Amazon to invest up to $25 billion more in Anthropic, locking in a 10-year AWS contract

Amazon has announced an investment of up to $25 billion in Anthropic, the maker of the Claude language model, in exchange for a 10-year agreement to run on AWS cloud infrastructure. It is one of the largest corporate bets in the history of generative AI, and it cements AWS as the primary cloud partner of one of the world's fastest-growing AI startups.

The new deal is Amazon's second consecutive megainvestment in a leading AI lab within two months. In February 2026, the company put about $50 billion into OpenAI, a structurally similar arrangement that ties the lab to long-term use of cloud infrastructure in exchange for funding. Amazon is now an anchor investor in both of the main players in the race for frontier language models, and that is no accident but a deliberate strategy.

Under the partnership, Anthropic commits to using AWS Trainium chips, including the future Trainium4 generation, as its primary computing base. In turn, Amazon guarantees the company access to capacity of up to 5 gigawatts. For scale: that is comparable to the output of five large thermal power plants, the kind of energy that industrial-grade AI clusters need to train and serve next-generation models.

Anthropic is growing fast. Its annual revenue has reached $30 billion, a figure that would have looked fantastical a year or a year and a half ago. Enterprise customers use Claude through Amazon Bedrock, and developers use it directly through the Anthropic API. Unlike OpenAI, which bet on mass consumers through ChatGPT, Anthropic focuses on the enterprise segment and positions itself as a responsible AI developer, with an emphasis on model safety and interpretability.

The deal makes strategic sense for Amazon on several levels. First, it strengthens AWS's competitive position against Azure, the main cloud partner of OpenAI and Microsoft, in the fight for enterprise demand for AI infrastructure. Contracts for training and hosting large language models are becoming one of the fastest-growing segments of the cloud market.

Second, tying Anthropic to Trainium chips helps scale Amazon's own semiconductor business, which competes directly with the dominant NVIDIA. Every petaflop of compute on Trainium is a potential contract that did not go to a competitor.

Finally, the 10-year horizon creates a predictable stream of infrastructure revenue regardless of how the AI model market itself develops.

For Anthropic, the terms address a key problem: a shortage of computing resources. Training frontier models requires enormous capacity that is scarce and expensive. The guaranteed 5 gigawatts remove one of the main operational risks for a fast-growing company and let it plan training of future Claude generations without worrying about GPU cluster availability. Anthropic formally remains an independent lab, despite deep integration with the infrastructure and strategy of a single investor.

The deal reflects a key dynamic of the current AI race: competition is playing out not only over model quality but over control of computing infrastructure. Cloud providers Amazon, Microsoft and Google are actively fighting to make their platform the standard for training and deploying next-generation AI systems.

Amazon's investment in Anthropic is not just a financial bet. It is an architectural decision about what the AI ecosystem will look like in ten years: who owns the chips, who runs the cloud, who builds the models, and how those three layers will be connected.