OpenAI will run and scale its core AI workloads on Amazon Web Services under a $38 billion agreement that begins immediately. The deal, set to expand over seven years, gives OpenAI access to large-scale compute capacity as it trains and deploys new models.
Notes
1. OpenAI will use AWS data centres to run advanced model training and inference.
2. AWS will supply OpenAI with Amazon EC2 UltraServers and large clusters of NVIDIA GPUs.
3. The agreement allows OpenAI to scale up to tens of millions of CPUs as demand grows.
The rapid growth in generative AI has driven demand for large clusters of compute. OpenAI will expand its compute footprint on AWS using hundreds of thousands of NVIDIA GPUs, with the option to scale further through 2027.
AWS already operates some of the largest GPU clusters used in commercial AI systems. The setup for OpenAI groups GB200 and GB300 GPUs on shared networking to keep latency low across large-scale workloads.
The capacity will support both training and serving. That means running ChatGPT at scale today and preparing new, larger models in the future.
Sam Altman said the deal gives OpenAI the compute needed to keep scaling its most intensive work. AWS CEO Matt Garman said the agreement reflects growing demand from companies building and deploying AI systems at scale.
OpenAI models are already available through Amazon Bedrock. Customers including Bystreet, Comscore, Peloton, Thomson Reuters, Triomics, and Verana Health are using them for tasks such as coding support, data analysis, and research workflows.
