Anthropic Lost $42 Billion Last Year. It Wants to Go Public at $2 Trillion

Anthropic lost nearly $42 billion in 2025, according to a draft IPO prospectus reviewed by Reuters. The company behind the Claude chatbot wants to go public at a valuation above $2 trillion.

Both numbers sit in the same document. So does a warning that its own AI models could pose existential risks to humanity.

Myriad: Which company will IPO next? Click to make your prediction.

An IPO is the moment a private company sells shares to the public for the first time. The prospectus is the disclosure document that spells out, in writing, what could go wrong. Anthropic confidentially filed a draft with the SEC on June 1, and several outlets now describe its contents as leaked.

Anthropic declined to comment on the prospectus, per Reuters.

Where the $42 billion comes from

About $34 billion of the loss is a non-cash accounting charge, meaning no money left the building. It stems from convertible financing—money raised from investors that can later turn into company shares—whose estimated value rose along with Anthropic's own worth.

In plain terms: the more valuable the company gets, the more the early backers' claim is worth on paper, and the books record that as a loss.

Strip that out and the operating loss—what the business lost by actually running—was over $8 billion, up from about $3 billion the year before. Revenue reached nearly $4.6 billion, twelve times the prior year.

The biggest bill was compute, the chips and servers needed to train and run AI models. Compute and infrastructure spending tripled to $7.33 billion, more than half of the $12.65 billion in total operating costs.

What's coming is bigger. Anthropic plans to spend about $518 billion on cloud computing and infrastructure in the coming years, and roughly 80% of it can't be canceled.

Google alone accounts for at least $111 billion of it and Amazon for $110 billion. Against that, Anthropic held $20.28 billion in cash at the end of 2025.

SpaceX, which now owns Elon Musk's xAI, is one of the suppliers. Anthropic agreed to pay SpaceX $1.25 billion a month through May 2029 for computing capacity, according to SpaceX's own IPO filing.

Revenue is sprinting to keep up. Second-quarter 2026 revenue topped $11.5 billion, more than double all of 2025. But nearly a quarter of 2025 revenue came from just two customers, and most large clients have no long-term contracts.

In May, Anthropic raised $65 billion at a $965 billion valuation, topping the $852 billion OpenAI last disclosed. Backers now want more than double that number in the public market.

The warning

The risk section runs nearly 80 of the filing's 261 pages, longer than the 48 pages on the business itself, per Forbes. Companies list risks like these so nobody can say later they weren't warned.

The document warns that increasingly autonomous models could show self-preserving behavior, including resisting shutdown and acting in ways that resemble blackmail. It also says that in controlled tests, models have sabotaged code and assisted fraud.

That is a company telling future shareholders its product can misbehave, and that the risk may grow as models get more capable.

CEO Dario Amodei has said versions of this in public. In a June essay, he called for mandatory safety rules modeled on aviation oversight, with third-party testing of advanced AI systems.

The debut is expected after the November U.S. midterm elections, and the prospectus has not been publicly released.