Rotating Data
Explore the map
Home/Deal Room/OpenAI expands AWS deal around 2 GW of Trainium capacity

Cloud capacity · Evidence brief

OpenAI expands AWS deal around 2 GW of Trainium capacity

OpenAI + Amazon + AWS

Primary sourcePress release only2026-02-27
Deal structureAI cloud-capacity expansion / custom silicon infrastructure partnership
GeographyAWS global cloud infrastructure
Infrastructure layersCloud capacity + chips / compute + data centers
Disclosed scale$100B expansion of OpenAI and AWS' existing $38B multi-year agreement over eight years, with OpenAI committing to consume approximately 2 GW of Trainium capacity through AWS infrastructure; Amazon also announced a $50B OpenAI investment tied to the broader partnership

Why it matters

The strategic read

Adds OpenAI to the AWS custom-silicon demand stack at gigawatt scale. This is not a generic model-distribution partnership: OpenAI is reserving roughly 2 GW of purpose-built Trainium capacity for Stateful Runtime Environment, Frontier and other advanced workloads, while Amazon is pairing cloud distribution, custom model work and a major equity investment with the compute commitment. It puts Trainium directly into the frontier-lab capacity race beside NVIDIA GPU, TPU and external neocloud lanes already represented in the Deal Room.

Source-supported terms

What was disclosed

  1. 01

    OpenAI and Amazon announced a multi-year strategic partnership on February 27, 2026.

  2. 02

    AWS will be the exclusive third-party cloud distribution provider for OpenAI Frontier, OpenAI's enterprise platform for building, deploying and managing AI agents.

  3. 03

    OpenAI and AWS are expanding their existing $38B multi-year agreement by $100B over eight years.

  4. 04

    The expansion includes OpenAI committing to consume approximately 2 GW of Trainium capacity through AWS infrastructure.

  5. 05

    OpenAI said the Trainium capacity will support demand for Stateful Runtime Environment, Frontier and other advanced workloads.

  6. 06

    OpenAI and Amazon will co-create a Stateful Runtime Environment powered by OpenAI models and available through Amazon Bedrock.

  7. 07

    OpenAI and Amazon will develop customized models to power Amazon customer-facing applications.

  8. 08

    Amazon said it will invest $50B in OpenAI, starting with an initial $15B investment and followed by another $35B when certain conditions are met.

  9. 09

    AWS separately said the 2 GW commitment spans Trainium3 and next-generation Trainium4 chips.

  10. 10

    DCD recirculated the partnership context on June 19, 2026 in coverage of Amazon potentially selling Trainium AI chips to data-center operators, noting that OpenAI had said in February it intended to use 2 GW of Trainium capacity on AWS.

Disclosure boundary

What is known—and what is not

Primary OpenAI announcement plus AWS follow-up. Public terms disclose the parties, $100B expansion, eight-year term, approximate 2 GW Trainium capacity commitment, OpenAI workload categories, AWS Frontier distribution role, Bedrock / Stateful Runtime integration, custom model cooperation and Amazon's planned $50B OpenAI investment. Public materials do not disclose the full cloud-services contract, annual minimums, exact Trainium generation mix by year beyond AWS follow-up language, region or data-center allocation, delivery milestones, power procurement, cooling or networking configuration, pricing, service-level terms, remedies, cancellation rights or whether Amazon's investment economics are directly linked to Trainium consumption.