AWS and NVIDIA announced an expanded strategic partnership on August 26, planning to deploy 2 million additional NVIDIA GPUs across AWS global infrastructure in 2027–2028. The announcement goes beyond chip supply: it also covers NVIDIA Vera CPUs, NVLink Fusion, Nemotron open models, data processing and robotics for large agentic and physical AI workloads.
Where will the 2 million GPUs go?
AWS says the added capacity will support model training and inference, scientific research, enterprise automation, robotics and other production workloads. The planned fleet includes Blackwell Ultra, Rubin and Rubin Ultra GPUs across AWS global infrastructure and AI factories.
The announcement builds on AWS’s expansion direction from NVIDIA GTC 2026, where it planned to add more than 1 million NVIDIA GPUs from 2026. AWS and NVIDIA say demand exceeded earlier expectations, leading to the larger 2027–2028 addition.
The partnership now spans the AI stack
Beyond GPUs, AWS will support NVIDIA Vera CPU infrastructure and integrate NVIDIA’s platform with AWS Nitro System and Elastic Fabric Adapter. These components affect security, networking and compute coordination across GPU clusters rather than a model feature users see in one API.
At the model layer, NVIDIA Nemotron open models will be available through Amazon Bedrock and SageMaker. At the data layer, the companies will use cuDF and cuVS to accelerate processing and vector indexing. Amazon Robotics will also adopt NVIDIA’s physical AI platform for warehouse automation.
Plans are not the same as capacity available today
The 2 million GPUs are a deployment plan for 2027–2028, not capacity that is already available to rent today. Availability will still depend on AWS product announcements, GPU type, region, EC2 instance and service.
Companies also cannot infer model cost or performance from GPU count alone. Results depend on model size, memory, networking, storage, batching, inference traffic and software optimization; AWS performance figures apply to specific configurations.
What this means for AI developers
The partnership shows that AI competition is expanding beyond which model is strongest to who can provide reliable training, inference, data and robotics infrastructure. Teams should evaluate model choice, GPU supply, cost control, data pipelines and deployment regions together rather than chase one hardware number.
For official details, see the AWS and NVIDIA announcement。
