Anthropic expands Amazon compute deal to up to 5GW

Anthropic and Amazon expanded their compute and investment partnership, adding up to 5 gigawatts of new AWS capacity for training and running Claude and committing Anthropic to spend more than $100 billion on AWS technologies over the following decade. Amazon paired the expansion with new money: $5 billion invested immediately, with up to $20 billion more tied to commercial milestones, on top of the roughly $8 billion Amazon had already put into Anthropic since 2023.

The new capacity runs on Amazon’s own Trainium chips rather than third-party GPUs, extending a relationship in which Anthropic already operated more than a million Trainium2 chips for training and serving Claude. The companies said significant Trainium2 capacity would come online in the second quarter of 2026, with Trainium3 following later in the year, and that nearly a gigawatt of combined Trainium2 and Trainium3 capacity was expected to be running by the end of 2026 — well short of the full 5GW figure, which represents a multi-year ceiling rather than an immediate commitment. The deal also covers Trainium4, a chip generation not yet available at announcement. Alongside the compute expansion, Anthropic gained access to inference capacity in Asia and Europe and a version of the Claude Platform integrated directly into AWS billing and account controls.

The announcement extended a pattern in which frontier labs increasingly tie themselves to a single hyperscaler’s custom silicon in exchange for capital and guaranteed capacity, mirroring OpenAI’s dependence on Microsoft’s infrastructure and Nvidia chips. It came five weeks before Anthropic’s Series H funding round, in which Amazon’s existing commitments were counted among the round’s hyperscaler backing.