Anthropic · 2026-04-20 · major
Anthropic & Amazon: $100B AWS Commitment, 5GW Compute for Claude
Anthropic commits $100B+ to AWS over 10 years and secures up to 5GW of compute (Trainium2–4) for training and serving Claude. Amazon invests $5B immediately with up to $20B more tied to milestones. Run-rate revenue surpassed $30B.
Anthropic locks in up to 5 gigawatts of AWS compute in the largest cloud AI infrastructure deal announced to date.
Key specs
| Aws commitment | $100B+ over 10 years |
|---|---|
| Amazon investment (immediate) | $5B |
| Amazon total investment potential | up to $25B |
| Compute secured | up to 5GW |
| Anthropic run rate revenue | $30B+ |
| Trainium2 capacity online | Q2 2026 |
| Trainium3 capacity | H2 2026 |
What is it?
Anthropic and Amazon signed an expanded infrastructure agreement on April 20, 2026, giving Anthropic access to up to 5GW of compute capacity across Amazon's Trainium2, Trainium3, and Trainium4 chips. In return, Anthropic commits over $100 billion to AWS technologies over the next decade. Amazon invests $5B in Anthropic immediately, with the potential for up to $20B more tied to commercial milestones — on top of $8B previously invested.
How does it work?
Trainium2 capacity begins in Q2 2026; scaled Trainium3 follows in H2 2026. The full Claude Platform will integrate directly into AWS accounts with unified billing, access controls, and monitoring. International inference expansion covers Asia and Europe. Anthropic already runs over 1 million Trainium2 chips and is the largest customer of custom AWS silicon.
Why does it matter?
Anthropic's run-rate revenue jumped from $9B at end of 2025 to $30B — fivefold growth in under five months. Securing 5GW of dedicated compute locks in supply before Trainium capacity is claimed by others, and the $100B AWS commitment is the largest publicly announced AI infrastructure pledge in industry history. The deal quotes both Andy Jassy and Dario Amodei, signaling deep bilateral commitment.
Who is it for?
Enterprise teams using Claude on AWS; ML engineers planning large-scale Trainium training runs