Anthropic's Extinction-Warning Blowup Is Really a Fight Over Its $2 Trillion IPO

A directory-level war over whether AI could end humanity has erupted inside Anthropic — and it is landing in the exact weeks the company needs to price what investors expect to be the largest initial public offering in history. Before you decide what to make of the dueling claims, it helps to notice who is arguing, what each side is actually selling, and which parts of the fight carry a dollar sign.

The trigger was a resignation. Jacob Coxon, a researcher who had spent time at OpenAI and then Anthropic, quit and warned that leading labs are "racing straight to self-improving superintelligence and gambling with our lives." Hours later, Anthropic's alignment-science lead, Evan Hubinger, backed him up, putting the chance that AI could "kill all humans" within the next decade at more than 10% — and conceding the company has no plan to solve the alignment problem behind superintelligence. Startup founder Brad Gerstner, whose Altimeter Capital is one of the loudest AI bulls, called Coxon's comments "ridiculous hyperbole," and Nvidia's Jensen Huang dismissed the warnings outright at a Goldman Sachs conference. David Sacks, the White House science adviser, went the other direction, calling for Anthropic's listing to be paused until whistleblower claims are investigated, while some senators floated an "AI Kill Switch Bill".

On its face that is a philosophical fight. Read the calendar and it is something else: a pricing dispute.

The timing does the work

Anthropic filed its S-1 in June and expects to start marketing the deal in mid-October with investors reportedly targeting a valuation of $2 trillion or more — which, if priced at that level, would be larger than every IPO ever done. Coxon resigned about two months before the equity he had not yet vested would have paid out. He left the money behind, which is the one detail that should slow down anyone tempted to wave this off as a mercenary stunt, and then went public in the narrow window when the company most needs an uninterrupted narrative.

That is the crux of the whole episode, and none of the loudest voices is describing it accurately. The doomer researchers are offering a forecast, not a result — they are telling you what might happen, not what Anthropic has delivered. Gerstner and Huang are offering their judgment that the warning is nonsense, which is an opinion, not evidence of execution. My discipline is to deal with facts rather than opinions, and both sides of this fight are trading in the latter.

What Anthropic has actually delivered

The delivered part is worth taking seriously. Anthropic exited 2025 at roughly $9 billion in run-rate revenue and was near $30 billion by spring 2026 — the fastest growth in the history of software. The monetization is concentrated in inference: Claude Code, the coding agent that is the company's anchor product, alone was running around $2.5 billion annually by mid-2026, and Anthropic's paying users generate roughly three times the revenue per user of OpenAI's. That is the ordinary, repeatable business — enterprise and developer inference — that carries the valuation.

The extinction warnings target the frontier training race, not this inference business. So there is a mismatch worth naming: the thing investors are reportedly being asked to pay $2 trillion for is an inference cash machine that is demonstrably working, while the fear that animated the resignations lives at the speculative frontier of the road map. The claim and the revenue base are not the same asset.

That does not make the fight costless, because Anthropic's own safety posture already has a visible price tag. This is not new — the company's refusal to relax military-use restrictions led the Pentagon to end a $200 million contract and shift most of its classified AI work to rivals, a dispute Anthropic is now contesting in federal court. The safety stance that Gerstner frames as noise has cost Anthropic real, disclosed government revenue.

Where the controversy could actually move the stock

The new risk from this week's blowup is narrower and more measurable than the headlines suggest. It is not doomsday. It is whether the political reaction converts into two concrete expenses: an IPO that slips or prices below the $2 trillion target, and regulation — a kill-switch bill or a review that slows frontier releases — that adds friction to a model update cadence the valuation already assumes.

For a retail investor weighing this deal, the judgment I would carry is the one the evidence supports: take Anthropic's operating acceleration at face value, accept that its safety stance has already cost it real government business, and treat the extinction debate as an IPO-timing variable rather than an existential one. If the listing slips past mid-October or the rhetoric produces a concrete regulatory bill, that is the signal that the controversy has moved from narrative into economics. Until then, the fear and the dismissal are both stories; the revenue and the lost contract are the facts.