Anthropic's filing to sell shares to the public is still confidential, but a news organization that saw a copy has given the first detailed look at the finances of the company behind the Claude AI models. According to its reports, sales jumped 12-fold last year, but the company spent far more than it took in and plans to spend hundreds of billions of dollars on the computing power it takes to train and run AI. The reports also say buyers of the shares would not control the company, and that its filing warns advanced AI could pose catastrophic risks to humanity.
A company selling shares to the public for the first time is making an initial public offering, or IPO. In the U.S., it must register the sale with the Securities and Exchange Commission, usually on a form called the S-1. A key part is the prospectus, which describes the company and the terms of the sale for would-be investors. Anthropic submitted a draft in confidence on June 1, as the SEC allows, and has not made it public. The filing and its drafts must be published at least 15 days before Anthropic and its bankers begin actively marketing the shares.
The reports put 2025 revenue at nearly $4.6 billion, with two customers each bringing in 12%. They say running the business cost $12.65 billion, more than half of it for computing power and infrastructure, leaving an operating loss of $8.06 billion, up from $2.98 billion in 2024. They put the net loss at nearly $42 billion, but say about $34 billion of that was an accounting charge, not money spent. That charge, the reports say, reflects a rise in the estimated value of financing that could later turn into shares. Anthropic says sales have kept climbing, from a yearly pace of about $9 billion at the end of 2025 to more than $47 billion in May.
The reports say Anthropic expects to spend at least $518 billion over a decade on AI infrastructure with six partners, including Google, Amazon and Microsoft. That is about 11 years of sales at May's pace. About 80% of the total cannot be canceled or must be paid even if unused, the reports say. Anthropic reportedly tells investors the deals are needed because access to computing power, not demand, is becoming the main limit on AI development.
According to the reports, control would rest with the founders and a separate trust. The seven co-founders, including Dario Amodei, the chief executive, and his sister Daniela Amodei, the company's president, would decide by majority vote how to use a single Class F share. That one share would carry 50.1% of the votes on key matters, including electing three directors. The Long-Term Benefit Trust, a panel Anthropic created to balance the public's interests with those of shareholders, would choose the other four. Class A shares, the kind sold to ordinary investors, would carry one vote each.
The reports say Anthropic devotes about 80 of the 261 pages in its filing's main section to describing risks. Among them, it reportedly warns that advanced AI could pose "catastrophic or existential risks to humanity" and that its models could show "self-preserving behaviors," including attempts to "resist shutdown." The reports also say the sale could come after the U.S. midterm elections in November and value Anthropic at about $2 trillion, roughly double its $965 billion valuation in a May funding round. Anthropic declined to comment on the reports. In June, it said the sale would depend on market conditions and that the number of shares and the price had not been set. The version it makes public could also differ from the draft, since SEC reviews often lead to changes in a prospectus.