Amazon Web Services and NVIDIA have expanded their cloud partnership to install 2 million extra graphics processors across global data centers. The computing expansion targets enterprise demand for agentic intelligence and physical robotics while introducing custom memory hardware to Amazon silicon. A major portion of the new hardware will support dedicated infrastructure built for government defense operations.
Enterprise demand for accelerated hardware continues to outpace earlier supply estimates. AWS plans to install 2 million Blackwell Ultra, Rubin, and Rubin Ultra processors across its global network through 2027 and 2028. This comes on top of the 1 million units already planned. The upgrade also introduces RTX PRO 4500 Blackwell Server Edition hardware to Amazon EC2 G7 instances, speeding up AI inference tasks by 4.6x and graphic rendering by 2.1x compared to older G6 hardware.
Demand is running ahead of every forecast. For 16 years, we have scaled NVIDIA computing in the cloud together. Now, we are expanding our partnership across the full stack to make agentic and physical AI real at an unprecedented pace and scale.
The updated agreement brings NVIDIA Vera processors into the AWS ecosystem. These central processors are built to manage heavy agentic tasks that require fast single thread calculations alongside massive graphics clusters. Amazon subsidiary Annapurna Labs is also upgrading its Trainium silicon to support NVLink Fusion with custom high bandwidth memory. This allows engineers to combine proprietary Amazon silicon and NVIDIA chips inside the same unified server racks.
System protection for these server clusters runs on the AWS Nitro System linked with Elastic Fabric Adapter networking. This foundation provides direct hardware isolation to keep corporate datasets private across distributed cloud environments.
Customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together. This expanded collaboration gives frontier labs, enterprises and governments even more ways to build and deploy AI on AWS.
A critical piece of the project focuses on dedicated computing centers for the United States government. AWS will set up 100,000 GPUs on isolated infrastructure built for federal and national security workloads rated at Impact Level 6 and higher. This gives intelligence agencies access to massive computing clusters without exposing private data to the public internet.
Software performance is also seeing direct speed improvements. Amazon EMR now processes data pipelines with the CUDA cuDF library, speeding up tasks by 3.7x with a 30% price reduction over traditional processors. Amazon OpenSearch offloads vector indexing to dedicated GPUs, completing indexing tasks 9x faster at 25% of the usual cost. For industrial robotics, Amazon Robotics is using the NVIDIA Omniverse and Isaac platforms to train next generation warehouse machines in digital simulation before deploying them onto physical factory floors.
