Amazon Web Services (AWS) and Nvidia have announced plans to enhance AWS's cloud infrastructure by deploying an additional two million graphics processing units (GPUs) by 2027 and 2028. This expansion is a response to the surging demand for resources dedicated to artificial intelligence, both in agentic and physical forms.

This new commitment builds on a prior agreement made during the GTC 2026 event, where AWS revealed intentions to introduce over one million Nvidia GPUs beginning in 2026. The goal is to establish a multi-million GPU fleet to support training, inference tasks, and emerging workloads for autonomous AI agents and robotics.

Integration of Nvidia Technologies

AWS is set to incorporate multiple generations of Nvidia platforms into its AI clusters, including Blackwell Ultra, Rubin, and Rubin Ultra. The companies are also collaborating on custom chips and memory solutions, with AWS preparing infrastructure based on the NVIDIA Vera CPU, designed for AI agent tasks like tool coordination and data processing.

Additionally, AWS is enhancing its Trainium chips with Nvidia technologies. This collaboration will utilize NVLink Fusion and high-speed memory solutions to optimize performance at the rack level.

Launch of New GPU Instances

AWS recently unveiled the EC2 G7 instances, which feature NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs. These instances are touted as the first of their kind among major cloud providers and offer substantial enhancements compared to the previous G6 generation, such as:

  • Up to 4.6 times more efficient AI inference.
  • Up to 2.1 times better graphical performance.

These GPU instances will utilize AWS's Nitro network infrastructure and Elastic Fabric Adapter, along with Nvidia's Spectrum networking technologies.

Focus on Government Applications

Part of the partnership also aims to serve government and defense sectors, with plans to create specialized AI factories for the U.S. government equipped with around 100,000 Nvidia GPUs, designed to handle classified data under stringent security requirements.

Advancements in Physical AI

Moreover, Amazon Robotics will standardize Nvidia platforms for physical AI applications, including Jetson and Omniverse, to facilitate warehouse automation and robotic system simulations. Open models from NVIDIA Nemotron will remain accessible through Amazon Bedrock and SageMaker, allowing for both managed services and custom deployments.

Nvidia highlighted that this agreement with AWS illustrates a shift in the AI infrastructure landscape, moving beyond just GPU counts to building comprehensive computing factories that integrate various technologies for advanced AI and robotics applications.