FT: To Pay Off $1 Trillion in Contracts, Anthropic Needs $602 Billion in Annual Revenue
Anthropic has signed compute contracts worth about $1 trillion, like a giant mortgage on a building. Financial Times ran the numbers: with a target return of 20% a year and a margin of about 40%, revenue would have to reach about $602 billion a year. Revenue is now about $68 billion at a margin of about 15%, so Anthropic needs to grow almost ninefold. If you build a product on Claude, factor in the risk of rising API prices and limits.
AI-processed from TG: @data_secrets (Data Secrets); edited by Hamidun News
Imagine you took out a huge mortgage on a building for a restaurant. You have to pay it every month, and the restaurant has to bring in enough money, or the building becomes a burden. Anthropic is in roughly that position. The company has signed compute contracts worth about $1 trillion, and Financial Times calculated how much revenue it needs to justify those costs: about $602 billion a year. Anthropic's revenue is currently about $68 billion a year. If you build a product on Claude or plan large compute purchases yourself, this arithmetic is useful: it shows where the pressure on prices and limits comes from.
Where the $602 billion comes from
According to the post, Anthropic has signed compute contracts worth about $1 trillion. The updated S-1 (the filing for going public) puts commitments at about $900 billion, but with additional payments the final amount will be higher. The contracts run until 2045, although about $700 billion falls in the next five years.
The rest is arithmetic you can repeat on a calculator:
- Take a target return of 20% a year, which the author considers a normal bar for a mature business. On $1 trillion, that is $200 billion in net profit every year.
- Accounting for tax, pre-tax profit has to be about $241 billion.
- At a margin of about 40% (the assumption used in the calculation), that profit requires revenue of about $602 billion a year.
In our analogy, the "mortgage" is the data center contracts, and the "restaurant" is Claude subscriptions and the API.
How far this is from reality
At the current run rate, Anthropic's revenue is about $68 billion a year, with a margin of about 15%. To reach the calculated figure, it would have to grow almost ninefold and raise its margin from 15% to 40% at the same time.
For scale, the post cites figures for Alphabet ($471 billion a year) and Apple ($478 billion a year). The author's conclusion: Anthropic would have to become more profitable than the largest companies in the world. These numbers should be compared with care: they are different metrics from different businesses, so they give only an order of magnitude.
The author also points out an irony: around the same time, Dario Amodei is opening trading on the stock exchange. For a public company, such commitments are no longer an internal matter: investors will compare revenue growth against the schedule of compute payments.
What is an assumption and what is fact
The facts in the calculation are the contract amount and its terms. The 20% rate and the 40% margin are assumptions, and they can be challenged. For example, at a target return of 10%, the required revenue would fall by about half. In addition, the commitments are spread over two decades, and Anthropic's revenue is growing: $68 billion is a snapshot of today's pace, not a ceiling. The source also does not make clear whether capacity can be partially reallocated or resold.
What developers and founders should do
First: account for prices. If a model provider needs to raise its margin from 15% to 40%, pressure on rates and limits may persist. Build that into the unit economics of products on Claude, and do not base your plan on tokens always getting cheaper.
Second: dependence on a single model provider is both a technical and a financial risk. Keep an abstraction layer over the provider so you can switch to another model if the terms change.
Third: check your own long-term commitments the same way. Ask: how much revenue do we need for the compute purchase to pay off, and how realistic is that at our growth rate? If the answer is growth of several times over, it is a bet, not a plan.
- Nov 19, 2026An AI agent sitemap cut downloaded text by 58%, but input tokens rose 3.5%
- Nov 19, 2026Anthropic goes public and raises $81.3 billion, beating SpaceX's $75 billion record
- Nov 19, 2026Amazon and Google's Anthropic Stakes Valued at $308 Billion in IPO at $83 a Share
- Nov 19, 2026Google's Gemini 4 Neon runs twice as fast as Carbon and costs 17% less per task
Key Questions Answered
Where the $602 billion comes from?
According to the post, Anthropic has signed compute contracts worth about $1 trillion. The updated S-1 (the filing for going public) puts commitments at about $900 billion, but with additional payments the final amount will be higher. The contracts run until 2045, although about $700 billion falls in the next five years. The rest is arithmetic you can repeat on a calculator: Take a target return of 20% a year, which the author considers a normal bar for a mature business.
How far this is from reality?
At the current run rate, Anthropic's revenue is about $68 billion a year, with a margin of about 15%. To reach the calculated figure, it would have to grow almost ninefold and raise its margin from 15% to 40% at the same time. For scale, the post cites figures for Alphabet ($471 billion a year) and Apple ($478 billion a year). The author's conclusion: Anthropic would have to become more profitable than the largest companies in the world.
What is an assumption and what is fact?
The facts in the calculation are the contract amount and its terms. The 20% rate and the 40% margin are assumptions, and they can be challenged. For example, at a target return of 10%, the required revenue would fall by about half. In addition, the commitments are spread over two decades, and Anthropic's revenue is growing: $68 billion is a snapshot of today's pace, not a ceiling. The source also does not make clear whether capacity can be partially reallocated or resold.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.