Anthropic is committing $100 billion to AWS infrastructure over the next decade to secure up to five gigawatts of compute capacity for training and deploying Claude as demand for its models accelerates.
The agreement includes new Amazon Trainium2 capacity coming online in the first half of this year, nearly one gigawatt of combined Trainium2 and Trainium3 capacity by the end of 2026, and expanded deployment through Amazon Bedrock.
Amazon also committed an additional $5 billion investment in Anthropic, with the option to invest up to $20 billion more.
The announcement is fundamentally about infrastructure scale. Anthropic is moving to secure long-term access to compute at a time when model providers are increasingly constrained by GPU shortages, rising inference costs, and growing enterprise demand.
Under the deal, Anthropic will use AWS as its primary cloud and training provider for mission-critical workloads, spanning Amazon’s Graviton processors and Trainium chips from Trainium2 through Trainium4, with options to adopt future generations of Amazon silicon.
The companies have worked together since 2023, and Anthropic said more than 100,000 customers now run Claude through Amazon Bedrock. Their earlier collaboration included Project Rainier, which Anthropic described as one of the world’s largest compute clusters. The company said it is currently using more than one million Trainium2 chips to train and serve Claude.
A major operational shift is also coming to enterprise customers. Anthropic said the full Claude platform will soon be available directly within AWS, allowing enterprises to access Claude through existing AWS accounts, billing structures, and governance controls without requiring separate contracts or credentials. That move reduces procurement friction for large organizations already standardized on AWS infrastructure.
Anthropic also said it is expanding inference capacity in Europe and Asia to support growing international demand. The company reported that annualized revenue has surpassed $30 billion, up from roughly $9 billion at the end of 2025, while usage across free, Pro, Max, and Team tiers has created reliability issues during peak periods.
The broader signal is clear: frontier model companies are increasingly behaving like infrastructure buyers as much as software vendors. For Anthropic, securing long-term compute access appears less about optional expansion and more about ensuring Claude remains reliably available as enterprise adoption and consumer usage continue to rise.