Anthropic's IPO filing warns its AI could end humanity. That's the honest part.
Jump to a section
See the story at a glance
If you've been using Claude to draft emails, talk through a hard decision, or just figure out what to make for dinner, you probably don't follow IPO filings. That's fair. Most of them are written by lawyers for investors and are interesting to approximately nobody else.
Anthropic's is different. Reuters reviewed Anthropic's S-1 prospectus on September 28, and buried inside it is a sentence no company has ever had to write in a legal filing before: a warning that the company's own core technology could pose "catastrophic or even existential risks to humanity." Eighty pages of risk factors. Including the possibility that its models might develop self-preservation behaviors.
That's not marketing. That's a legal disclosure. The SEC requires companies to tell investors the actual risks of their business before taking their money. So Anthropic had to write it. And they wrote it.
Source spread
- Reuters / CNBC — S-1 review, September 28 [hype] — first look at the prospectus, emphasizes growth and the sweep of the vision
- Bloomberg — IPO timeline, October 1 [builder] — confirms the November target and the $2T+ valuation goal
- SeekingAlpha — investor analysis [skeptic] — focuses on loss structure and spending commitments
- InvestmentNews — compute bill breakdown [skeptic] — best breakdown of the $518B non-cancellable commitment structure
Pros & cons
What's real:
- The 12-fold revenue growth is genuine. $4.6 billion in 2025 revenue from roughly $383 million in 2024 is the kind of growth that explains the $2T valuation target even alongside the losses.
- The $42B "loss" headline is misleading. The real operating loss is $8.06 billion, up from $2.98B in 2024. That's bad, but it's not 2008-financial-crisis bad. The other $34B is a paper accounting item from their financing instruments rising in value as the company's estimated worth climbed. When your company goes from $965B to $2T in estimated value in less than a year, the accounting math gets weird in a way that the headline number doesn't explain.
- The risk section disclosures are required by law. The humanity warning is remarkable not because Anthropic is being reckless. It's remarkable because the SEC's rules forced a company to put, in permanent public record, a statement about the severity of what they're building. Most companies would hire lawyers to soften it. The Anthropic version didn't soften it.
What deserves a side-eye:
- $518 billion in total commitments, 80% non-cancellable. Those are contracts with cloud providers — Google, Amazon, Microsoft — that Anthropic has to pay regardless of whether the models keep improving or the customers keep coming. That is the most aggressive infrastructure bet in tech history. It's also the one bet they can't take back.
- Going public means shareholders. Shareholders want returns. Anthropic was founded as a safety company. Those two things aren't always in tension. But when they are, the S-1 doesn't tell you which one wins.
- $7.33 billion in compute spend in one year — tripling from 2024 — means Claude users are being subsidized by investor money right now. What happens to pricing when that subsidy runs out is the unasked question inside the filing.
What to do about it
- If you're a free Claude user: nothing changes right now. IPOs don't affect the product overnight. But pay attention to what happens to free-tier limits in 2027. That's when the pressure typically starts.
- If you pay for Claude Pro or Max: you have a product Anthropic is actively investing in keeping strong, because paid subscribers have more leverage than free users when a company starts optimizing for margin. You're in a better position than you might think.
- If you're a builder using the API: the $518B compute commitment is actually a reassurance, in one sense. Anthropic is locked into running serious infrastructure for a long time. They can't quietly scale back. API stability over the next three years is more likely, not less, because of how deep the infrastructure commitments run.
- For everyone: the humanity warning in the S-1 is the kind of honest disclosure that public accountability produces. IPOs don't only create shareholder pressure. They also create scrutiny. More people will be watching Anthropic's safety work now than before. That's not nothing.
Further reading
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.
Keep following the thread
A little more to explore.
Anthropic is a public company now. The $26B nonvoting stake matters more than the IPO price.
Anthropic priced its IPO at $83 per share on November 18, raising $81.3 billion — and separately closed a $26 billion concurrent private placement from seven tech companies including Google, Microsoft, Nvidia, and AMD. ANT begins trading on Nasdaq today. Here's what actually changes for builders on the API.
LangChain raised $300M. The number that matters is 7,000.
LangChain closed a $300 million Series C led by CapitalG at a $3 billion valuation on November 13. The headline is the round; the signal is the 7,000 organizations using LangSmith. Here's what I think this bet says about where enterprise agent tooling goes next.
Anthropic prices at $70–$78. It goes public November 19. Here's the part that actually matters.
Anthropic set an IPO price range of $70 to $78 per share on November 10, 2026, targeting up to $1.4 trillion in fully diluted valuation and trading on Nasdaq as ANT starting November 19. The financials are interesting. The governance structure — a single Class F share giving co-founders 50.1% of votes — is what builders should actually care about.