Anthropic's finances have been one of the AI industry's best-kept secrets. If Reuters has read them right, they're wild in both directions.

Reuters says it has seen Anthropic's IPO prospectus. One big caveat before any numbers: this is not a public filing. Anthropic confidentially submitted a draft S-1 to the SEC on 1 June, and no public version has appeared yet. Everything below is according to Reuters' reporting, and Anthropic declined to comment.

Exclusive: Anthropic's IPO prospectus shows sweeping AI vision and surging costs as it seeks $2 trillion valuation. Read more - https://t.co/0eQpYCMCAu

— Reuters Tech News (@ReutersTech) September 29, 2026

The headline figure: Anthropic reportedly posted a net loss of $42 billion in 2025. That sounds apocalyptic, but most of it wasn't cash out the door. According to Reuters, roughly $34 billion of it was an accounting charge tied to the rising estimated value of financing that could eventually turn into Anthropic shares.

The operating picture is still steep. Reuters says the operating loss widened to $8.06 billion in 2025, up from $2.98 billion in 2024, while revenue grew 12-fold to nearly $4.6 billion. Total operating expenses reportedly hit $12.65 billion, with $7.33 billion going to compute and infrastructure, triple the 2024 figure.

And the bills are only getting bigger. The prospectus reportedly lists $518 billion in cloud, computing and infrastructure obligations over the coming years. At the end of 2025, Anthropic had $20.28 billion in cash, cash equivalents and short-term investments, Reuters says.

Two more details stand out. Nearly a quarter of 2025 revenue reportedly came from just two customers, and many large clients aren't locked into long-term contracts. TechCrunch, citing the Financial Times, adds that second-quarter 2026 revenue alone reached $11.5 billion. If that holds, growth has sped up dramatically this year.

So what about the listing itself? Reuters says the IPO could value Anthropic at more than $2 trillion, over double the $965 billion valuation from its May funding round. It has also previously reported, citing sources, that the debut will likely slip to after the US midterm elections in November. Anthropic hasn't confirmed a date, a valuation or a share price.

Then there's the part you don't usually see in an IPO pitch. In a separate Reuters exclusive, the prospectus 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." Risk factors reportedly take up about 80 of its 261 main pages.

It's a strange sales pitch: invest in us, and by the way, this could go very badly. It's also very Anthropic. The company shipped Claude Sonnet 5.5 on Monday, even as CEO Dario Amodei keeps urging the industry to slow the pace of new releases.

Until Anthropic files publicly, treat all of this as Reuters' read of a confidential document, not figures anyone else can check yet. We'll update this when the S-1 goes public.