Anthropic's IPO filing reveals the company has locked in $518 billion in AI infrastructure commitments from clients and infrastructure partners.

Anthropic expects to commit at least $518 billion to AI infrastructure over the next decade, according to disclosures in the company's confidential IPO prospectus reviewed by Reuters. The scale of these commitments reflects the computing infrastructure Anthropic expects to require as it expands its AI models and services.

More significant than the headline figure is the structure of those obligations. Roughly 80% of the total is either non-cancelable or requires the company to make payments regardless of how much computing capacity it ultimately uses. Anthropic told prospective investors that access to computing power is becoming a central constraint on AI development, with future demand for advanced AI systems expected to exceed available supply.

A substantial portion of Anthropic's infrastructure obligations stems from relationships with major cloud providers. The company expects to spend at least $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft under long-term infrastructure service agreements. Google's contract runs from April 2026 through July 2033, while Amazon's spans May 2026 through April 2036. If Anthropic's actual spending falls below the required amount, it must pay Google the difference—similar terms apply to Amazon. Microsoft's commitment runs from November 2026 through May 2033 and is non-cancelable except in the event of an uncured material breach by Microsoft. These arrangements provide Anthropic with long-term access to computing resources while creating substantial financial obligations that persist even if actual usage falls below expectations.

Beyond cloud services, Anthropic disclosed approximately $161.2 billion in Broadcom-related equipment lease obligations, which are largely non-cancelable. Together with its cloud commitments, these obligations represent a significant shift toward securing computing capacity through long-term contractual arrangements rather than relying entirely on flexible, on-demand cloud usage.

Anthropic also disclosed additional infrastructure relationships with xAI and AMD. The company has agreements with xAI that could result in up to $84.5 billion of spending through 2029 for NVIDIA-based computing capacity—notably, these agreements are largely cancelable with 90 days' notice. AMD has committed to purchasing up to $5 billion of Anthropic stock and supplying AI computing capacity expected to exceed $20 billion. These arrangements give Anthropic access to multiple sources of computing infrastructure as it seeks to reduce reliance on any single hardware or cloud provider.

Anthropic is increasingly building its own AI infrastructure, combining dedicated data centers with directly leased chips and other computing equipment. This shift is partly driven by the need to secure sufficient capacity as AI model demand grows. The company has highlighted the complexity of relying on Amazon, Google and Microsoft, which occupy several roles in its ecosystem as investors, customers, cloud infrastructure providers, distributors and competitors developing their own AI models. Anthropic warned that these relationships could create misaligned incentives and identified the risk that third-party compute could be curtailed, repriced or terminated.

For rapidly scaling AI companies, the economics of frontier AI increasingly extend beyond model development and software. Building and operating advanced AI systems requires large amounts of accelerators, data-center capacity, networking infrastructure and power. Securing those resources typically requires long-term commitments well before all capacity is actually consumed. Anthropic's strategy provides greater certainty around future computing availability while creating obligations that remain in place regardless of actual usage changes. With roughly 80% of its disclosed $518 billion infrastructure plan either non-cancelable or payable regardless of usage, Anthropic is committing substantial resources to ensuring access to the computing infrastructure it expects to need as its AI systems continue to scale.