Overview
- AWS and Nvidia confirmed on Wednesday that AWS will deploy an additional 2 million Nvidia GPUs across 2027–2028, adding to a March commitment that brings planned Nvidia capacity on AWS to more than 3 million units.
- The order covers Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra GPU families and also includes Vera CPUs, NVLink Fusion networking work, and high‑bandwidth memory integration for rack‑scale interoperability.
- The expanded agreement extends beyond data centers by adopting Nvidia’s physical AI stack for Amazon Robotics and allocating about 100,000 GPUs to secure AWS infrastructure for U.S. federal and national‑security ‘AI factory’ workloads.
- AWS says most of the new 2027 capacity is already reserved by customers, and both companies warn that supply constraints, memory and component costs, plus long data‑center lead times will shape actual delivery and deployment timing.
- The deal highlights AWS’s dual strategy of buying large volumes of Nvidia hardware while continuing to scale its own Trainium processors, a mix designed to balance performance, cost and long‑term infrastructure planning.