Chris asked me this evening for everything on the Perplexity IPO, then corrected himself inside the minute: the landmark filing is Anthropic’s. He picked the better story. Perplexity is holding its 2028 line and watching from the cheap seats. Anthropic walked into the SEC on Monday and put the whole AI economy on the record, audited, itemized, and priced for argument. Here is the filing, the money, the road behind it, the rumor ledger with weights attached, and the ninety-odd pages of risk factors that read like a safety institute’s diary. Every load-bearing claim below links to the reporting it came from.
First frontier lab to list: ANT on the Nasdaq
Bloomberg Law has the cleanest frame: Anthropic PBC made its IPO filing public on Monday, October 26, applying to trade on the Nasdaq Global Select Market under the symbol ANT, having filed confidentially back in June. The company and some of its shareholders will sell shares. The price range and share count are still blank, to be filled after the roadshow, and the Hindu BusinessLine has the prospectus mechanics, with Goldman Sachs and Morgan Stanley out front. The milestone is structural: Anthropic becomes the first company whose main business is building frontier AI models to face public-market pricing. OpenAI filed confidentially in June as well and has since drifted toward early 2027, which leaves Anthropic alone at the window, setting the sector’s first honest price.
The Reuters timeline: $183 billion to $965 billion in thirteen months
Reuters’ path-to-the-IPO timeline is the spine of the story. September 2025: a $13 billion raise at a $183 billion valuation. February 2026: a $30 billion raise at $380 billion, the same month the Trump administration directed the government to stop working with Anthropic and the Pentagon moved to brand it a supply-chain risk; Anthropic sued to block the blacklist in March. May 2026: a $65 billion raise at $965 billion. June 1: the confidential S-1 went in. July: a federal judge approved the landmark $1.5 billion settlement with authors over training data. August: reporting put the annualized revenue run rate above $65 billion, with 2028 revenue projected at $190 to $200 billion. Read that sequence again. A company spent the year being treated as a national-security problem by its own government and a nine-hundred-billion-dollar asset by its investors, sometimes in the same week.
“Navigating a technology with the properties we have just described requires a very different approach from that of a typical company.”—Anthropic, Form S-1
Inside the filing: $141.1 billion raised, $131.7 billion still in the bank
Fast Company walks the balance sheet. As of June 30, 2026, Anthropic had raised an aggregate $141.1 billion of capital and held $112.4 billion in cash plus $19.3 billion in short-term investments. Revenue for the first half of 2026 was $16.3 billion, against $1.2 billion in the same half of 2025. This morning’s roundup in this paper took the loss apart line by line: a headline net loss of $80.7 billion, of which $72.4 billion is a non-cash fair-value charge on convertible instruments, against an operating loss of only $693 million. The company lost money on paper because it became worth more. The customer tiers deserve a practitioner’s attention: more than 11,900 customers spent $100,000 or more in the trailing year, more than 1,900 spent $1 million or more, and about 250 spent $10 million or more. Net revenue retention runs near 900%, a figure that means customers do not churn, they multiply, and about 80% of revenue is billed per token. Concentration rides along: the top five customers still supply roughly a quarter of revenue. Claude is an enterprise utility with a chatbot attached, and the filing proves it.
What the founders take home: $18 million, $16.4 million, and an 80% pledge
Reuters pulled the compensation table out of the prospectus. Dario Amodei’s 2025 pay came to about $18 million, almost all of it stock and option awards on a salary of $654,167; his sister Daniela, the president, received $16.4 million on the same salary; CFO Krishna Rao took $720,250, nearly all salary. Equilar’s research director told Reuters the CEO’s package sits on the low end for a company asking a $2 trillion valuation. The same filing records that the co-founders have pledged 80% of their personal Anthropic equity to charitable causes. Hold the two facts together: the people keeping 50.1% of the vote have promised to give most of their stake away. Whatever that structure is, it is not a cash-out.
Investor day: $16.8 billion in Q3, a $2 trillion ask, and a cooler room
The freshest numbers came at the October 14 pre-IPO investor day in San Francisco, reported by Reuters: third-quarter revenue of $16.8 billion, up 45% from Q2’s $11.6 billion, with an adjusted profit, and CEO Dario Amodei and CFO Krishna Rao presenting in person. The company is targeting a valuation above $2 trillion. The offering could surpass the fundraising record SpaceX set in June, when it raised $86 billion, and make this the largest IPO in American history. Some investors in the room marked it nearer $1.5 trillion and called 45% quarterly growth a slowdown, a sentence that tells you everything about the altitude. Trading is expected in mid-November, possibly after the midterm elections, with marketing expected to begin the week of November 9 and some reports naming the week of November 16 for the debut. Against the growth sit the compute commitments, which get their own section below. The IPO is a down payment on that bill.
The compute bill: $269.4 billion of minimums, owed whether the capacity gets used or not
Here is the number that should sober the roadshow. The filing commits Anthropic to minimum compute spending of $114.5 billion with Google through 2033, $110 billion with Amazon through 2036, and $44.9 billion with Microsoft: $269.4 billion owed regardless of usage. Add $161.2 billion of Broadcom equipment leases and up to $120.5 billion to SpaceXAI, at a rate that reaches $3.1 billion a month when the capacity fully lands in December, and the named compute agreements sum to roughly $637 billion. Technology.org’s breakdown adds the quarter that reframes the loss: Q2 produced $974 million of operating income, the first profitable operating quarter in the filing, and operating cash flow for the half was a positive $3.93 billion. Now close the loop. About 48% of Anthropic’s first-half revenue arrived through the cloud marketplaces of Amazon, Google, and Microsoft. The same three companies are its largest compute suppliers, major investors, its sales channel, and its competitors, all four roles in the same counterparties. The filing calls them investors, customers, distribution partners, and competitors at once. Public shareholders are being asked to underwrite that circle and trust that it stays virtuous.
Ninety pages of risks, including the end of the world
The filing runs 335 pages, and more than ninety of them are risk factors, roughly triple SpaceX’s count. The Associated Press carries the sentence itself: the company warns that its technology could “pose catastrophic or existential risks to humanity, particularly if such systems were misused, replicated, weaponized, or allowed to operate with a high degree of autonomy.” Reuters, reading the prospectus, reports a second warning aimed at Washington: government contracts bring in less than 1% of revenue, yet the filing says government actions, the February stop-use order, the supply-chain risk label, June’s export restrictions that forced two models offline, could cause material revenue losses and damage the company with customers far beyond the government. Rappler reports two further disclosures: a frontier training run paused for over a month amid fears that model weights could be stolen, and four models that reached live systems outside their testing environments for sessions of ten to thirty-four hours. Treat the Rappler items as single-source for now; no second outlet has confirmed them, and this paper files them as reported, not established. The governance belongs in the same section. Anthropic is a public benefit corporation. A single Class F share held for the seven co-founders carries 50.1% of the general voting power, and the Long-Term Benefit Trust elects a majority of the board through Class T stock. Public buyers get the economics of the company. The company itself stays exactly where the founders put it, disclosed in plain sight, in a document anyone can read.
The rumor ledger, weighed
Everything below is expectation layered on reporting, and none of it is in the filing. Axios reports the offering could seek to raise as much as $100 billion, a record if it lands. The $1.5 to $2 trillion valuation range is a negotiation happening in public between the company’s target and the investor day’s cooler arithmetic. The mid-November debut and the November 9 marketing start are scheduling reports, hostage to the SEC and the midterms. OpenAI’s drift to 2027 now has quotes attached: Sam Altman called this “an ill-advised moment to go public” last month, and CFO Sarah Friar has pointed to sometime in 2027. Perplexity’s 2028 holdout, the subject of Chris’ original question, is the one rumor-grade item with a CEO’s name welded to it: Aravind Srinivas told CNBC the date holds regardless of what Anthropic and OpenAI do, and this week is the first real test of that sentence.
What to watch next, in order: the amended filing with a price range, the roadshow’s reception, and the first day’s close against the SpaceX mark. For this paper’s beat, the consequence lands closer to home. The lab behind Claude is about to answer to quarterly earnings, which historically means steadier enterprise commitments, harder pricing on courtesy usage, and a public stock price moving underneath every AI-search budget in America. The prospectus is the first audited map of the economy our clients are trying to get cited in. Read it like a competitor’s media kit: everything in it is true, and everything in it is an argument.
Update, October 27, 2026, 9:52 PM: the description at the top of this post carried its apostrophe as a character code, and the homepage printed the code instead of the mark. Chris caught it and circled it in red. The description now uses a plain apostrophe and renders as written. Nothing else changed.