Amazon and Anthropic expand AWS deal for 5GW Claude compute

Anthropic has signed a new 10-year AWS agreement to secure up to 5GW of capacity for training and running Claude, including major Trainium2/3 ramps. Amazon is also investing $5B now, with up to $20B more, as demand and revenue surge.

Amazon and Anthropic expand AWS deal for 5GW Claude compute

TL;DR

  • **Amazon–Anthropic compute deal (Apr. 20):** Up to **5 GW** for training/deploying Claude; deeper partnership since 2023
  • **Trainium roadmap:** New **Trainium2** capacity in H1; nearly **1 GW Trainium2/3** by end of 2026
  • **Long-term AWS commitment:** **>$100B over 10 years**; capacity spanning **Graviton, Trainium2–Trainium4**, plus future silicon options
  • **Deployment expansion:** Incremental capacity for Claude on **Amazon Bedrock**, including inference growth in **Asia and Europe**
  • **Platform integration:** Full **Claude Platform in AWS** with same account/controls/billing; no extra credentials; “coming soon”
  • **Funding and demand pressures:** Amazon invests **$5B now**, up to **$20B later**; reliability affected; meaningful compute in three months, nearly **1 GW** before year-end

Anthropic has signed a new agreement with Amazon that it says will deepen the companies’ partnership and secure up to 5 GW of capacity for training and deploying Claude, including new Trainium2 capacity coming online in the first half of this year and nearly 1GW total of Trainium2 and Trainium3 capacity by the end of 2026.

The announcement, dated Apr. 20, extends a collaboration that Anthropic says has been in place since 2023. The company states that more than 100,000 customers now run Claude on Amazon Bedrock, and that it currently uses over one million Trainium2 chips to train and serve the model. Anthropic also links the new deal to Project Rainier, which it describes as one of the largest compute clusters in the world.

Anthropic says the agreement expands the relationship in three main ways. First, it commits the company to "more than $100 billion" over the next ten years to AWS technologies, with access to up to 5GW of new capacity spanning Graviton and Trainium2 through Trainium4 chips, plus the option to buy future generations of Amazon’s custom silicon as they become available.

The company also states that significant Trainium2 capacity is coming online in Q2, with scaled Trainium3 capacity expected later this year. Anthropic says it will use incremental capacity for Claude in Amazon Bedrock, including expanded inference in Asia and Europe. The firm adds that AWS remains its primary training and cloud provider for mission-critical workloads.

Amazon CEO Andy Jassy, in a quote included in the announcement, characterizes the company’s custom AI silicon as offering "high performance at significantly lower cost" and suggests that Anthropic’s commitment to AWS Trainium for the next decade reflects progress on custom silicon.

The second piece of the deal is closer integration with AWS. Anthropic says the full Claude Platform will be available directly within AWS with "same account, same controls, same billing," and no additional credentials or contracts. The company also states that Claude remains the only frontier AI model available across AWS, Google Cloud, and Microsoft Azure. It adds that Claude Platform on AWS is "coming soon."

The third element is new funding. Amazon is investing $5 billion in Anthropic today, with up to another $20 billion possible later, according to the announcement. Anthropic notes that this follows Amazon’s prior $8 billion investment.

Dario Amodei, Anthropic’s CEO and co-founder, says the company needs to "build the infrastructure to keep pace with rapidly growing demand." Anthropic adds that enterprise and developer demand has accelerated in 2026, alongside a sharp rise in consumer usage across its free, Pro, and Max tiers.

The company says its run-rate revenue has risen to more than $30 billion from about $9 billion at the end of 2025. It also acknowledges that the surge has strained infrastructure, with reliability and performance affected for free, Pro, Max, and Team users, especially during peak hours.

Anthropic says the new capacity should arrive quickly, with meaningful compute in the next three months and nearly 1GW in total before the end of the year. It points to additional capacity expansions and a diversified hardware strategy as part of its effort to keep Claude "at the frontier" and serve its customer base more reliably.

Source: Anthropic

Continue the conversation on Slack

Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.

Join our community