动察 Beating AI Flash: The Information's analysis found that Anthropic has signed at least 14.8GW of compute capacity over the past 11 months, to be deployed gradually over the coming years. Based on current public contracts, the potential total cost could reach as high as $517 billion, with most spending occurring over the next decade. Combined with the 1–2GW locked in before October last year, Anthropic's total signed compute capacity stands at approximately 16GW.

This expansion is primarily driven by demand for Claude. This year, growth in Claude Code and Cowork has exceeded Anthropic's expectations, prompting the company to aggressively secure compute resources. Amazon's new agreement provides up to 5GW, while Google and Broadcom add another 5GW; Microsoft and NVIDIA contribute up to roughly 1GW. Anthropic has also leased all of SpaceX's Colossus 1 compute capacity, securing over 220,000 NVIDIA GPUs in one go, including H100, H200, and GB200 models.

Yet despite this massive investment, Anthropic's compute planning may still lag behind OpenAI. Anthropic has locked in approximately 16GW, with many contracts extending beyond 2030; OpenAI has set a target of 30GW by 2030 for investors, with cumulative compute spending projected at around $750 billion by that year.

The $517 billion figure represents The Information's estimate of the potential maximum cost based on existing cloud service, chip, and data center contracts. Some compute capacity may be delayed or even unused, and certain contracts allow for early cancellation.