The previous commitment of one million chips was consumed faster than expected. Amazon tripled the bet and extended the agreement to processors, networking and robotics.
Announced on August 26, the new agreement between Amazon Web Services and Nvidia focuses on the deployment of an additional two million GPUs — Blackwell Ultra, Rubin and Rubin Ultra — on the cloud provider's global infrastructure in the 2027s and 2028s. The previous commitment, presented to GTC 2026 and covering more than 1 million chips from 2026 onwards, was absorbed faster than both companies anticipated.
The partnership goes well beyond the only supply of accelerators. Nvidia's Vera processors will join the AWS catalog for Gourmet Agential Loads in General Computing: Code Execution, Tool Call, Safety Isolation, Analytical and Pipeline Orchestration. The future Trainium home chips will host the NVLink Fusion interconnection as well as the NVHBM memory, a sign that both catalogues cease to be strictly competing and become partially interoperable.
Two components complete the package. Both companies plan to build "IA plants" for the U.S. government, with 100,000 GPUs installed on a certified AWS infrastructure for IL6 and beyond classified loads. On the warehouse side, Amazon Robotics adopts Nvidia's physical AI battery — Jetson, Omniverse and Isaac — for its next generation of automation.
Read the full article on the original source
Go to the source →