As infrastructure demand surges beyond expectations, NVIDIA and AWS will expand their collaboration to government security-level AI factories.

NVIDIA and Amazon Web Services have announced a significant expansion of their sixteen-year strategic partnership, substantially increasing artificial intelligence infrastructure capacity and deepening integration across the entire technology stack to advance agents and physical AI.

The collaboration now covers GPU and CPU hardware, advanced networking, open models, and robotics platforms, aiming to meet the higher-than-expected demands from enterprises and governments. This partnership also highlights an industry-wide shift from pilot projects to production-scale deployments, spanning scientific discovery, enterprise automation, and robotics.

The core of this expansion involves deploying an additional two million NVIDIA GPUs in AWS's global infrastructure during 2027 and 2028. These GPUs will include NVIDIA Blackwell Ultra, Rubin, and Rubin Ultra accelerators. Previously, AWS announced plans to add more than one million GPUs starting in 2026. AWS will also launch instances powered by NVIDIA Vera CPUs, which are purpose-built for agent AI workloads, and deploy NVIDIA NVLink Fusion along with custom high-bandwidth memory technology developed in partnership with Annapurna Labs to increase the speed and energy efficiency of the GPU and Trainium architecture.

To enable large-scale AI training, the two companies are jointly developing NVIDIA Spectrum networking to optimize GPU cluster performance. In the government sector, AWS and NVIDIA plan to build dedicated AI factories, deploying 100 thousand GPUs on secure-certified infrastructure capable of handling federal workloads (impact level 6 and above). All new instances will continue to use AWS Nitro systems and elastic network adapters, ensuring that as cluster sizes grow, security and reliability standards are maintained.

Software, Data Processing, and Physical AI

Beyond hardware expansions, the partnership also focuses on software layer optimization and emerging application scenarios. NVIDIA Nemotron open models will continue to be offered as fully managed products on Amazon Bedrock and as deployable resources on Amazon SageMaker, providing developers with a broader range of model choices. With NVIDIA cuDF GPU acceleration on Amazon EMR, data processing capabilities will be significantly enhanced, delivering up to 3.7 times greater speed compared to CPU configurations and a 30% improvement in cost-effectiveness.

Additionally, Amazon OpenSearch Service will leverage GPU-accelerated vector indexing technology to achieve up to nine times faster index build speeds at roughly a quarter of the cost, addressing bottlenecks in retrieval-augmented generation and semantic search pipelines. AWS will also become the first major cloud provider to offer EC2 G7 instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, delivering 4.6 times the AI inference performance and 2.1 times the graphics performance of the previous generation G6 instances.

In robotics, Amazon Robotics will integrate NVIDIA's 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-simulation validation.