NVIDIA and Amazon Web Services have announced a substantial expansion of their sixteen-year strategic collaboration, significantly scaling AI infrastructure capacity and deepening integration across the full technology stack to advance agentic and physical AI.

The partnership now spans GPU and CPU hardware, advanced networking, open models, and robotics platforms as both companies respond to enterprise and government demand that has exceeded existing projections. The announcement reflects a broader industry shift from pilot projects to production-scale deployment across scientific discovery, enterprise automation, and robotics, where compute availability, networking, and software integration are increasingly treated as a single system.

The centerpiece of the expansion is the deployment of two million additional NVIDIA GPUs across AWS global infrastructure during 2027 and 2028. This capacity will include NVIDIA Blackwell Ultra, Rubin, and Rubin Ultra accelerators, building on previously announced plans to add over one million GPUs starting in 2026. AWS will also introduce instances powered by the NVIDIA Vera CPU, purpose-built for agentic AI workloads, and will implement NVIDIA NVLink Fusion alongside custom high-bandwidth memory technology developed with Annapurna Labs to enhance speed and power efficiency across both GPU and Trainium architectures.

For large-scale AI training, the companies are collaborating on NVIDIA Spectrum networking to optimize performance across GPU clusters. On the government front, AWS and NVIDIA plan to construct dedicated AI factories delivering one hundred thousand GPUs on secure infrastructure certified for federal workloads at Impact Level 6 and above. All new instances will continue to utilize the AWS Nitro System and Elastic Fabric Adapter, maintaining security and reliability standards as the fleet scales.

Software, Data Processing, and Physical AI

Beyond hardware expansion, the collaboration emphasizes software-layer optimization and emerging use cases. NVIDIA Nemotron open models will remain available as fully managed offerings on Amazon Bedrock and as deployable resources on Amazon SageMaker, broadening model choice for developers. Data processing capabilities will be enhanced through GPU acceleration on Amazon EMR using NVIDIA cuDF, promising up to 3.7 times faster processing with 30 percent better price performance compared to CPU configurations.

Separately, Amazon OpenSearch Service will utilize GPU-accelerated vector indexing to deliver up to nine times faster index construction at roughly one-quarter of the cost, addressing bottlenecks in retrieval-augmented generation and semantic search pipelines that depend on fast retrieval over large datasets. AWS will also become the first major cloud provider to offer EC2 G7 instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, which deliver 4.6 times the AI inference performance and 2.1 times the graphics performance of prior-generation G6 instances.

In the robotics sector, Amazon Robotics will integrate NVIDIA full-stack physical AI technologies, including the Jetson platform, Omniverse libraries, and Isaac development framework, to advance warehouse automation through simulation, synthetic data generation, route optimization, and real-to-sim validation.