Amazon Web Services will deploy an additional 2 million NVIDIA GPUs across its global infrastructure in 2027 and 2028, following an earlier commitment to install 1 million or more chips beginning in 2026. The combined program brings AWS’s publicly disclosed NVIDIA GPU commitment to more than 3 million units across roughly three years, the largest such commitment publicly disclosed by a single cloud provider. The new order covers Blackwell Ultra, Rubin and Rubin Ultra systems. AWS and NVIDIA are also broadening their partnership through NVLink Fusion and NVIDIA’s NVHBM memory architecture, allowing AWS Trainium chips and NVIDIA GPUs to operate within a more integrated rack-scale system. The companies plan AI factories for U.S. government workloads using 100,000 GPUs, while AWS will add NVIDIA’s Nemotron models to Bedrock and SageMaker and Amazon Robotics will use NVIDIA’s physical AI platform for warehouse automation. Amazon shares opened at $256.32, down 1.52%, as investors weighed the larger NVIDIA commitment against AWS’s continued development of internal Trainium processors.