Amazon Puts $50 Billion Into OpenAI in Sweeping Cloud Deal

Amazon will invest $50 billion in OpenAI, the two companies announced today, in a multi-year strategic partnership that also makes Amazon Web Services the exclusive third-party cloud distribution provider for OpenAI Frontier, the company's most advanced enterprise platform.

The investment starts with an initial $15 billion, followed by another $35 billion in the coming months once certain conditions are met. The size of the commitment places Amazon among OpenAI's largest backers and signals how sharply the competitive lines between cloud providers and model developers have shifted: Amazon, which builds its own Nova family of foundation models and its own Trainium AI chips, is now funding and distributing its most formidable rival's technology.

The deal matters beyond the two companies. It redraws the map of enterprise AI distribution, ties OpenAI's production workloads more deeply to non-Nvidia silicon, and gives AWS customers direct access to OpenAI's frontier models inside Amazon Bedrock — the managed service where enterprises already assemble production AI applications.

A new runtime for agents, built on Bedrock

At the center of the technical partnership is a jointly developed Stateful Runtime Environment, powered by OpenAI's models and available through Amazon Bedrock. The companies describe stateful developer environments as the next generation of how frontier models will be used, enabling models to access compute, memory, and identity. The runtime lets developers keep context, remember prior work, operate across software tools and data sources, and draw on compute as needed — capabilities designed for ongoing projects and workflows rather than one-off prompts.

The Stateful Runtime Environment will be trained to run optimally on AWS infrastructure and integrated with Amazon Bedrock AgentCore and AWS infrastructure services, so that customers' AI applications and agents run cohesively with the rest of their workloads on AWS. The companies expect the runtime to launch in the next few months.

The move targets the fastest-growing segment of enterprise AI spending: agents that persist across sessions and act inside real business systems. OpenAI and Amazon are betting that the runtime layer — where context, memory, and identity live — becomes the contested ground for the next phase of AI deployment.

Frontier comes to AWS — and only to AWS among third-party clouds

Under the agreement, AWS becomes the exclusive third-party cloud distribution provider for OpenAI Frontier. Frontier lets organizations build, deploy, and manage teams of AI agents that operate across real business systems with shared context, built-in governance, and enterprise-grade security, without managing underlying infrastructure.

For enterprises moving from experimentation to production AI, the companies frame Frontier as a way to integrate AI into existing workflows quickly, securely, and at global scale. Exclusivity means AWS is now the only third-party cloud through which organizations can buy Frontier — a notable win for Amazon as demand for AI deployment accelerates across industries, and a constraint on where OpenAI's enterprise platform can otherwise be consumed.

2 gigawatts of Trainium, and a $100 billion expansion

The compute commitments are among the largest in the industry. OpenAI and AWS are expanding their existing $38 billion multi-year agreement by $100 billion over eight years. As part of that expansion, OpenAI has committed to consume approximately 2 gigawatts of Trainium capacity through AWS infrastructure to support demand for the Stateful Runtime Environment, Frontier, and other advanced workloads.

The companies say the agreement lowers the cost and improves the efficiency of producing intelligence at scale. Under the structure, OpenAI secures long-term capacity while working with AWS to deploy purpose-built silicon alongside its broader compute ecosystem, letting enterprises consume intelligence on demand without managing infrastructure themselves.

The commitment spans Trainium3 and next-generation Trainium4 chips. Trainium4, expected to begin delivery in 2027, is designed to deliver significantly higher FP4 compute performance, expanded memory bandwidth, and increased high-bandwidth memory capacity — specifications aimed at supporting increasingly capable AI systems at scale. That timeline gives AWS and OpenAI a multi-year runway to shift a substantial share of OpenAI's training and inference workloads onto Amazon's custom silicon.

Custom OpenAI models for Amazon's own products

The partnership also reaches into Amazon's consumer-facing products. OpenAI and Amazon will collaborate to develop customized models available to Amazon developers to power Amazon's customer-facing applications. Amazon teams will be able to tailor OpenAI models for AI products and agents that serve customers directly.

These capabilities will complement the models Amazon developers already have access to, including Amazon's own Nova family — effectively giving Amazon's teams another tool to build and deliver at scale alongside in-house models rather than replacing them.

What the numbers say about where this is heading

The aggregate stakes are substantial: a $50 billion equity investment, a $100 billion expansion of a $38 billion compute agreement, 2 gigawatts of committed Trainium capacity, and exclusive third-party cloud rights to OpenAI's enterprise platform. Together they bind OpenAI's roadmap to AWS infrastructure for the better part of a decade — spanning Trainium4 chips that will not arrive until 2027 — while giving Amazon both a financial stake in OpenAI's success and a distribution moat around its most advanced enterprise offering.

The first concrete test comes in the next few months, when the Stateful Runtime Environment launches on Amazon Bedrock and customers can begin judging whether jointly built, infrastructure-optimized agent runtimes deliver on the promise of production-scale AI.