Anthropic’s IPO Filing: $4.6B Sales, a $518B Compute Bill

The numbers behind Claude are finally on paper. Anthropic’s IPO prospectus, a 261-page document seen by Reuters, shows revenue of $4.59 billion in 2025, up from $386 million a year earlier, alongside an operating loss of $8.06 billion. The company filed confidentially in June at a $965 billion private valuation; backers are now aiming for more than $2 trillion at a listing expected after the US midterm elections in November. The filing has gone to a limited set of parties and is not yet public on the SEC’s EDGAR system, The Decoder notes.

The headline net loss looks far worse than the business. Anthropic reported a GAAP net loss of about $41.97 billion for 2025, against $8.31 billion in 2024, but roughly $34 billion of that is a non-cash accounting charge: financing that can later convert into stock has to be marked at a higher estimated value as the company’s worth climbs. The operating figures are the ones that describe the cost of the work. Compute and infrastructure spending tripled to $7.33 billion, more than half of all operating costs, which means Anthropic spent roughly $1.60 on compute for every dollar it brought in. It ended the year with $20.28 billion in cash.

This year the picture flips. Revenue reached $4.73 billion in the first quarter and $11.5 billion in the second, according to Bloomberg figures compiled by PYMNTS, and by the first quarter compute had fallen to about 71 cents per dollar of revenue. The run rate that stood near $9 billion at the end of 2025 hit $30 billion in April. That is the growth the $2 trillion target is betting on, and the reason Anthropic was talking about operating profit well before anyone expected it.

The biggest number in the document points the other way. Anthropic lists about $518 billion in future cloud, compute and infrastructure commitments, around 80% of it non-cancelable, per Reuters. The largest pieces are $161.2 billion in Broadcom equipment leases, $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft, most running seven to ten years with cancellation fees. There is also an $84.5 billion agreement with xAI for capacity on Colossus that, unlike the rest, can be cancelled on 90 days’ notice. Put simply, the company has signed up for more than a hundred times its 2025 revenue in future spending.

The risk section is long even by IPO standards, roughly 80 of the 261 pages. Two customers, unnamed, each made up about 12% of 2025 revenue, and many large clients are not on long-term contracts. The document also repeats in legal language what Anthropic’s leadership has said in public, that AI could pose “catastrophic or existential risks to humanity,” and it describes models behaving unexpectedly in controlled tests, including sabotaging code and helping with fraud. That sits alongside a live legal fight: a federal appeals court upheld the Pentagon’s supply-chain risk label on Anthropic this month.

What the prospectus does not yet contain is a ticker, an exchange or a price range, and those will come with the public filing. When it does land on EDGAR, the figures to watch are the second-quarter margin and whether the xAI deal survives, since it is the one large commitment Anthropic can walk away from quickly. Investors will be buying a company whose sales grew twelvefold in a year and whose bills are already contracted out to the middle of the next decade.