes they do. They have to file an S-1 with the SEC, which will be made public about a month before the IPO.
The S-1 has to include, among other things, three years of audited financial statements, plus interim statements (unaudited). It will cover both revenue and expenses, the latter breaking out things like cost of revenue, R&D, sales and marketing etc.
Based on the (unofficial but reported) IPO target date of late Sep to early Oct, the S-1 will have to be made public in a few weeks from now.
They do, but the accounting rules change based on context. They will file an S1 today outlining how they are rolling in profits then, come tax time, suddenly they are hemorrhaging money. Flesh-and-blood investors have lost all trust in financial filings. See SpaceX.
It's obviously not in their interest. Disclosing would only work as marketing for their IPO and it's the most anticipated IPO in history and needs no marketing. There is like zero positive side to disclosing audited financial statements and massive liability.
"There is like zero positive side to disclosing audited financial statements and massive liability."
Why are you posting about stuff you have zero clue about?
Oh theres no positive side... yes there is. There is a huge amount of failure risk weighing on both OAI and Anthropic - investors don't care about how great you claim your technology is gonna be. They want to know if a viable buisiness model is taking form and whether you will be around long enough given the investment time horizon of the investor.
Right now China is making that failure risk even larger. This directly affects the IPO.
They reported a $47B run rate in May. This article claims $4.7B revenue in Q1 and $11.5 B in Q2. It all aligns with a very high growth rate. Here's one set of numbers that fits (though I would guess the actual growth was spikier than this):
Jan $1.0 B
Feb $1.5 B
Mar $2.2 B
Apr $3.0 B
May $3.9 B
Jun $4.6 B
We already know they are not profitable in the truest sense w.r.t valuation.
FCFF = EBIT(1-t) - Reinvestment.
This is how the operating assets are valued via intrinsic valuation.
Could they generate immense earnings and cash flows net of reinvestment? Sure. DO I believe so? Nope. They've got way too out infront ahead of their skies about where this technology belongs and operates best.
Aren't there providers who will run open models for you? If those are profitable, or haven't taken large investments, then it's reasonable to assume that the same thing is profitable for other providers.
Many people have reported that their use would be drastically more expensive with the API. None have (afaict) reported how much the average subscription is used.
That means that Anthropic with its lousy revenues should have an IPO for $25 billion and not $2 trillion. All growth scenarios are a complete fantasy. They aren't even profitable and will never be.
rogerkirkness | 5 hours ago
rvz | 5 hours ago
There's a reason why a company like Stripe can stay private far longer than Anthropic or OpenAI can.
These AI companies have taken in all the capital from private investors and are still losing hundreds of billions and have no choice but to hype up the IPO and dump some of the stock at a purposefully inflated valuation to retail investors.
mlnj | 5 hours ago
WarmWash | 5 hours ago
Even in my non-SWE job, paying $100/mo for my current $20/mo plan would still be a no-brainer.
I don't think there is much concern about open models either. Compute is constrained for the foreseeable future, and money is what will determine who gets it. Nevermind that the US will likely block Chinese model imports or China will block exports at some point. The cold war has already begun here.
mlnj | 5 hours ago
eddq | 5 hours ago
Most people shouldn't open their mouths / write anything re. valuation TBH.
WarmWash | 4 hours ago
Just look at software over the last 20 years. People pay based on the value they receive, not the cost the produce or serve it. That fact is literally is the backbone of tech, and why it has been an absolute money machine.
I think the worst case scenario for the labs is current (or next gen) SoTA models reaching a point where cheap consumer hardware can fully run them. But the labs practically have a monopsony on compute, and getting the kind of long context current models thrive on out of 16GB GDDR6 is gonna be a trick.
3dd3 | 4 hours ago
Another bozo who read a intro microeconomics textbook, learned a fancy word, and doesn't know how to apply it! LOL.
Wow you people on here are really funny.
gizmo686 | 4 hours ago
Lerc | 4 hours ago
alasdair_ | 4 hours ago
The average American family won’t be willing to pay more than a Netflix subscription.
I think in five years, it will only be power users that use a model in its raw form - everyone else will mostly consume using wrapper apps.
derwiki | 4 hours ago
lumost | 4 hours ago
bitmasher9 | 4 hours ago
We still have 2billion+ people offline. Looking at global population is the wrong reference frame for selling a $100/mo service.
sottol | 3 hours ago
I think more realistically we'll have something like the Google/social media US-vs-world profit split of 40-50% US vs rest of world combined. Even those numbers can work out but then I don't see tremendous growth.
But the sector valuation is already priced for wholesale workforce replacement or massively expanded productivity and AI platform providers taking a lot of that pie for themselves.
With corporate profits already near all-time highs with respect to GDP, who is going to buy all those new products (from expanded productivity ) if all the gains only go to OIA and Anthropic employees?
It's an interesting time.
Bender | 4 hours ago
lumost | 4 hours ago
At the investment scales being discussed, CUDA/architecture and other advantages do not matter - you could spend 1 billion on building a new chip architecture. The ram/fab inputs have been a commodity market for years. Heck, even the model bottleneck doesn't seem real when it's only 1-4 billion or less to get a state of the art model.
At some point the compute bottleneck will be relieved, you can see NVidia hedging their strategy with both open models and on-device chips targeted for local inference. The 200 dollar a month plan will absolutely be taken over by local hardware in the future.
HarHarVeryFunny | 4 hours ago
There is competition everywhere, and it is intensifying and catching up, not fading away. Open weight models are becoming more common, both within the US as well as elsewhere. Treasury secretary Scott Bessent just praised Meta's open weight models.
There is demand for AI at all different price points, and as all models at all price points become more capable, it seems that increasingly developers are seeing the most expensive ones as specialized tools, not daily drivers.
Compute/memory may be constrained for a few years until production capacity catches up, but this does not mean that demand for cheaper and open weight models will go away, else it would already be happening. Anthropic would like to sell an expensive Ferrari to everyone on the planet, but 99.99% of those people have no need for anything more than a Yugo.
WarmWash | 4 hours ago
HarHarVeryFunny | 3 hours ago
DeepSeek recently said that their super-low pricing let's them recoup the cost of the hardware it runs on in 10 months, so there is evidentially plenty of profit to be had over a projected 3+ year lifespan of a "GPU".
Some in the AI industry, or breathing the same air (Dwarkesh) project that limited GPUs will only be used to serve the most expensive models with the highest profit margins, but it is just not what we are seeing. If the only LLMs available were ones at Opus/Fable price points then the GPU scarcity would disappear since the demand at that price is just not there. It's remarkably like trying to fill all the seats on a plane - you can fill a few at 1st class prices, but most of the plane better be coach if you want to sell all the seats.
For a GPU, "selling all the seats", keeping it busy 24x7, is critical to profitability since the primary cost to serving is the GPU which has a limited lifespan.
WarmWash | 2 hours ago
There is a clear trend of popularity and price.
HarHarVeryFunny | an hour ago
For example, here we have Fable 5 at $3.14/task vs Kimi K3 at $0.84/task, with very little difference between them in coding capability (and this isn't even a coding/agentic fine tune of Kimi).
https://artificialanalysis.ai/models
We now have models like Qwen 3.8 27B, small enough to run locally, with coding capability similar to Opus 4.5 based on challenging tasks like the Anthropic Kernel challenge.
I think we are rapidly getting to the "good enough" stage of LLMs, just like we did long ago with PCs. A cheap PC/LLM is all you need for 99.9% of normal use cases. Maybe nothing can touch whatever latest greatest models Anthropic and OpenAI have when it comes to solving Erdos problems, but most developers are working on problems more like the Anthropic Kernel challenge in complexity (or in fact typically way simpler ones).
delecti | 3 hours ago
kurthr | an hour ago
I don't know where the 5% of world population came from, because that's clearly not just professionals or people making a lot of money. That's Uber drivers, and retirees in the developed world or tech workers in Asia making <$10000/year. Those don't look like great markets. This needs to be 2x higher value than their cell phone and internet that they might spend $300/year on today (that's a new iPhone every 3 years on an ATT plan). It's not like it can replace their plan, because they need that connectivity to use it!
Who's getting this value other than SWEs? There aren't 40 million SWEs and I don't see them spending over $6000/year. If their business does, it still has to pass on the cost to consumers and/or fire SWEs.
lokar | 5 hours ago
GolfPopper | 4 hours ago
budsniffer952 | 4 hours ago
Why on earth do data centres need to be built from cashflow???
keeda | an hour ago
tim333 | 41 minutes ago
dominotw | 5 hours ago
seafoam | 5 hours ago
Analemma_ | 5 hours ago
williamcotton | 5 hours ago
https://www.sec.gov/resources-small-businesses/exempt-offeri...
estearum | 5 hours ago
brookst | 4 hours ago
__natty__ | 4 hours ago
pineaux | 4 hours ago
wtf_is_up | 4 hours ago
aix1 | 4 hours ago
The S-1 has to include, among other things, three years of audited financial statements, plus interim statements (unaudited). It will cover both revenue and expenses, the latter breaking out things like cost of revenue, R&D, sales and marketing etc.
Based on the (unofficial but reported) IPO target date of late Sep to early Oct, the S-1 will have to be made public in a few weeks from now.
sandworm101 | 4 hours ago
budsniffer952 | 4 hours ago
The financials were always bad. Everyone with half a brain avoided the IPO. Did you?
sandworm101 | 4 hours ago
budsniffer952 | 4 hours ago
This is the biggest capital buildout on history. Saying money will be wasted is not insightful, it's obvious.
arjie | 4 hours ago
eddq | 4 hours ago
Why are you posting about stuff you have zero clue about?
Oh theres no positive side... yes there is. There is a huge amount of failure risk weighing on both OAI and Anthropic - investors don't care about how great you claim your technology is gonna be. They want to know if a viable buisiness model is taking form and whether you will be around long enough given the investment time horizon of the investor.
Right now China is making that failure risk even larger. This directly affects the IPO.
dominotw | 2 hours ago
solomatov | 5 hours ago
yread | 4 hours ago
jonas21 | 4 hours ago
cactusplant7374 | 5 hours ago
kingkongjaffa | 5 hours ago
eddq | 5 hours ago
FCFF = EBIT(1-t) - Reinvestment.
This is how the operating assets are valued via intrinsic valuation.
Could they generate immense earnings and cash flows net of reinvestment? Sure. DO I believe so? Nope. They've got way too out infront ahead of their skies about where this technology belongs and operates best.
esafak | 5 hours ago
eddq | 4 hours ago
Both OAI and Anthropic tried to time their pricing to look good heading into an IPO window.
They got screwed as China has kept up. Wonder how they're gonna overcome this problem - protectionism? Maybe.
cactusplant7374 | 4 hours ago
dgellow | 4 hours ago
Very impressive growth for sure (they doubled in around a month?), but we don’t know the ratio of paying seats
nozzlegear | 4 hours ago
You're projecting your preference for these products onto a hugely fragmented group of people.
brookst | 4 hours ago
arjie | 4 hours ago
dgellow | 4 hours ago
Arnt | 3 hours ago
WithinReason | 27 minutes ago
Arnt | 3 hours ago
Many people have reported that their use would be drastically more expensive with the API. None have (afaict) reported how much the average subscription is used.
ashrt1154 | 4 hours ago
The market cap however is only $50 billion:
https://www.macrotrends.net/stocks/charts/MBGYY/mercedes-ben...
That means that Anthropic with its lousy revenues should have an IPO for $25 billion and not $2 trillion. All growth scenarios are a complete fantasy. They aren't even profitable and will never be.
readthenotes1 | 2 hours ago
Why do you believe the business of Anthropic and MercedesBenz are similar?
tim333 | 44 minutes ago