Anthropic’s S-1 shows an $80.7 billion loss. The real risk sits in its compute contracts

On 26 October Anthropic published its S-1 prospectus with the US Securities and Exchange Commission and is preparing to list on Nasdaq under the ticker ANT. It filed confidentially in June. The public document is the first look at audited figures, risks and the contracts the company rests on.

The $80.7 billion loss grows with the company’s valuation, operations lost $693 million

Revenue has climbed steeply: $41 million in 2023, $386 million in 2024 and $4.6 billion in 2025. In the first half of 2026 it reached $16.3 billion, against $1.2 billion a year earlier. That matches what Anthropic told investors in October about its first two quarters, according to the Financial Times.

Net loss for the same half year was $80.7 billion. Of that, $72.4 billion is the change in fair value of convertible notes, derivative liabilities and SAFEs, driven mainly by the rise in Anthropic’s own valuation. The more the company is worth, the bigger the paper loss it reports. The operating loss was $693 million, down from $2.1 billion a year earlier. As of 30 June Anthropic held $112.4 billion in cash and another $19.3 billion in short-term investments.

Compute suppliers double as the main sales channel

The heaviest part of the prospectus is the commitments. Anthropic has agreed to spend at least $44.9 billion with Microsoft through May 2033, at least $114.5 billion with Google through July 2033 and at least $110 billion with Amazon through April 2036. It pays regardless of actual usage, and the contracts cannot be cancelled. A separate deal with SpaceXAI, formerly X.AI and the company behind Grok, secures Nvidia GPUs for up to $120.5 billion through 2029. On top of that comes a five-year equipment lease of about $125.2 billion.

The same clouds also sell Claude. Their marketplaces accounted for 48% of Anthropic’s revenue in the first half. Amazon and Google are long-standing investors as well. The big clouds therefore sit on three sides of the table: they supply compute, sell the product and hold equity. For companies buying Claude through AWS or Google Cloud, that means a stable channel. For shareholders, it means much of the bargaining power sits outside Anthropic.

Anthropic measures its own growth and has not named a share price yet

The prospectus highlights net revenue retention of about 900% as of 30 September and more than 11,900 customers paying over $100,000 a year. Its own risk factors, however, say these metrics have not been independently verified. Anthropic also warns that it does not expect today’s growth rate to last.

The filing still leaves the price range and the size of the offering blank. Among the risks are debates over whether AI systems could develop forms of sentience, and a warning that no single insurer, and possibly not the insurance industry as a whole, could cover a catastrophic AI incident. Control stays with the Long-Term Benefit Trust, which elects a majority of the board through special Class T shares.

The price range and the first quarter of full compute bills will decide

The first test comes with the price range, which will show how much investors will pay for growth the company measures itself. The second comes with the first quarters as a public company. According to the prospectus, SpaceXAI capacity should be fully delivered in December 2026 and then cost about $3.1 billion a month. That is over $9 billion a quarter, more than half of the $16.8 billion the company reportedly earned in the third quarter, according to the FT.

If revenue grows faster than the compute bills, the $80.7 billion loss stops being the headline. If it does not, contracts signed in 2026 will set Anthropic’s pace.

Lilith's verdict

Anthropic arrives on the stock market like a tenant who has signed non-cancelable leases with the big clouds while selling almost half its goods over their counters. Most of the $80.7 billion loss is on paper, but the rent will be paid in cash.