Overview
- OpenAI has begun running workloads on AWS now, with all contracted capacity targeted to be in place by the end of 2026 and an option to expand into 2027.
- AWS will cluster hundreds of thousands of Nvidia GB200 and GB300 GPUs via Amazon EC2 UltraServers, with the ability to scale to tens of millions of CPUs for both ChatGPT inference and next‑generation training.
- This is OpenAI’s first direct contract with AWS, signaling reduced reliance on Microsoft even as OpenAI continues large commitments with Azure and other suppliers.
- AWS executives said some capacity is already available and in use as part of a straightforward customer purchase agreement, and Amazon shares rose about 5% after the news.
- The pact joins OpenAI’s broader spree of massive compute commitments reported at roughly $1.4 trillion across vendors, drawing scrutiny over financing, power availability, and execution timelines.