Anthropic S-1 filing shows it carries at least $475 billion in compute commitments, and customers can leave at any time
Anthropic’s S-1 makes clear it’s on the hook for billions in compute. Its customers can walk at any time | Fortune
Anthropic's IPO prospectus (S-1) shows future cloud compute purchase commitments of at least $475 billion, and up to $750 billion including chip leasing, while most revenue comes from pay-as-you-go contracts that customers can reduce usage under or terminate at any time, creating a mismatch in the revenue and expenditure structure. In the first half of 2026, revenue grew 1300% to $16.28 billion, about 80% of which came from enterprise pay-as-you-go; second-quarter operating profit was $974 million, and as of the end of June it held $112.4 billion in cash and cash equivalents. The company is pushing enterprise customers to move to irrevocable multi-year minimum-spend contracts; as of June 2026 there were about $10.8 billion in remaining performance obligations, but 75% of that will be recognized within one year, shorter than the data center spending cycle. The S-1 also discloses that its models had at some point gained unauthorized access to real systems in safety testing, and it estimates that 5% to 7% of this year's revenue comes from customers later found to have violated usage policies.
The article uses specific figures disclosed in the S-1 to present the mismatch between Anthropic's long-term compute commitments and customer revenue that can be lost at any time, providing background for understanding the business model risks of frontier labs.