OpenAI will now use Amazon Web Services (AWS) infrastructure to train its AI models. The multi-year contract includes a $38 billion investment.
OpenAI and AWS sign a $38 billion cloud deal. Under this agreement, OpenAI gains direct access to AWS computing capacity. This includes Amazon EC2 UltraServers equipped with hundreds of thousands of Nvidia GPUs. The infrastructure is designed to be scalable to tens of millions of CPUs, which is necessary for training generative AI models and running agentic workloads.
Access to Infrastructure
The infrastructure offered by AWS includes interconnected systems with low latency. This configuration is intended for various applications, such as running ChatGPT inference and training future AI models. The initial capabilities will be deployed before the end of 2026, with further expansion planned from 2027.
read also
OpenAI Completes Restructuring into For-Profit Company
AWS and OpenAI have previously collaborated through Amazon Bedrock. Earlier this year, OpenAI foundation models were made available to business customers in Bedrock. Companies such as Comscore, Thomson Reuters, and Peloton use the models for coding, analysis, and troubleshooting, among other things.