Amazon Commits $25 Billion to Anthropic and 5 Gigawatts of Compute for Claude

Amazon and Anthropic announced on April 20 that they are expanding their partnership into one of the largest AI compute commitments in industry history. Amazon will invest an additional $5 billion in Anthropic immediately, with a path to $25 billion more tied to development milestones, while Anthropic pledges to spend over $100 billion on AWS technologies across the next decade. In exchange, the two companies will jointly build and deploy up to 5 gigawatts of new Trainium-based compute capacity to train and serve Claude. Together with Amazon's previous $8 billion investment, the deal brings Amazon's total financial exposure to Anthropic to roughly $33 billion.

The scale makes this the single largest expansion of Anthropic's infrastructure arrangements, surpassing even the Akamai inference deal signed in May and the earlier SpaceX Colossus agreement. It also cements AWS as Anthropic's primary cloud provider for the foreseeable future, a position Amazon has been building toward since its initial investment in 2023.

The Compute Side of the Deal

Five gigawatts is an unusual unit for a business announcement, but in AI infrastructure it is the clearest way to express what Anthropic is actually acquiring. A single gigawatt of Trainium capacity translates to hundreds of thousands of accelerators, enough to run training runs and inference workloads at a scale that Anthropic could not sustain on rented third-party capacity alone. The companies have agreed that nearly 1 gigawatt of combined Trainium2 and Trainium3 will be live by the end of 2026, with the remainder of the 5-gigawatt envelope rolling out as demand warrants.

The underlying hardware is Project Rainier, the cluster Amazon built with Anthropic and announced last year. Rainier currently houses close to half a million Trainium2 chips, making it one of the largest AI training clusters in the world. The new deal extends the Rainier model to dedicated inference capacity and adds the next-generation Trainium3 accelerators, which Amazon says will roughly double per-chip performance compared to Trainium2.

For Anthropic, the practical effect is reliability. Annualized revenues have crossed $30 billion, up from approximately $9 billion at the end of 2025, and that trajectory places severe strain on available inference capacity. Eighty-fold first-quarter growth is the kind of number that turns latency into a customer-retention issue. The AWS commitment provides the headroom to absorb consumer and enterprise demand through the rest of 2026 without the reliability degradations that have affected some competitors operating in tighter capacity environments.

"We believe Amazon is the best partner for the next phase of our growth. The combination of Trainium infrastructure, AWS global reach, and our shared investment in making Claude available at scale is unlike anything we could build through any other arrangement." Anthropic, announcing the expanded AWS partnership, April 20, 2026

What Amazon Gets

For Amazon, the deal locks in Anthropic's $100 billion cloud-spend commitment, which is revenue that flows directly to AWS. Claude is among the most-used AI models on Amazon Bedrock, the managed AI service through which most AWS enterprise customers access foundation models, and deeper integration with Trainium infrastructure allows Amazon to position Bedrock as the natural home for organizations building Claude-based applications. The investment also shields Amazon from the risk that Anthropic could shift its workloads to Google Cloud or Microsoft Azure as capacity constraints ease — the $100 billion pledge makes that scenario economically difficult.

The deal does not give Amazon equity control. Anthropic remains independent, and the governance structure leaves Anthropic's board and safety policies unchanged. That distinction matters in the current regulatory environment, where several major jurisdictions are scrutinizing whether large cloud providers are using AI investment structures to acquire de facto control over competitors without triggering standard merger-review processes. The UK's Competition and Markets Authority has been investigating Amazon's earlier Anthropic investment along those lines, and the new deal is likely to prompt a fresh look.

Infrastructure as Strategy

The Amazon announcement arrives as Anthropic navigates a period of rapid expansion on multiple fronts: new enterprise partnerships with KPMG, PwC, and Bristol-Myers Squibb, new developer tooling through Claude Code and Managed Agents, and a legal dispute with the Pentagon that has introduced operational uncertainty for some government-adjacent customers. Against that backdrop, locking in 5 gigawatts of dedicated compute removes one category of risk: the company does not need to scramble for capacity in a market where Nvidia GPUs are still constrained and third-party cloud slots are at a premium.

The 10-year $100 billion AWS commitment is the part of the deal that will draw the most scrutiny. Critics will note that it binds Anthropic to Amazon's cloud ecosystem through at least 2036, a period that will encompass multiple generations of AI hardware and likely several inflection points in the cost structure of inference. Anthropic's counterargument is that the arrangement provides price and capacity certainty that no spot-market strategy could match at current growth rates.