Anthropic and Amazon Go All In: $100 Billion Compute Deal and a Lot More Trainium Chips

Anthropic just dropped a big one. They’ve signed a new agreement with Amazon that goes way beyond the usual cloud partnership. We’re talking up to 5 gigawatts of compute capacity for training and running Claude, and a commitment of over $100 billion to AWS technologies over the next ten years. That’s not a typo.

Let’s break down what this actually means, because the numbers are wild even by AI industry standards.

First, the compute. Anthropic is already using over a million Trainium2 chips for Claude. They launched Project Rainier together, one of the largest compute clusters on the planet. Now they’re locking in 5GW of new capacity, spanning Graviton and Trainium chips from generation 2 through 4, with the option to buy future generations. Significant Trainium2 capacity hits in Q2 of this year, and scaled Trainium3 capacity is expected by the end of 2026. Nearly 1GW total of Trainium2 and Trainium3 capacity will be online by then.

Second, the money. Amazon is investing $5 billion in Anthropic today, with the potential for up to an additional $20 billion down the road. That’s on top of the $8 billion Amazon has already put in. Total potential investment: $28 billion. This is not a small bet.

Third, the platform. The full Claude Platform will be available directly within AWS — same account, same controls, same billing. No extra credentials or contracts. That’s a big deal for enterprises that want Claude but need to keep everything under their existing governance and compliance frameworks. Claude remains the only frontier AI model available on all three major clouds: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That kind of multi-cloud flexibility is rare and smart.

The timing makes sense. Anthropic says their run-rate revenue has surpassed $30 billion, up from roughly $9 billion at the end of 2025. That’s tripling in a year. Consumer usage across free, Pro, and Max tiers has surged, and it’s putting serious strain on infrastructure. Reliability and performance have taken hits during peak hours, especially for free and Pro users. This deal is meant to fix that fast — meaningful compute in the next three months, nearly 1GW before year-end.

Andy Jassy, Amazon’s CEO, framed it around custom silicon performance and cost. Dario Amodei, Anthropic’s CEO, talked about keeping Claude at the frontier. Both are right, but the subtext here is clear: demand is exploding, and the infrastructure race is real. Anthropic is betting big on AWS’s custom chips over NVIDIA’s GPUs. That’s a strategic bet that could pay off if Trainium delivers on its promise of lower cost and high performance. If not, it’s a massive lock-in.

I’ve been watching this space for a while, and I have to say — the scale of this commitment is staggering. $100 billion over ten years is not just a partnership, it’s a near-merger of AI and cloud infrastructure. The fact that Anthropic is also expanding inference capacity in Asia and Europe tells me they’re thinking globally, not just about US demand.

One thing I find interesting is the diversification strategy. Anthropic says they’re spreading workloads across a range of chips, not just Trainium. That’s smart. No single hardware provider should own your entire stack. But the bulk of this deal is clearly tied to Amazon’s silicon roadmap. If Trainium4 doesn’t deliver, there’s a lot of money at stake.

For now, the immediate impact is that Claude should get more reliable and faster, especially during peak hours. That’s good news for the growing number of users — over 100,000 customers on AWS Bedrock alone. The Claude Platform on AWS is coming soon, so if you’re an enterprise customer, reach out to your AWS account team to get early access.

This is the kind of deal that reshapes the competitive landscape. Google and Microsoft are not going to sit still. Expect more announcements from the other hyperscalers soon. The AI infrastructure arms race is just getting started.