Anthropic expects to spend at least $518 billion on AI infrastructure with six partners over roughly a decade, with about 80 percent of the total covered by agreements it cannot cancel or must pay regardless of usage. The figures come from a confidential IPO prospectus reviewed by Reuters; Anthropic has not released the filing publicly.

The largest disclosed cloud obligations are at least $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft, according to the filing. Those service agreements run across seven to 10 years. Anthropic says it would owe Google the difference if its actual spending fell short of the contracted amount, and describes similar terms with Amazon. It says the Microsoft agreement is non-cancelable except if Microsoft fails to remedy a material breach.

The prospectus also lists about $161.2 billion in equipment lease obligations tied to Broadcom that are largely non-cancelable. Another agreement could send up to $84.5 billion to Elon Musk’s xAI for Nvidia-based computing capacity through 2029. Unlike the biggest cloud and lease commitments, that arrangement is largely cancelable on 90 days’ notice. The document says AMD has committed to supply capacity expected to exceed $20 billion and to buy up to $5 billion in Anthropic stock.

Anthropic says nearly a quarter of its 2025 revenue came from two customers. It warns that many large clients lack long-term contracts and could reduce or stop spending. Its own cloud and equipment agreements extend for up to a decade, with most of the projected payments mandatory.

The company spent $7.33 billion on compute and infrastructure in 2025, about three times its 2024 figure, the prospectus says. Revenue rose to nearly $4.6 billion that year. Anthropic reported a net loss near $42 billion, though roughly $34 billion came from an accounting charge related to financing that could convert into shares, rather than the cost of running the business.

Anthropic has also arranged private-credit financing for its compute expansion and is moving toward dedicated data centers and directly leased chips. The prospectus shows its continuing dependence on Google, Amazon and Microsoft. Between them, those companies are investors, customers, cloud providers, distributors and competitors, with incentives Anthropic says may not align with its own.

Anthropic announced its confidential S-1 submission in June, but the document itself is still unavailable for independent inspection. The company’s earlier deal for xAI’s Colossus capacity established one part of the relationship; the reported filing terms supply its potential dollar value and cancellation provisions.

Sources: Reuters, Reuters, Anthropic

–
By the Control Plane Editorial Team