Reuters’ reporting on the Anthropic IPO prospectus describes a fast-growing AI company with an unusually large set of future infrastructure obligations. The colorful characterizations circulating around the filing are commentary. The reported facts are plainer: Anthropic recorded nearly $4.6 billion in 2025 revenue while disclosing at least $518 billion in planned spending over a decade on AI infrastructure.
Anthropic filed confidentially for an initial public offering with the Securities and Exchange Commission in June, Reuters reported. Its filing had not been publicly released when Reuters reviewed the prospectus, so the disclosures remain reported accounts of a confidential document rather than a public SEC filing.
Reuters reported that revenue rose roughly 12-fold in 2025 to nearly $4.6 billion. The prospectus also showed a net loss of $42 billion for the year. Separately, Reuters reported an operating loss of more than $8 billion, excluding writedowns of liabilities that were mostly connected to earlier fundraising.
How much of Anthropic’s $518 billion infrastructure plan is non-cancelable?
About 80% of the $518 billion total is non-cancelable or must be paid whether Anthropic uses the capacity or not, according to Reuters’ account of the prospectus. That makes the figure a future, multi-year commitment, not a measure of 2025 spending or a current-year loss.
The reported commitments include at least $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft in long-term infrastructure obligations, generally spanning the next seven to 10 years and payable regardless of use. Anthropic also reported about $161.2 billion in Broadcom-related equipment lease obligations that are largely non-cancelable.
Reuters said Anthropic’s Google commitment runs from April 2026 through July 2033, its Amazon agreement from May 2026 through April 2036, and its Microsoft commitment from November 2026 through May 2033. Under the Google arrangement, Anthropic said it must pay the difference if actual spending falls short; it said similar conditions apply to Amazon. The Microsoft agreement is non-cancelable except in the event of an uncured material breach, Reuters reported.
What risks did Anthropic identify in the filing?
Anthropic told prospective investors that compute availability, rather than demand, is likely to limit advanced AI development. It is also moving beyond a cloud-only approach toward dedicated data centers and directly leased chips, Reuters reported.
The same arrangements leave Anthropic dependent on companies that occupy several roles at once. Amazon, Google and Microsoft provide infrastructure and distribution, and are also investors, customers and developers of competing AI models. Anthropic said their incentives may not fully align with its own, and warned that curtailed, repriced or terminated third-party compute access could harm its business and financial results.
The prospectus identified a separate revenue risk: two customers supplied nearly one-quarter of 2025 revenue, Reuters reported, while many of Anthropic’s largest clients were not tied to long-term contracts. Together, the disclosures put the essential investor question in view: whether growth can keep pace with a set of long-duration obligations that are, in many cases, payable even if the capacity goes unused.
This story draws on original reporting from Daring Fireball.