On April 20, 2026, Anthropic and Amazon announced an agreement that secures up to 5 gigawatts of new capacity for training and running Claude, backed by an Anthropic commitment of more than $100 billion over ten years to AWS technologies. Amazon agreed to invest $5 billion in Anthropic immediately and up to $20 billion more later. Anthropic also acknowledged that its growth had strained its infrastructure and hurt reliability for consumer users at peak hours, which makes this deal as much about service quality as about training frontier models.

Key Facts#

  • Announced: April 20, 2026, by Anthropic. The post was updated on April 21 to clarify that Claude Platform on AWS is coming soon.
  • Capacity: up to 5 GW of new capacity to train and deploy Claude.
  • Commitment: more than $100 billion over the next ten years, spanning Graviton CPUs and Trainium2 through Trainium4 chips, with an option to buy future generations of Amazon's custom silicon.
  • Near-term timeline: significant Trainium2 capacity in the second quarter of 2026, and nearly 1 GW of Trainium2 and Trainium3 capacity in total by the end of 2026.
  • Current scale: Anthropic says it already uses more than one million Trainium2 chips to train and serve Claude, and that over 100,000 customers run Claude on Amazon Bedrock.
  • Investment: $5 billion from Amazon now and up to $20 billion in the future, on top of the $8 billion Amazon had previously invested.
  • New product: Claude Platform on AWS, the full Claude Platform available inside AWS accounts, announced as coming soon.
  • Demand: Anthropic said its run-rate revenue had passed $30 billion, up from about $9 billion at the end of 2025.

What Happened#

Anthropic described the agreement as expanding its Amazon relationship in three ways. The first is infrastructure. The 5 GW figure is a ceiling that spans several chip generations, from Trainium2, which Anthropic already uses at scale, through Trainium3 later in 2026 to Trainium4, with an option on later generations. Anthropic said it will also use incremental capacity for Claude in Amazon Bedrock, and that the agreement expands inference in Asia and Europe for international customers.

The second is distribution. Claude Platform on AWS will offer the complete Claude Platform directly within AWS, with the same account, controls and billing that customers already use, and without separate credentials or contracts. Anthropic said this gives organizations direct access to Claude, with more Claude Platform features, while meeting their existing governance and compliance requirements.

The third is capital. Amazon's new investment adds to the $8 billion it had already put into Anthropic. Amazon CEO Andy Jassy said the commitment to run Anthropic's models on Trainium "for the next decade" reflects progress the two companies have made together on custom silicon. Anthropic CEO Dario Amodei said the company needs infrastructure that keeps pace with rapidly growing demand.

The post's most practical passage concerned reliability. Anthropic said its consumer growth in 2026 had "impacted reliability and performance" for Free, Pro, Max and Team users, especially during peak hours, and that the agreement would deliver meaningful compute within three months.

Background#

Anthropic has worked with Amazon since 2023 and calls AWS its primary training and cloud provider. Together they built Project Rainier, which Anthropic describes as one of the largest compute clusters in the world. Anthropic has also deliberately avoided depending on one chip supplier. It trains and runs Claude on AWS Trainium, Google TPUs and NVIDIA GPUs, and says this lets it match each workload to the hardware best suited for it.

The Amazon deal was the latest in a run of large commitments. In October 2025, Anthropic said it would use up to one million Google TPUs. In November 2025, it agreed to buy $30 billion of Azure capacity as part of a partnership with Microsoft and NVIDIA, and it announced a $50 billion investment in U.S. data centers with Fluidstack. Two weeks before the Amazon announcement, it signed with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027. The common thread is demand: Anthropic said that by early April more than 1,000 business customers were each spending over $1 million a year, double the number it reported in February.

Why It Matters for Developers#

Capacity deals rarely change code, but this one changes the operational picture for teams that depend on Claude.

  • Plan for capacity limits. Anthropic acknowledged that demand outran supply. New capacity arrives in stages, so build retries with backoff, timeouts and fallback models into AI features. Our resilience guide shows how to do this with Polly and Microsoft.Extensions.Http.Resilience.
  • Pick the route that fits your governance. For .NET teams on AWS, Claude is already available through Bedrock via the AWSSDK.BedrockRuntime package, and the AWSSDK.Extensions.Bedrock.MEAI package implements the Microsoft.Extensions.AI abstractions for Bedrock. Claude Platform on AWS adds a second route inside the same account. Teams on Azure can use Microsoft Foundry, and the official Anthropic NuGet package calls the Claude API directly.
  • Keep the provider swappable. Because Claude runs on several clouds, coding against IChatClient, as described in our Microsoft.Extensions.AI guide, lets you change routes when prices, limits or compliance needs change.
  • Check where inference runs. The planned expansion of inference in Asia and Europe may help teams with data residency requirements, but confirm which regions your chosen route actually offers before you promise customers anything.

Architects should also track usage and cost per route. When a provider is adding capacity quickly, rate limits and performance can shift from month to month, and good telemetry is the only reliable way to notice. Our LLM observability and cost guide covers the setup.

What's Next#

Anthropic said Trainium3 capacity would scale up later in 2026 and that Claude Platform on AWS would follow, with access requested through AWS account teams. On May 6, it announced a deal to use all of the compute at SpaceX's Colossus 1 data center and raised usage limits, listing the Amazon agreement's nearly 1 GW of new capacity by the end of 2026 among its sources of supply. On May 28, Anthropic's Series H announcement said its run-rate revenue had crossed $47 billion and that the round included $15 billion of previously committed hyperscaler investment, including $5 billion from Amazon.

Open questions include where the rest of the 5 GW will be built, how quickly Trainium4 arrives, and whether the extra capacity ends peak-hour degradation for consumer users.

Sources#