OpenAI has signed a $38 billion deal with Amazon Web Services that will give it access to AWS’s vast computing infrastructure to support the development and scaling of next-generation AI models. The deal marks OpenAI’s first major cloud partnership since Microsoft and signals a clear shift to diversify its infrastructure as the company grows.
The press release states that OpenAI will “immediately” begin using the AWS computing system to train its AI models, with “full capacity to be utilized by the end of 2026 and will be able to continue expanding in 2027 and beyond.”
Last week, OpenAI completed a for-profit restructuring and announced a new agreement with Microsoft that will give the company the rights to OpenAI’s technology until it reaches advanced general intelligence (AGI). Under the agreement, OpenAI can collaborate with third parties to build some AI products and release certain open-source models.
While the company will continue to heavily partner with Microsoft (another $250 billion is reportedly being committed to Azure), the AWS deal provides crucial flexibility and reduces the risk of over-reliance on a single vendor. The dual-cloud approach enhances resilience and supports long-term scale.
A key element of the AWS partnership is access to NVIDIA’s new Blackwell GPUs, including the GB200 and GB300 chips. These high-performance processors are deployed in tightly integrated AWS EC2 clusters, providing the seamless compute power needed for OpenAI’s evolving model architecture.
AWS already operates GPU clusters with over 500,000 chips, offering the scale and technical depth OpenAI requires. While NVIDIA remains the focus, the agreement leaves room for AWS to explore proprietary silicon options, such as Trainium, which “Anthropic” currently uses.
The deal also reflects a broader shift in how infrastructure solutions shape the AI ecosystem. AWS CEO Matt Garman called the deal a confirmation of AWS’s readiness to manage AI development at scale.
OpenAI models are already integrated into Amazon Bedrock, an AWS-managed service for accessing leading AI systems. Companies like “Peloton”, “Thomson Reuters”, and “Triomics” are using the tools for everything from automation and research to software development.
Credits:
