On October 26, 2026, Anthropic PBC publicly filed Form S-1 with the SEC for an IPO: Class A shares are planned for listing on the Nasdaq Global Select Market under the ticker “ANT.” Along with the filing, the company disclosed its financials for the first time: revenue for the first half of 2026 reached $16.3 billion with a near-zero operating loss, while the record net loss of $80.7 billion was almost entirely formed by non-cash accounting revaluations of convertible instruments that grow with the company's valuation, rather than a loss of cash.

image
image
image

What happened

The filing was published on SEC EDGAR with the status “Subject To Completion”: this is a preliminary prospectus where the offering price and number of shares are left blank until subsequent editions of the document. Class A shares are planned for listing on the Nasdaq Global Select Market under the ticker “ANT.” Goldman Sachs, Morgan Stanley, and J.P. Morgan are leading the deal organization, joined by a syndicate of more than twenty banks, including Citigroup, Barclays, Wells Fargo, and BofA. The company's financial history was disclosed publicly for the first time: revenue grew from $41 million in 2023 and $386 million in 2024 to $4.59 billion in 2025 and $16.3 billion in the first half of 2026, representing a 1300% year-over-year increase. Compute and infrastructure costs for the same six months amounted to $8.9 billion, and the operating loss was $693 million. The net loss for the first half reached $80.7 billion, however, $72.4 billion of this amount is attributable to the non-cash change in fair value of convertible notes, derivatives, and SAFE instruments due to the rise in the company's valuation, rather than cash burn. In the previous 2025, according to the same file, the operating loss was $8.06 billion for the year.

Context

The market entry was prepared in an unusual configuration for an IPO. Anthropic is registered as a public benefit corporation, its capital is divided into Classes A, N, T, and F, and the majority of the board of directors is elected by the holder of Class T — the Long-Term Benefit Trust, meaning ordinary investors will have fewer corporate guarantees than in a standard offering. The filing also records a deep tie to hyperscalers: NPA lines of $77 billion link Anthropic with Google, Amazon, and Broadcom, meaning the expansion of the compute base relies on external power providers. The organizational scale grew rapidly: the staff increased from approximately 940 people at the end of 2024 to approximately 5,300 people by September 30, 2026.

Why this matters for the industry

For the industry, this is the first verifiable cross-section of a frontier lab's economics, replacing estimates and leaks. The ratio of revenue to compute costs for the first half of 2026 was approximately 1.8x ($16.3 billion vs. $8.9 billion) — the first public baseline for the payback of frontier model inference, and the popular thesis that scaling eternally consumes money and returns nothing is not directly confirmed by these figures. Calculations based on the file data show that compute and infrastructure account for approximately half of all operating expenses: the sum of revenue ($16.3 billion) and operating loss ($693 million) yields about $17.0 billion, meaning the lab's economy has essentially become an inference infrastructure economy, not just a research one. The offering itself will be the largest market test of the “controlled company” structure with a public benefit corporation: if the scheme with the Long-Term Benefit Trust survives the IPO and the first quarters of listing, it will set a precedent for a public frontier lab with a preserved long-term mandate. Public status will also add regularity: quarterly reports from the largest API player will make its pricing and product decisions more observable, and for the founders, the disclosed figures are a signal that the enterprise-AI market is colossal and nearly self-sufficient in cash flow.

Why this matters for users

For Claude users, nothing changes immediately: the S-1 is a financial document, not a release; tariffs, limits, SLAs, and the API remain the same, and the filing does not entail migrations or deprecations. The practical value for readers lies elsewhere. The file on SEC EDGAR is available as an open source of data on the ratio of revenue to compute costs for a frontier vendor — it can be used in one's own calculations of model operating costs. When reading headlines, it is worth remembering that the giant net loss figure is mostly an accounting revaluation tied to the rise in the company's valuation, so “shorting” Anthropic because of it is a misreading of the financials. For teams whose product critically depends on a single vendor, public status provides a reason to conduct a dependency audit: where is the fallback, and what would move to another API if conditions change. Next, one should watch for the price and number of shares to be filled in future editions of the prospectus, the listing date for “ANT,” and Q3-2026 figures. If the payback dynamics are maintained, servicing frontier models could become a marginable business, which historically would mean a reduction in per-token inference costs and more predictable prices for developers.

What is currently unknown / limitations

The key caveat is the single data point: conclusions are based on one half-year, and Q3-2026 figures are not yet in the current document, so the sustainability of the revenue-to-compute-cost ratio, as well as the absence of deceleration after the 1300% growth, is not yet proven. The $8.9 billion cost is given as an aggregate of compute and infrastructure without a breakdown between training and inference; with the next training cycle, this line item could spike, so it is premature to draw conclusions about inference margins. For the same reason, it should not be considered proven that the near-zero operating loss gives the leader room for pricing maneuvers. Finally, the prospectus is preliminary: the offering price, number of shares, and listing date will appear only in subsequent editions of the document, and how the controlled company structure will behave in public status will only be shown by the listing itself.

Sources

Author

Look at AI, editorial team