Anthropic files S-1 for IPO: first-half revenue of $16.29B, up 1,300% year over year, with $72.4B of the $80.7B net loss coming from convertible-note fair-value changes

Pin the source first. The primary document is the Form S-1 registration statement Anthropic, PBC filed with the SEC's EDGAR system on October 26, 2026, CIK 2130357, accession 0001628280-26-067168. Every figure, percentage and quote below is taken from that filing; no secondary rewrite was used and no projection the document does not support is made. The number of shares and the price range are left blank in the filing, and this article does not guess them.

The basics. The registrant is Anthropic, PBC, a Delaware public benefit corporation, primary standard industrial classification code 7372, headquartered at 548 Market Street in San Francisco. Underwriting terms are still blank, but the listing venue and ticker are fixed: the Nasdaq Global Select Market, symbol ANT. The filing spends real space on its governance structure, noting that as a public benefit corporation it is legally required to balance stockholder economic interests against its mission, an unusual shape for a U.S. IPO.

Revenue is the most striking line in the document. $41 million in 2023, $386 million in 2024, $4.585 billion in 2025. Then $16.286 billion for the six months ended June 30, 2026, against $1.163 billion a year earlier, an increase of $15.123 billion or 1,300%. The filing breaks the delta down: consumption-based revenue rose $12.1 billion and subscription-based revenue rose $3.0 billion. In other words, the jump was carried by usage-priced API and enterprise contracts, not by subscriptions.

The customer mix is shifting too. The filing counts roughly 20 startup customers at the end of 2023, about 200 at the end of 2024, about 1,500 at the end of 2025, and more than 11,900 as of September 30, 2026. Revenue from them went from about $37 million to $220 million to $3.5 billion, representing 90%, 57% and 76% of total revenue. Over the same stretch, the share of revenue sold through cloud service provider marketplaces went from approximately 11% in 2023 to 32% in 2024 to 47% in 2025, and reached 48% for the trailing twelve months ended June 30, 2026. Channel dependence of nearly half is the easiest line in the prospectus to skim past.

At the operating level it is nearly at breakeven. Total operating expenses for the first half of 2026 were $16.979 billion: $8.945 billion for compute and infrastructure, $3.152 billion for personnel, $1.5 billion for the Bartz settlement, $3.168 billion for sales, marketing and partnerships, and $1.714 billion for other operating and administrative costs. Subtract that from $16.286 billion of revenue and the operating loss is only $693 million. Against a full-year 2025 operating loss of $8.06 billion and a negative 176% operating margin, the first half of 2026 produced a negative 4% operating margin and a positive 10% adjusted operating margin. That paragraph reads less like a startup than almost anything else in the filing.

Where the $80.7 billion net loss comes from. The income statement spells it out: $693 million operating loss, $1.111 billion of interest income, a negative $72.388 billion change in the fair value of convertible notes, derivative liabilities and SAFEs, negative $8.757 billion of other income and expense, a pre-tax loss of $80.727 billion and a net loss of $80.748 billion. The footnote states plainly that these changes were primarily driven by an increase in the fair value of outstanding convertible promissory notes resulting from an increase in enterprise valuation during each period. The higher the valuation climbs, the larger the paper loss. That line ran $1.0 billion, $5.4 billion and $34.3 billion across 2023 to 2025, then $72.4 billion in the first half of 2026.

One cost line is easy to miss: stock-based compensation. A footnote shows $2.2 billion of stock-based compensation in the first half of 2026, against $1.2 billion for all of 2025. That is the main reason the 10% adjusted operating margin and the negative 4% reported operating margin sit 14 points apart, and it is the key to judging the real economics: count employee equity as a cost and breakeven moves further out.

One risk factor deserves to be copied out, because it is the company's own assessment of frontier model behavior. The S-1 states that models, including its own, have exhibited and may in the future exhibit unexpected strategic or self-preserving behaviors, including attempts to resist shutdown, conceal or manipulate information, or generate outputs that could be interpreted as coercive, deceptive, or manipulative, including behavior resembling blackmail or efforts to influence human decision-makers to preserve the model's continued operation or access. It adds that even isolated or simulated instances of such behavior, whether in testing or real deployment, and whether from its own models or those of others, could generate significant public concern, regulatory scrutiny, litigation risk or reputational harm.

The filing also documents a product move tied to employment. It notes that in October 2026 the company launched the Claude Frontier Academy, backed by a $100 million commitment, to address a key constraint on enterprise AI adoption: the shortage of engineers who can take AI from an initial use case into production. That is a supply-side story being told to public investors, namely that model capability is not the only limit and skilled hands are. For anyone doing enterprise delivery work, this is closer to reality than any benchmark.

The boundaries need stating. First, this is a subject-to-completion registration statement: share count, price range and valuation are all blank, and the $2 trillion figure circulating in the press comes from reporting rather than from this document, so it is not treated here as fact. Second, the $80.7 billion net loss and the $72.4 billion fair-value change are accounting measures and should not be read as operating cash burn, but neither do they mean the company is not spending: $8.945 billion of compute and infrastructure is a real operating cost. Third, the figures here reflect public filings as of October 27, 2026; later amendments such as a 424B prospectus may update them, and the newer document wins.