Anthropic flags existential AI risks in its IPO filing
The prospectus pairs 4.6 billion dollars in revenue and an operating loss above 8 billion with a warning that AI could threaten humanity.
In short
In its own IPO prospectus Anthropic lists existential risks to humanity as a risk factor, while targeting a valuation above 2 trillion dollars.
At a glance
- 2025 revenue: 4.6 billion dollars, roughly twelve times the prior year.
- 2025 operating loss: 8.06 billion dollars, up from 2.98 billion.
- Committed cloud, compute and infrastructure spending: 518 billion dollars.
- Target valuation: above 2 trillion dollars, listing reported for November 2026.
- Close to a third of the prospectus consists of risk factors.
Anthropic has placed existential risks to humanity in the risk-factor section of its own IPO prospectus, and is pursuing the listing regardless. Reporting on the S-1 describes risk factors filling close to a third of the document. Among them: the possibility that powerful models operate outside their predicted parameters despite safety controls, plus manipulation and blackmail by AI systems.
A growth chart and a loss column
Revenue for 2025 comes to 4.6 billion dollars, roughly twelve times the prior year. Against that sits an operating loss of 8.06 billion dollars, up from 2.98 billion. Compute and infrastructure alone consumed 7.33 billion dollars, about three times the year before. The headline net loss of roughly 42 billion dollars is softer than it reads: some 34 billion of it is an accounting charge tied to convertible financing.
The second quarter of 2026 is reported at 11.5 billion dollars in revenue by the-decoder. Concentration is the counterweight: two customers accounted for nearly 25% of 2025 revenue, with few long-term contracts locking that in.
The number that forces a listing
The figure that explains the IPO is not a historical one. Anthropic has committed 518 billion dollars to cloud, compute and infrastructure over the coming years, more than a hundred times its 2025 revenue. Capital on that scale does not arrive through customer invoices, and public markets are one of the few places it exists.
Above 2 trillion dollars, after the midterms
The target valuation sits above 2 trillion dollars, more than double the 965 billion assigned in May. That would put Anthropic ahead of the roughly 1.8 trillion SpaceX carried into its June listing. The reported window is November 2026, after the US midterm elections. OpenAI filed confidentially for its own IPO in June 2026, so both leading labs would reach public markets within months of each other.
Why the warning is not boilerplate
Risk-factor language is a legal genre investors are trained to skim. What is unusual here is the substance: a company selling growth states in the same document that its product could become an existential danger. The reporting traces to a Financial Times account of the filing, which cites the possibility of powerful models running beyond predicted parameters despite safety controls. Anthropic has made that case publicly since it was founded six years ago; the liability document is the new venue.
What we could not verify
A third report on the filing, which also covers proposals for the leadership to retain control, was not retrievable for this article. Share structure and individual governance clauses are therefore left open here rather than asserted. The prospectus itself is not public, and every figure above comes from reporting on the filing, not from a document we read.
FAQ
How much revenue does Anthropic make?
The filing puts 2025 revenue at 4.6 billion dollars, roughly twelve times the prior year, and the second quarter of 2026 is reported at 11.5 billion dollars.
What valuation is Anthropic targeting in its IPO?
Above 2 trillion dollars, more than double the 965 billion valuation from May. The listing is reported for November 2026, which is not a confirmed date.
Why does Anthropic list existential risk in its IPO filing?
Reporting on the filing cites the possibility that powerful models act beyond their predicted parameters despite safety controls, including manipulation and blackmail.
