The Information reported that Anthropic signed at least 14.8GW of new compute capacity over the past 11 months, with that capacity scheduled to be used over the next several years.
Based on publicly known contracts, the potential maximum cost of those commitments could reach $517 billion, with most of the spending expected over the next decade. Adding the 1GW to 2GW Anthropic had already secured before October last year brings the company’s total contracted compute footprint to about 16GW.
Claude demand drove the latest expansion
The report said the push was largely driven by demand for Claude. Growth in Claude Code and Cowork this year came in above Anthropic’s expectations, prompting the company to move aggressively to secure more compute.
Amazon, Google, Broadcom, Microsoft, and NVIDIA are part of the supply mix
Under current agreements, Amazon could provide as much as 5GW. Google and Broadcom together add another 5GW, while Microsoft and NVIDIA account for about 1GW.
Anthropic has also leased the full capacity of SpaceX Colossus 1, giving it access to more than 220,000 NVIDIA GPUs, including H100, H200, and GB200.
Still may remain below OpenAI’s planned scale
Even with commitments of that size, Anthropic’s compute buildout may still fall short of OpenAI’s stated plan. Anthropic has locked in about 16GW, with many of those contracts extending beyond 2030. OpenAI, by contrast, has told investors it aims to reach 30GW by 2030 and expects cumulative compute spending of about $750 billion by that year.
The report added that the $517 billion figure represents The Information’s estimate of the potential upper-end cost based on existing cloud services, chip, and data center contracts. Some of the compute could be delayed or never used, and some agreements allow early cancellation.