Anthropic and Amazon expand collaboration for up to 5 gigawatts of new compute

Source: Anthropic Date: 2026-04-20 URL: https://www.anthropic.com/news/anthropic-amazon-compute

Summary

Anthropic and Amazon announced an expanded partnership securing up to 5 GW of compute capacity for training/deploying Claude through 2026. Anthropic commits $100B+ in AWS technologies over ten years. Amazon invests an additional $5B immediately (with up to $20B more possible). New hardware: Trainium2 and Trainium3. The full Claude Platform will be available directly within AWS with unified account controls and billing. Run-rate revenue cited as $30B+ (up from $9B at end of 2025); infrastructure reliability strained during peak usage periods.

Implications

• Capital / compute infrastructure thread. $100B+ Anthropic AWS commitment over ten years is the largest cloud spending commitment Anthropic has made — and the most public. This level of capital commitment creates financial dependency on AWS that’s difficult to unwind.
• 5 GW compute. Adding 5 GW alongside Google’s separate 1 GW TPU deal means Anthropic’s compute roadmap involves 6+ GW of new capacity — an extraordinary infrastructure investment that would rank Anthropic among the largest AI compute operators globally.
• $5B immediate + $20B possible. The tiered Amazon investment ($5B confirmed, $20B contingent) is a performance-linked investment structure — Anthropic likely needs to hit growth targets to unlock the full $20B. This creates an incentive alignment between Amazon and Anthropic’s revenue growth.
• Claude Platform within AWS. Native AWS integration with unified billing means enterprise customers can purchase Claude directly through their existing AWS accounts — removing the separate Anthropic billing relationship. This is the same distribution lock-in strategy as Microsoft Foundry.
• Infrastructure reliability strain. Acknowledging peak usage infrastructure problems is unusual transparency — it suggests demand has outpaced supply, and this deal is specifically the capacity response.
• Watch: whether Trainium2/3 performance justifies the AWS dependency; how the $100B AWS commitment interacts with the Google Cloud TPU partnership; when the reliability issues are resolved at scale.