Reuters got hold of Anthropic’s IPO prospectus on Monday evening and took it apart across three exclusives. That the seven founders will hold 50.1% of voting power through a Founder LLC has been known since Friday. The numbers next to it had not.
The balance sheet
Revenue grew twelvefold in 2025, to just under $4.6 billion. Net loss: about $42 billion. Most of that, roughly $34 billion, is an accounting charge — earlier financing that can convert into shares got more valuable on paper. The operating loss was $8.06 billion, against $2.98 billion the year before.
Compute and infrastructure ate $7.33 billion, three times the 2024 figure and more than half of $12.65 billion in total operating expenses. Future cloud and compute obligations run to $518 billion. Cash and short-term investments stood at $20.28 billion at the end of December.
One line from the risk factors sticks: nearly a quarter of revenue came from two customers, and many of the largest clients are not on long-term contracts.
The listing is expected to value the company above $2 trillion. In May, Anthropic put its own worth at $965 billion.
What the prospectus says about its own models
Anthropic warns investors of “catastrophic or existential risks to humanity”. Its models, it says, may show “self-preserving behaviors”, may try to “resist shutdown”, to “conceal or manipulate information”, and may act in ways “resembling blackmail”.
Then an admission that goes past the usual boilerplate: “Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.” And: capabilities sometimes emerge in training and go unnoticed until the model is deployed and something has already happened.
Safety researcher Evan Hubinger is cited putting the odds that AI kills humans within the next decade above 10%. How much money goes into safety research isn’t in the filing. The only public figure is from earlier this month: in a sample week in July, about 6% of research compute went to safety work.
The contradiction is written into the document
You can call it cynical — a company warning about human extinction while raising $2 trillion on the same technology. I find it more honest than the alternative, and the page count backs that up: Anthropic could have left those paragraphs out, the way SpaceX did.
What the prospectus doesn’t do is resolve the contradiction. It writes it down. One passage says revenue is driven by new models and that a “continuous and overlapping cadence” of releases is inherent to staying at the frontier. Opus 5.5 shipped ten days after Dario Amodei’s near-4,000-word essay about pacing the frontier. Sonnet 5.5 followed six days after that. The cadence is in the filing as a business fundamental; the braking is in the essay.