AWS and NVIDIA Plan 2 Million More GPUs for Agentic and Physical AI

A much larger AWS-NVIDIA AI infrastructure expansion

AWS and NVIDIA announced on August 26, 2026 that they plan to deploy 2 million additional NVIDIA GPUs across AWS global infrastructure during 2027-2028. The companies describe the expansion as a response to rising demand from frontier labs, enterprises, startups and governments for agentic AI, scientific computing, enterprise automation and physical AI.

This is a forward-looking infrastructure commitment, not 2 million GPUs that are already online today. NVIDIA says the planned fleet will include Blackwell Ultra, Rubin and Rubin Ultra GPUs and will be spread across AWS global infrastructure, including AI-factory deployments.

The partnership goes beyond GPU capacity

The announcement spans CPUs, memory, networking, open models, data processing and robotics. AWS and NVIDIA say they are working to bring NVIDIA Vera CPU-based infrastructure to AWS, extend NVLink Fusion with NVIDIA NVHBM for future Trainium designs, and continue integrating NVIDIA technology with AWS Nitro and Elastic Fabric Adapter.

Those components have their own product timelines. Vera infrastructure and the additional Rubin-class GPU capacity should therefore be treated as planned capabilities rather than currently available AWS instance families.

100,000 GPUs are planned for U.S. government AI factories

The companies also say they plan AI factories for U.S. federal and national-security workloads, including 100,000 GPUs on secure AWS infrastructure. That figure is part of the announced roadmap and should not be interpreted as an existing deployed government cluster.

Nemotron, OpenSearch and accelerated data processing are part of the stack

NVIDIA Nemotron open models remain available through Amazon Bedrock and SageMaker. AWS and NVIDIA are also collaborating on GPU-accelerated data processing with cuDF and vector indexing with cuVS. NVIDIA says the planned OpenSearch acceleration can deliver up to 9x faster vector-index construction at a quarter of the cost in its cited configuration, while EMR with G7 and cuDF can deliver up to 3.7x faster processing and 30% better price-performance than the compared CPU setup. These are vendor-reported workload results, not universal guarantees.

Physical AI is becoming a first-class cloud workload

Amazon Robotics is also collaborating with NVIDIA around Jetson, Omniverse and Isaac for simulation, synthetic data, robot training, route optimization, functional safety and real-to-sim validation. That makes the announcement notable not only for raw GPU volume but for how cloud infrastructure is being tied directly to robotics development pipelines.

Why the 2 million GPU plan matters

The scale illustrates a shift from isolated accelerator deployments toward integrated AI factories combining compute, CPU orchestration, memory, networking, model catalogs, vector infrastructure and robotics software. It also gives developers a clearer view of where AWS expects demand to grow: long-running agents, large-scale inference, scientific workloads and physical AI.

The most important status distinction is timing: the 2 million additional GPUs are planned for 2027-2028, while several supporting services and models are already available today.