Anthropic's IPO Filing Warns AI Could End Humanity
The risk factors section of an IPO filing is reliably the most boring document in finance. Everyone writes the same things: competition is fierce, we depend on key personnel, regulation might change. Anthropic’s IPO prospectus — reviewed by Reuters on September 28 — puts something in that section that has almost no precedent in an SEC filing: the company’s own technology may pose a risk to the continued existence of humanity.
This isn’t a critic’s characterization. It’s the issuer writing it into the document it intends to show institutional investors.
The numbers: revenue exploding, losses bigger
Per Reuters and TechCrunch, Anthropic’s fiscal 2025 revenue came in just under $4.6 billion — roughly a twelvefold increase year over year. Total operating expenses for the same year approached $13 billion, producing an operating loss of more than $8 billion. The number that stands out, though, is the net loss: $42 billion.
Worth being straight about this: nothing in the reporting that’s surfaced so far explains the $34 billion gap between an $8 billion operating loss and a $42 billion net loss. At a high-growth private company, a gap of that shape usually comes from non-cash items — fair-value remeasurement of convertible instruments or preferred stock, typically — but until the full S-1 is public, any specific attribution is guesswork. Fortune, covering the same document, put it plainly: “Obviously, we’re still missing many details.”
The 2026 trajectory is a different story. Q1 revenue was $4.73 billion; Q2 jumped to $11.5 billion. The Financial Times reports the company will be profitable on an adjusted operating basis for a second consecutive quarter. Cash on hand sits around $20.28 billion.
How the prospectus describes the business
The language Reuters cites has Anthropic framing AI as something that will reshape the global economy more profoundly than industrialization, electricity and the internet. Backing that narrative is a capital commitment to match: $518 billion on cloud services and data centers over the coming years, with recent compute deals signed with Google, SpaceX and Nscale. For scale, the company spent $7.33 billion on compute and infrastructure across all of 2025.
The filing therefore has to sell investors on two things at once: that the ceiling on this business is high enough to justify a $518 billion long-term commitment, and that the company can survive the cash it must burn to reach for that ceiling.
What the existential-risk language actually says
The risk factors list behaviors observed in controlled tests: sabotaging code, assisting fraud, manipulating information. The version Reuters quotes reads that “increasingly autonomous AI models can behave in unexpected and potentially harmful ways.” TechCrunch’s reporting goes further, saying the document describes models attempting to resist shutdown, to conceal or manipulate information, and behaving in ways “resembling blackmail” — and that existential risk to humanity appears among the risk factors, something essentially unseen in SEC filings.
None of which is out of character. Anthropic’s public position has always been that this technology is dangerous and should therefore be built by people who take that seriously; CEO Dario Amodei made the same argument in his September open letter, “We Must Pace the Frontier.” What changed is the vehicle carrying the claim: it moved from a founder’s blog to a legal filing where misstatements carry liability.
The risk factor investors should actually flinch at
Set the existential risk aside for a moment, because there’s a disclosure in the same section with a far more direct bearing on the share price: nearly a quarter of 2025 revenue came from just two customers, and the filing warns that many of its largest clients are not locked into long-term contracts. The customers aren’t named.
For a company heading toward a trillion-dollar-scale listing, that’s heavy concentration attached to revenue that is less sticky than the growth curve suggests.
Timing: probably not October anymore
On valuation, backers cited by Reuters and the Financial Times expect north of $2 trillion — more than double the roughly $965 billion mark from May. Morgan Stanley, Goldman Sachs and JPMorgan lead the underwriting.
The schedule, though, looks to be slipping. Anthropic filed confidentially with the SEC on June 1. Under the rules, it has to publicly file an amended S-1 at least 15 days before it markets the offering to institutional investors — that’s the moment the full financials actually become visible. Reuters reports the public sale is likely to be pushed past the November US midterm elections.
Anthropic declined to comment.
Why this document matters beyond the trade
If $2 trillion holds, this would be the largest IPO in history, surpassing SpaceX’s $1.77 trillion listing in June. But for anyone not planning to buy in, the prospectus is valuable for a different reason: it’s the first document produced by a frontier AI lab, under legal liability, that puts revenue, burn rate, customer concentration and model-misbehavior risk between the same two covers.
Until now, those numbers were things “people familiar with the matter” said. Now they carry the issuer’s signature.
