> The filing directly discloses for the first time that Anthropic has agreed $895.4bn in purchase commitments and lease agreements for computing power and infrastructure, of which $447.5bn is expected to be incurred before the end of 2030.
So $448 billion committed, with unclear cancellation terms, through the next ~48 months.
Clearly this will be rising over time, but even a naive division results in $10bn per month of compute spending.
That honestly doesn't seem so bad relative to some of the claims I've seen? Their Q3 revenue is about $5.6bn per month and it's rising. They have so much money in the bank waiting to be deployed that they get a billion dollars of interest per year. Investors are desperate to put more money into the company.
They spend a lot on compute, but rolling it up into cumulative spend numbers makes it sound scarier.
If I’m mathing right, your own numbers suggest they’re spending nearly double their quarterly revenue, every month. That sounds far from reasonable to me.
And half of it will happen by the end of 2030. I don't know finance well enough to conjecture that these purchases and lease agreements via equity, but maybe that's their plan to reduce the hit. Or, if they will trade tokens for goods and services as part of these agreements. No clue if that is something they could do.
It just seems like, given what we know about how things have worked so far in the AI space, that paying for the entirety of these agreements is not likely.
And also, because no one seems to see it, negative profit for the foreseeable future whilst competitors have locked in "good enough" and a fraction of the price.
their this year projection at a minimum is above 60B? and its not going to be the same every year.. so not sure how 30 years of revenue is your estimate.
The commitments are over 19 years, but they expect to spend half of that, $447.5bn, by the end of 2030 - so about $105bn per year for the first few years.
Their 2026 revenue looks like it will be ~$67bn so clearly they're anticipating continued significant revenue growth for this to make sense.
> If we are unable to differentiate our models, platform, and applications from competing offerings through enhanced functionality, superior performance, better safety features and reliability, lower cost, or other factors, demand for our models, platform, and applications may decline. The rapid pace of technological change in AI means that models, platforms, and applications can quickly become outdated. We must continuously invest in research and development to maintain our competitive position, but there is no guarantee that our investments will result in models, platforms, and applications that customers prefer over competing alternatives.
My thought is that any competition to Anthropic is going to need to actually have the compute to serve all of Anthropics customers. Chinese labs certainly do not. It seems to be the crux of all these compute agreements.
Even then, Anthropic has the luxury of knowing exactly what hardware (even what datacenter!) will serve their models. This gives them the ability to co-design their model for the exact chips it will run on, something that Open Weights models are completely unable to do - even to the point that Deepseek fully open-sourced their inference engine, revealing that they're still running on only H100s and Huawei Ascend chips (https://github.com/deepseek-ai/DeepInfer/blob/main/README.md).
Anthropic will be running their 2027 Opus and Fable models on Vera Rubin, a chip which produces tokens for about 1/100th the cost of the H200. Seriously. 100x more efficient on GLM 5.3: https://inferencex.semianalysis.com/inference/glm-5-3 compared to what DeepSeek is able to use for inference.
Combined with Kimi K3.1's underperformance I'm honestly not too bullish on Chinese AI being able to catch back up. As an example, according to https://artificialanalysis.ai/, Kimi K 3.1 would need to be run 10x cheaper than its current inference cost in order to hit the Paraeto Frontier (matching 6.1 Sol Medium)
That's a really good point, I hadn't thought about that. If Anthropic have tied up almost a trillion dollars in compute commitments, that's going to make it a lot harder for their competitors to steal their customers (though OpenAI seem to be fine on that front, they've been hoarding compute for years.)
One of Anthropic's most interesting technical moats is that they can serve their models on NVIDIA hardware or Google TPUs or AWS Trainium - they talked about that in an incident postmortem last September: https://www.anthropic.com/research/a-postmortem-of-three-rec...
I think that’s OP’s exact point. It’s not just the capital costs of the build out but the operations costs of running them that can be hard to predict.
Zero percent chance they can grow fast enough to meet their financial obligations without taking on some extremely expensive debt that will increase and compound their losses.
At this point I think the goal is:
Step 1: Go public
Step 2: Allow exit for early investors
Step 3: file bankruptcy
Step 4: Amazon or Google “rescues” the company at pennies to the dollar on the already heavily diluted valuation.
If they slowdown just a bit more (and growth just slowed) they’re bankrupt.
If they grow like no startup in history they just eek by and manage to barely pay their bills. Impressive, but not worth anything close to $1T plus valuation.
ohh boy, guess who's gonna pay for all of this ? Retail investors of course. VCs want to get out quickly hence the crazy valuation, institutionnal investors are already heavily positionned on this through energy, semiconductors and cloud actors ..
Like all other problems of the future, this too can simply be solved by throwing more AI at it. The next great work of the frontier labs must not merely be to replace human labor, but also human consumption. With the ouroboros so complete, all such petty human economic concerns will cease to have relevance, as we all cleanly divide out of the equation at last.
That didn't stop any of the previous financial bubbles, financial crises, and progress-induced suffering.
If you make it through the devastation, you'll be one of the richest and most important people in the world. If not, throw a random frontman in front of the wolves and disappear into the background.
> The filing directly discloses for the first time that Anthropic has agreed $895.4bn in purchase commitments and lease agreements for computing power and infrastructure, of which $447.5bn is expected to be incurred before the end of 2030.
So $448 billion committed, with unclear cancellation terms, through the next ~48 months.
Clearly this will be rising over time, but even a naive division results in $10bn per month of compute spending.
That honestly doesn't seem so bad relative to some of the claims I've seen? Their Q3 revenue is about $5.6bn per month and it's rising. They have so much money in the bank waiting to be deployed that they get a billion dollars of interest per year. Investors are desperate to put more money into the company.
They spend a lot on compute, but rolling it up into cumulative spend numbers makes it sound scarier.
If I’m mathing right, your own numbers suggest they’re spending nearly double their quarterly revenue, every month. That sounds far from reasonable to me.
They gave revenue for quarter per month, not per quarter.
$900b contract spending / $30b = 30 years of revenue.
That’s wild. They have committed to absolutely insane growth with no margin for error.
And half of it will happen by the end of 2030. I don't know finance well enough to conjecture that these purchases and lease agreements via equity, but maybe that's their plan to reduce the hit. Or, if they will trade tokens for goods and services as part of these agreements. No clue if that is something they could do.
It just seems like, given what we know about how things have worked so far in the AI space, that paying for the entirety of these agreements is not likely.
And also, because no one seems to see it, negative profit for the foreseeable future whilst competitors have locked in "good enough" and a fraction of the price.
their this year projection at a minimum is above 60B? and its not going to be the same every year.. so not sure how 30 years of revenue is your estimate.
Hey they're just like me and my mortgage! And I consider myself very high AAA investment grade.
The commitments are over 19 years, but they expect to spend half of that, $447.5bn, by the end of 2030 - so about $105bn per year for the first few years.
Their 2026 revenue looks like it will be ~$67bn so clearly they're anticipating continued significant revenue growth for this to make sense.
If you are going bankrupt, you may as well go bankrupt spectacularly.
Or if you know a bailout is coming, might as well go full steam ahead
The George Costanza approach to bankrupting a business.
So basically it works out if they become a Microsoft level company
It works out if they can keep up basically one more quarter of their previous growth rate and then plateau for the rest of time.
The challenge there is competition. If other providers can match or beat their model quality at a lower cost that revenue could dry up pretty quickly.
Or as Anthropic put it on page 42 of their S-1: https://www.sec.gov/Archives/edgar/data/2130357/000162828026...
> If we are unable to differentiate our models, platform, and applications from competing offerings through enhanced functionality, superior performance, better safety features and reliability, lower cost, or other factors, demand for our models, platform, and applications may decline. The rapid pace of technological change in AI means that models, platforms, and applications can quickly become outdated. We must continuously invest in research and development to maintain our competitive position, but there is no guarantee that our investments will result in models, platforms, and applications that customers prefer over competing alternatives.
Thank you for the reply!
My thought is that any competition to Anthropic is going to need to actually have the compute to serve all of Anthropics customers. Chinese labs certainly do not. It seems to be the crux of all these compute agreements.
Even then, Anthropic has the luxury of knowing exactly what hardware (even what datacenter!) will serve their models. This gives them the ability to co-design their model for the exact chips it will run on, something that Open Weights models are completely unable to do - even to the point that Deepseek fully open-sourced their inference engine, revealing that they're still running on only H100s and Huawei Ascend chips (https://github.com/deepseek-ai/DeepInfer/blob/main/README.md).
Anthropic will be running their 2027 Opus and Fable models on Vera Rubin, a chip which produces tokens for about 1/100th the cost of the H200. Seriously. 100x more efficient on GLM 5.3: https://inferencex.semianalysis.com/inference/glm-5-3 compared to what DeepSeek is able to use for inference.
Combined with Kimi K3.1's underperformance I'm honestly not too bullish on Chinese AI being able to catch back up. As an example, according to https://artificialanalysis.ai/, Kimi K 3.1 would need to be run 10x cheaper than its current inference cost in order to hit the Paraeto Frontier (matching 6.1 Sol Medium)
That's a really good point, I hadn't thought about that. If Anthropic have tied up almost a trillion dollars in compute commitments, that's going to make it a lot harder for their competitors to steal their customers (though OpenAI seem to be fine on that front, they've been hoarding compute for years.)
One of Anthropic's most interesting technical moats is that they can serve their models on NVIDIA hardware or Google TPUs or AWS Trainium - they talked about that in an incident postmortem last September: https://www.anthropic.com/research/a-postmortem-of-three-rec...
You mean East India Company
Pretty insane that sama's $7T target in 2024 wasn't that far off... https://www.wsj.com/tech/ai/sam-altman-seeks-trillions-of-do...
How much to power those data centers? It's kind of meaningless if you can't power them.
They can simply connect to the existing grid and let others look for alternative or cheaper generation. Who's going to stop them?:)
You can just stop using electricity. Have you tried living in a tent? But seriously, rising prices is the solution.
I think that’s OP’s exact point. It’s not just the capital costs of the build out but the operations costs of running them that can be hard to predict.
Zero percent chance they can grow fast enough to meet their financial obligations without taking on some extremely expensive debt that will increase and compound their losses.
At this point I think the goal is:
Step 1: Go public
Step 2: Allow exit for early investors
Step 3: file bankruptcy
Step 4: Amazon or Google “rescues” the company at pennies to the dollar on the already heavily diluted valuation.
You genuinely think that the goal of this business is to go bankrupt and be sold for scrap?
I am being sardonic, but I honestly don’t see a path to profitability.
If they slowdown just a bit more (and growth just slowed) they’re bankrupt.
If they grow like no startup in history they just eek by and manage to barely pay their bills. Impressive, but not worth anything close to $1T plus valuation.
If they achieve AGI money becomes worthless.
So where’s the upside for investors here?
It makes more sense if you view themselves collectively as Claude, and their IPO is a prompt that ends with "make no mistakes."
I'd also like to reveal a Chapter 11 restructuring of Anthropic in 2029 to cancel these contracts.
I mean, I can't reveal that because I'm just some schmuck. But I'd like to.
Discussion: https://news.ycombinator.com/item?id=50337439
ohh boy, guess who's gonna pay for all of this ? Retail investors of course. VCs want to get out quickly hence the crazy valuation, institutionnal investors are already heavily positionned on this through energy, semiconductors and cloud actors ..
I dont want to be a bag holder, but the indices seem to be being gamed to get retail investors involved
Ive been wondering what to do to protect indexes from this type of practice. Its always had an element of this but now its worse
i'm pretty sure the plan is for labor to pay for it, by giving up their pay, involuntarily.
Going to be hard for companies to pay for inference if consumers don't have any money to buy their products.
Like all other problems of the future, this too can simply be solved by throwing more AI at it. The next great work of the frontier labs must not merely be to replace human labor, but also human consumption. With the ouroboros so complete, all such petty human economic concerns will cease to have relevance, as we all cleanly divide out of the equation at last.
I think that's more or less where we'll end up, sadly.
That didn't stop any of the previous financial bubbles, financial crises, and progress-induced suffering.
If you make it through the devastation, you'll be one of the richest and most important people in the world. If not, throw a random frontman in front of the wolves and disappear into the background.
[dead]
makes me wonder if these retail investors are noticing how everyone keeps mentioning them about holding the bag
The safest bet here is to put your money into datacenter infrastructure indexes :D
If’s fine. They’re the smartest guys in the room.
They will have to be bailed out. If they go public, leaders will take their money and run and everyone else will be buying up their parts.
Anyone have a non paywall link?
https://wallhop.io/https://www.ft.com/content/4d91dbab-6fe4-...
[dead]
[flagged]