Anthropic's confidential IPO prospectus, reviewed by Reuters on September 28, shows a company growing fast and spending faster. Revenue rose about twelvefold in 2025 to nearly $4.6 billion, the net loss was $42 billion, and Anthropic has committed to at least $518 billion in computing and infrastructure over roughly a decade. About 80% of that must be paid whether or not the capacity gets used. The filing has not been made public, and Anthropic declined to comment.

Table of contents

The $42 billion loss, explained

The headline loss looks worse than the operating picture. Reuters reports that about $34 billion of the $42 billion net loss is an accounting charge. It reflects a higher estimated value of financing that could eventually convert into Anthropic shares, rather than cash spent running the business. Strip that out and the operating loss was a little over $8 billion.

That is still large. Total operating expenses were $12.65 billion in 2025, against revenue of nearly $4.6 billion. Compute and infrastructure made up $7.33 billion of those expenses, three times the 2024 figure and more than half the total. At the end of 2025 the company held $20.28 billion in cash and short-term investments, according to the Calcalist summary of the filing.

Where the $518 billion goes

Reuters says the money is split among six infrastructure partners over seven to ten years. The named amounts, as reported by TrendForce and others, are:

  • At least $111.1 billion with Google
  • At least $110 billion with Amazon
  • At least $31.4 billion with Microsoft
  • About $161.2 billion in Broadcom-related equipment lease obligations
  • Up to $84.5 billion with SpaceX through 2029, for Nvidia-based computing capacity
  • More than $20 billion of capacity from AMD, which has also agreed to buy up to $5 billion of Anthropic stock

The SpaceX deal is the odd one out. Reporting says it is largely cancelable on 90 days' notice, unlike the cloud commitments. Stocktwits describes it as nearly double the initial agreement.

Anthropic's own explanation, per the prospectus, is that demand for advanced AI will likely outrun supply. It says AI development will be "limited principally by the availability of compute," and the commitments are meant to secure that capacity early.

Why the non-cancelable share matters

The 80% figure is the part analysts keep coming back to. A cloud contract you can walk away from is a cost. One you must pay regardless of use is closer to debt. If Anthropic's revenue grows as fast as it hopes, the commitments look prudent. If demand slows, or if models get cheaper to run, the company would be paying for capacity it does not need.

Equity analyst Ross Hendricks, author of the Ross Report, questioned in the coverage whether commitments of this size are sustainable, and tried to sketch a financing plan around them. Reuters and the other outlets did not report any specific plan from Anthropic itself beyond the IPO. The prospectus also acknowledges that its business and results could be adversely affected if it cannot meet these obligations.

There is a circular quality here too. Amazon and Google have both invested billions in Anthropic while also supplying the cloud capacity it trains on. Alphabet's own results show a paper profit from its roughly 14% stake, according to one write-up. Money flows from those companies to Anthropic as investment, then back to them as cloud spending.

Customer concentration

The prospectus also shows how much of 2025 revenue came from a few buyers. Implicator reports that two customers supplied nearly a quarter of it. That is a normal risk disclosure for a young company with large enterprise contracts, but it matters more when fixed costs are this high.

Valuation and timing

Anthropic is reportedly seeking a valuation above $2 trillion. Reuters notes that is more than double its own estimate of $965 billion in May. According to Reuters, the listing is now likely to come after the US midterm elections in November rather than before.

Anthropic also signed the voluntary safety pledge at the White House this week, which we covered in our report on the Super Intelligence accord. The prospectus itself reportedly pairs its pitch to investors with warnings about catastrophic AI risk, according to Calcalist, a mix not often seen in an IPO document.

Nothing here is final. The document is confidential and the numbers come from a reporter's review of it. Terms can change before a public filing, and Anthropic has not confirmed any of the figures.

FAQ

How much did Anthropic lose in 2025?

Its net loss was $42 billion, but about $34 billion of that was an accounting charge tied to financing that could convert into shares. The operating loss was just over $8 billion.

What is the $518 billion?

It is the minimum Anthropic expects to spend on cloud, computing and infrastructure with six partners over roughly a decade. About 80% is non-cancelable or payable regardless of usage.

When will Anthropic go public?

Reuters says the IPO will likely come after the US midterm elections in November. No date has been announced.

Has Anthropic confirmed these figures?

No. The prospectus is confidential, and Anthropic declined to comment on Reuters' reporting.