TL;DR:

  • Anthropic has committed more than £75bn (0bn) to AWS over the next ten years, securing up to 5GW of training and inference capacity on Trainium2 through Trainium4 silicon.
  • Amazon is investing a further £3.8bn (bn) now with up to £15bn (bn) to follow, stacking on top of the £6bn (bn) already committed.
  • Inference capacity will expand in Europe and Asia, bringing Claude closer to UK enterprises already running on Amazon Bedrock.

Anthropic’s expanded compute pact with Amazon is one of the largest single infrastructure commitments in the AI industry to date, and lands at a point where Anthropic’s self-reported run-rate revenue has climbed from around £6.8bn (bn) at the end of 2025 to more than £22.6bn (bn) today. The deal is part capacity agreement, part capital injection, and part distribution play — all three tied into a single AWS-centric bet.

A capacity commitment, not a cloud headline

The headline £75bn figure covers Graviton general-purpose silicon alongside Trainium2 through Trainium4 generations, with options on future Amazon chips. Anthropic says “significant” Trainium2 capacity is already coming online this quarter, and that roughly 1GW total will be live by year-end. That places AWS firmly at the centre of Claude’s training pipeline, alongside the existing million-plus Trainium2 chips already in use through Project Rainier.

What changes for UK customers

For UK buyers, two details matter more than the headline. First, the agreement explicitly extends inference capacity into Europe — Claude service performance in-region has been a repeated pain point as consumer growth strained peak-hour reliability. Second, Anthropic is rolling Claude Platform into AWS directly: same account, same billing, same governance controls, currently in private beta. That simplifies procurement for organisations already standardised on AWS under existing Crown Commercial Service or FCA-registered frameworks — a meaningful friction removal compared with today’s multi-contract setup.

Looking forward

Anthropic is now the only frontier lab with first-party availability across AWS, Google Cloud and Azure, a positioning that hedges customer lock-in concerns at the expense of deeper single-vendor integration elsewhere. The real test is whether this compute uplift closes the reliability gap flagged on Pro, Max and Team tiers during peak hours — UK enterprise buyers planning Claude rollouts will want to see service-level stability hold through summer before committing production workloads.

Share this article