Anthropic filing: $4.6B in, $42B out

The Verge reported on September 29 on Anthropic’s leaked draft IPO filing — which Reuters said it had reviewed — the first checkable ledger from a frontier lab: 2025 revenue grew roughly twelvefold to nearly $4.6 billion (from about $400 million a year earlier), the operating loss exceeded $8 billion, and some accounting treatments push the headline loss figure as high as $42 billion. The document keeps generating news: per The Information on October 4, Anthropic has also woven a charity gift-matching program worth more than $660 million into its pitch to prospective IPO investors. Multiple reports point to a mid-November listing target, weeks after the safety-pledge theatre of early October. The two-trillion-dollar valuation had circulated for months; this draft is where the numbers finally land.

47% of sales via two clouds

One of the draft’s sharpest numbers: roughly 47% of sales are routed through Amazon and Google channels. Both companies are simultaneously Anthropic’s largest investors, its compute suppliers, and its distribution path — and after a listing, those arrangements will surface as paper gains on Amazon’s and Alphabet’s balance sheets. Concentration of that order means bargaining power and dependence arrive together: the growth story has to answer what happens if the cloud partners change terms. With regulators already circling the labs, the dependency map is now in their files too — an investor-supplier-channel triple role is almost a custom-built antitrust exhibit.

Paper gains and the shareholder map

The channel concentration matters beyond operations because it connects to the post-listing shareholder map: Amazon’s and Alphabet’s stakes become markable positions after the offering, while the same two companies remain the biggest recipients of Anthropic’s compute bill — the shovel-sellers are shareholders, and the shareholder depends on shovel discounts. Nobody scrutinizes that loop in private markets; as a public company it gets re-disclosed every quarter, and AI’s “related-party transactions” face listed-company transparency standards for the first time.

A prospectus that confesses risk

The second leaked document shows Anthropic warning in its own filing that its products could pose “catastrophic” risks. Writing the company’s internal risk assessment into an offering document has almost no precedent among tech listings. It reads two ways at once — as an industry answer to this autumn’s safety-team turmoil, putting the risk acknowledgment in front of future shareholders in black and white, and as an elegant liability shield: the risk was disclosed; buyer beware. Both readings hold, which is precisely its sophistication, and precisely the lawyers’ victory. Either way, the confession raises the bar for every prospectus that follows.

Read it next to the Jev rumor

The ledger’s biggest function is resetting the anchor for every valuation story in the industry. Set TypeSafe’s reported three-week, 50x jump against it: Anthropic’s twelvefold annual revenue growth arrives paired with an $8 billion annual loss — “the growth is real and the burn is real” are, for the first time, simultaneously true and checkable. Any startup claiming a faster trajectory now has to answer one question first: what does your loss curve look like?

The first checkable ledger

Until now, every financial number attached to a frontier lab came from “people familiar with the matter.” The draft puts growth, burn, and channel concentration in one frame for the first time, and the charity-matching detail shows how far IPO marketing has evolved — even philanthropy is now roadshow material for institutional investors. A November listing — if it holds — will be the first open-market test of the AI burn narrative: a two-trillion-dollar valuation prices in profits that do not yet exist. The number worth watching is the marginal change in operating losses; beyond the revenue multiple, that is what decides how long the race can run.