Amazon Web Services (AWS) and Nvidia have announced an expansion of their strategic partnership, under which two million additional graphics processing units will be deployed within AWS cloud infrastructure over the course of 2027 and 2028. This agreement builds on previous commitments announced at the GTC 2026 conference, when AWS revealed plans to install more than one million Nvidia GPUs starting in 2026. Taken together, the deals point to the formation of a multi-million-unit accelerator fleet that AWS intends to use for model training and inference, as well as for new workload types associated with autonomous AI agents and robotics.
As part of the expanded collaboration, AWS will integrate several generations of Nvidia platforms into its AI clusters, including Blackwell Ultra, Rubin, and Rubin Ultra. The companies are also deepening cooperation in the area of custom chips and memory. AWS is preparing infrastructure based on the NVIDIA Vera CPU, developed specifically for tasks arising around AI agents — tool coordination, code execution, data processing, and simulations. This solution is designed to complement GPUs and enable more efficient use of accelerators in multi-step agentic scenarios. Another area of focus is the integration of Nvidia technologies with AWS Trainium chips — Amazon's Annapurna Labs division and Nvidia are expanding NVLink Fusion support, including the use of high-bandwidth memory and scalable interconnects in rack-level configurations.
AWS has also introduced EC2 G7 — cloud virtual servers powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, which the company describes as the first such offering among major cloud providers. According to Nvidia, G7 instances deliver up to 4.6x improvement in AI inference efficiency and up to 2.1x in graphics performance compared to the previous-generation G6. GPU instances and Trainium-based systems will continue to leverage AWS Nitro and Elastic Fabric Adapter networking infrastructure alongside NVIDIA Spectrum networking technologies.
A dedicated area of the partnership focuses on serving government and defense customers. AWS and Nvidia plan to create specialized AI factories for the US government, which will receive approximately 100,000 Nvidia GPUs. This infrastructure will be designed to handle classified data and meet security requirements at Impact Level 6 and above. The partnership is also extending into the physical world — Amazon Robotics is standardizing on the Nvidia platform for physical AI, including Jetson, Omniverse, and Isaac, for warehouse automation, synthetic data generation, and simulation of robotic system behavior. NVIDIA Nemotron open models will remain available through Amazon Bedrock and SageMaker, both as managed services and for custom deployment.
Nvidia noted that the agreement with AWS confirms a trend in the competitive AI infrastructure market, where the focus has shifted beyond simply counting GPUs to building entire compute factories that integrate accelerators, CPUs, memory, networking, software, and specialized systems. In the context of this development, SpaceX AI announced in August an expansion of its use of Nvidia infrastructure for the development of agentic systems and Grok, with the adoption of Vera processors and the deployment of the Vera Rubin platform in space.