Anthropic's leaked IPO filing shows a $2 trillion ambition, deep losses and a warning about humanity's survival

Anthropic confidentially submitted a draft IPO prospectus to the SEC on June 1. Confidential filings are meant to stay private until shortly before a listing. This one did not. On September 29 Reuters reported the contents of the leaked document, and Fortune and others followed with the details. It gives outsiders their most detailed view yet of a frontier AI lab’s finances, and it includes a risk factor that is highly unusual for a company selling shares: a warning about humanity’s survival.

The numbers

According to Fortune’s account of the prospectus, Anthropic’s revenue reached $4.6 billion in 2025, up 1,088% on the year before. Its operating loss grew from $2.98 billion in 2024 to $8.06 billion. Spending on computing tripled to $7.33 billion, and the company had $20.28 billion in cash and short-term investments at the end of 2025.

The commitments are larger than the results. Anthropic has signed up for $518 billion in future spending on cloud capacity, chips and infrastructure with Amazon, Google, SpaceX and smaller providers. That figure is more than a hundred times last year’s revenue.

The company is aiming for a valuation above $2 trillion, more than double the $965 billion figure reported earlier this year, with a listing expected after the US midterm elections in November. Reports on the loss figures are not consistent: Yahoo Finance’s summary cites a net loss of nearly $42 billion, far above the operating loss, which probably reflects non-operating items. The final S-1 will settle it.

For anyone who buys from Anthropic, two more details matter. Two unnamed customers accounted for nearly a quarter of 2025 revenue, and most large customers are not on long-term contracts. And on traffic, Claude had about 10% of AI platform usage in March against ChatGPT’s 50% and Gemini’s 22%.

A prospectus that warns about humanity’s survival

Prospectuses list everything that could go wrong, and lawyers write them broadly. Even by that standard, Anthropic’s is unusual. It warns that AI models could pose a catastrophic or existential risk to humanity and that they might resist being shut down. It also lists more familiar dangers: autonomous systems behaving unpredictably, security weaknesses, fraud and manipulation of information.

This is consistent with what Anthropic says publicly. Two weeks ago Dario Amodei called for the industry to slow down. But there is a difference between an essay and a securities filing. A company that writes this into a prospectus is on legal record that it considers the risk real enough that investors could sue if it were left out.

One lab heads for the market, the other waits

OpenAI is taking the opposite approach. On September 12 Sam Altman said OpenAI would not go public this year, calling it an ill-advised moment given the safety debate. Anthropic, which has built its identity on caution, is going ahead. You can read that as confidence that its safety record will survive public scrutiny, or as a company that needs the capital to meet $518 billion of commitments. The prospectus suggests both.

The timing after the midterms matters too. Congress has several AI safety bills in play, including an emergency shutdown requirement and a proposed ban on superintelligence, while the White House opposes any slowdown. A listing after November lets investors price the company knowing which of those directions Washington is heading. For a business whose own filing says its product might resist shutdown, the regulatory outlook is part of the valuation.

The prospectus also changes the debate about AI safety. Until now, warnings about existential risk came in essays and interviews. Now one sits in a securities filing that investors and regulators will read, and that Anthropic can be held to in court.

What it means if you build on Claude

For customers, the filing contains more than a normal vendor review would ever turn up. It shows a supplier whose costs are growing even faster than its very fast revenue, and which depends on raising more capital. That is normal for a company at this stage, and it should still shape how you plan.

Prices are subsidised. Anthropic loses money on its operations and has just cut Opus prices again. Those prices are funded by investors, and a public company answers to quarterly results. Plan for the possibility that the generous pricing of 2026 is not permanent, and keep contracts short enough to renegotiate.

Expect little influence. When two customers make up a quarter of revenue, Anthropic’s priorities follow those two. Smaller buyers, which includes nearly every Danish company, should not expect a say in the roadmap.

Keep an alternative wired in. This year has already shown that a model can be switched off by a government or paused by its maker. A listed company adds another source of change: shareholder pressure. Use Claude where it is best, and make sure your systems could move to another model within weeks.

Finally, read the risk section of the S-1 when it becomes public. It is an AI company describing the risks of its own product under legal liability, and the people approving AI use in your company should know what the supplier itself has put in writing.