OpenAI spreads the imaginary wealth beyond Microsoft with $38B AWS deal
Summary
OpenAI has signed a seven-year, $38 billion agreement with Amazon Web Services to run large-scale AI compute on AWS EC2 UltraServers, with access starting immediately and planned scaling over the next seven years. The deal includes customised infrastructure using Nvidia Blackwell GB200 and GB300 GPUs clustered with EC2 UltraServers for high-efficiency AI workloads. The agreement follows OpenAI’s recent transition to a for-profit public benefit corporation that removed Microsoft’s right of first refusal, allowing OpenAI to diversify its hyperscaler partners.
The arrangement is distinct from AWS’s separate Project Rainier and does not commingle compute for that project with OpenAI’s workloads. Despite the new AWS deal, OpenAI still maintains a strong commercial relationship with Microsoft: Redmond holds 27% of the new corporate structure and OpenAI remains committed to spending up to $250 billion on Azure. Financial transparency remains limited — OpenAI’s funding sources for such commitments weren’t disclosed, and the company reported very large losses in the prior quarter.
Key Points
- OpenAI and AWS inked a seven-year, $38 billion cloud-compute agreement to scale AI workloads on EC2 UltraServers.
- AWS will deploy custom infrastructure using Nvidia Blackwell GB200 and GB300 GPUs clustered with EC2 UltraServers for maximum AI efficiency.
- Access to AWS compute starts immediately and will scale to very large capacity (described as “tens of millions of CPUs”).
- The deal follows OpenAI’s for-profit restructuring, which removed Microsoft’s prior right of first refusal and allowed multi-cloud sourcing.
- OpenAI still has a substantial Azure commitment and Microsoft stake (27% and a $250bn Azure spend commitment), so the AWS pact diversifies rather than replaces Microsoft ties.
Context and relevance
This is a material industry move: OpenAI diversifying compute partners shifts the hyperscaler dynamics around AI infrastructure. It strengthens AWS’s position as a go-to provider for large agentic AI workloads and highlights the continuing importance of Nvidia GPU supply and custom datacentre design. For organisations tracking cloud-provider strategies, vendor lock-in, or AI-capex trends, the deal signals increasing multicloud competition for AI platform business.
Why should I read this?
Because this is where the money — and therefore the future of large-scale AI services — flows. OpenAI just opened its options beyond Microsoft and handed AWS a giant seat at the table. Quick read, big implications.