Anthropic's roadshow: $74 a share, $17.0 billion of revenue and a $21.9 billion accounting charge
Anthropic's roadshow site prints $17.0B of Q3 revenue, a $23.3B net loss and a 15% adjusted margin beside a $74 assumed IPO price that implies about $968B.

Anthropic's IPO marketing site went live this week with an eleven-chapter roadshow video, a founders' letter, a fact sheet and a calendar: the price is expected to be set on November 18, and shares are expected to begin trading on November 19 under the ticker ANT. The site is not a filing, but the prospectus it links to now carries a working number, $74.00 a share, and its highlights print $17.0 billion of quarterly revenue beside a $23.3 billion net loss.
A retail-facing page with a deadline attached
The roadshow began on November 10, according to the site's own FAQ, and most of the page is written for someone who has never opened a prospectus. Participation runs through a brokerage — Schwab, Fidelity and Robinhood are listed with links — by submitting an indication of interest, which the text treats as a request rather than an order: "There is no guarantee that any investor will receive all or any of the shares they apply for."
Beyond the document, the site supplies sequencing. The video runs opportunity, product, frontier and safety research, five customer segments including Bristol Myers Squibb, Accenture and AIG, compute, then financial model. The downloadable fact sheet is the more useful artifact: it prints the quarterly revenue series, the compute expense behind it, and the adjusted-margin series back to the first quarter of 2025.
Seventy-four dollars, in a document that still says "assumed"
The October S-1 left the offer price blank, which is normal at that stage, and it still meant the central number was missing. The prospectus now on the site fills it in: 825,000,000 Class A shares, an assumed price of $74.00, and estimated net proceeds of about $84.6 billion, or $93.7 billion with the underwriters' option for a further 123,750,000 shares. The net figure exceeds 825 million times $74 because it includes a concurrent private placement of Class N stock, in which Google's purchase exhausts its option under a 2026 agreement.
Put the placeholder against the share count and it becomes a valuation. After the offering and the placement Anthropic expects 13,086,672,593 shares outstanding across its classes; at $74 that is roughly $968 billion, well below the $1.5 trillion to $2 trillion that investors and the Wall Street Journal were discussing in October. The gap is not a correction: $74 exists to compute proceeds, not to promise where the book clears.
Revenue rose twelvefold; revenue per compute dollar did not
The fact sheet's quarterly series reconciles to the audited half-year: $4.7 billion in the first quarter of 2026 plus $11.6 billion in the second is the $16.3 billion the S-1 reported for the six months to June. The third quarter adds $17.0 billion, 12.1 times the $1.4 billion of a year earlier — the ">12x" on the highlights page.
| Quarter | Revenue | Compute & infrastructure | Revenue per $1 of compute |
|---|---|---|---|
| Q1'25 | $0.4B | $0.9B | $0.40 |
| Q2'25 | $0.8B | $1.4B | $0.57 |
| Q3'25 | $1.4B | $2.0B | $0.68 |
| Q4'25 | $2.0B | $2.9B | $0.69 |
| Q1'26 | $4.7B | $3.4B | $1.40 |
| Q2'26 | $11.6B | $5.6B | $2.08 |
| Q3'26 | $17.0B | $9.9B | $1.72 |
The bottom row is the one nobody put in the video. Compute and infrastructure spending rose to $9.9 billion from $5.6 billion while revenue rose to $17.0 billion from $11.6 billion, so revenue per dollar of compute fell to $1.72 from $2.08 — the first sequential decline in the series. The ">4x compute efficiency gain since Q1'25" is measured against the $0.40 ratio in that quarter; it says nothing about the last two.
The $17.0 billion also closes a smaller gap. Our October 16 report carried $16.8 billion, relayed by investors who attended the October 14 meeting rather than published by the company. The site prints $17.0 billion for the same quarter, with no reconciliation: 1.2%, the difference between a number in a conference room and a number in a document.
Fifteen percent adjusted, negative three percent as filed
The highlights page shows two operating margins and lets the reader pick. Adjusted operating margin was 15% in the third quarter. Operating margin was (3)%. On $17.0 billion of revenue, eighteen points is roughly $3 billion of difference in three months, and the only explanation on the page is a footnote pointing to the prospectus section on "Non-GAAP Financial Measures."
The adjusted series has already turned over. Minus 221% in the first quarter of 2025, then minus 129%, minus 93%, minus 84%, minus 13%, then 19% in the second quarter of 2026 and 15% in the third. The company's preferred metric peaked one quarter before the quarter it is advertising to retail. That does not make 15% wrong. It makes it a number whose best quarter is behind it, presented without the footnote's contents.
Twenty-three billion of loss that is mostly a marking
The net loss is the figure that would stop a first-time reader, and the footnote defuses most of it: $23.3 billion includes a $21.9 billion non-cash charge tied to the change in fair value of convertible notes, derivative liabilities and SAFEs, which rises when the company's valuation rises. What is left is about $1.4 billion of everything else — real, and larger than the $0.5 billion operating loss the (3)% margin implies.
That charge runs both ways: the same line shrinks if the valuation falls after listing. The obligation book underneath it does not. The fact sheet prints roughly $900 billion of compute commitments as of September 30 and afterwards, with about $70.7 billion due in the twelve months after that date and about $451.6 billion by the end of 2030, against $96 billion of cash. Our earlier reading of the S-1 put commitments at $542 billion to $567 billion. Either several hundred billion of capacity was signed in six weeks or the two counts measure different things; the page does not say which.
Logos, and no contract values
The customer studies are what a procurement team will read, and the numbers are specific. Legora serves more than 1,500 law firms and in-house teams and reports 4.3 non-billable hours recaptured per lawyer per week in a study of 31 firms, with Claude Code seats growing from about 50 to 700. Eli Lilly gives 50,000-plus employees Claude Enterprise access and says more than 85% in key functions engage with it.
Every one of those figures is the customer's or the vendor's, published on a page whose purpose is to sell shares, and none arrives with a contract value or a renewal term. The duller counts sit on the fact sheet: more than 11,900 customers spending at least $100,000 on a trailing-twelve-month basis, about 1,900 above $1 million, roughly 250 above $10 million, net revenue retention near 900%. An average Fortune 500 customer spending under 0.1% of its total expenses on Claude is a base with room to grow — and, on the same evidence, one with no particular reason to stay.
What November 18 actually settles
Tuesday's pricing will resolve the things that are still genuinely open. The price, and with it the valuation: the prospectus's $74 is an assumption, and the real number comes out of the book. Demand, though only second-hand, since book coverage is reported by people told about it and never published. Dilution, and whether the extra 123.75 million shares are sold. And the use of proceeds, which the prospectus sets out and the site does not summarise.
What pricing day will not settle is the argument the highlight page makes against itself. The $3 billion bridge from (3)% to 15% needs the reconciliation buried in the MD&A. The ~900% expansion figure needs a cohort old enough to renew. Revenue per compute dollar needs a quarter where it climbs again, because the last two went the other way. The calendar tells you when the price arrives. Whether the business underneath has stopped spending more than it earns is a question for the first public quarter, and the threads already arguing about it are not waiting for the answer.


