Anthropic's $518 billion AI infrastructure plan relies on mostly non-cancelable partner contracts

Anthropic's confidential IPO prospectus outlines a $518 billion, ten-year AI build-out plan, with roughly 80% of the amount bound by contracts that cannot be cancelled or must be paid regardless of consumption. The company will allocate $111.1 billion to Alphabet's Google, $110 billion to Amazon, and $31.4 billion to Microsoft under long-term service obligations spanning 2026-2036. It also holds $161.2 billion of non-cancelable equipment lease commitments with Broadcom.

Further agreements include up to $84.5 billion of spending with Elon Musk's xAI for Nvidia-based compute, and a partnership with Advanced Micro Devices that involves a $5 billion stock purchase and more than $20 billion in AI computing capacity. Anthropic notes that reliance on Amazon, Google, and Microsoft—who also act as investors, customers, and rivals—creates alignment challenges that could affect its operations if access to third-party compute is reduced or re-priced. The filing, submitted confidentially to the SEC in June, has not been publicly released.