Amazon Web Services and NVIDIA have expanded their strategic collaboration, with AWS planning to add 2 million NVIDIA GPUs across its global infrastructure during 2027 and 2028. The companies said the deployment is intended to address rising demand for AI computing capacity.

The planned rollout includes NVIDIA Blackwell Ultra, Rubin, and Rubin Ultra GPUs. It follows AWS’s announcement at NVIDIA GTC 2026 that it would begin installing more than 1 million NVIDIA GPUs during the year.

AWS and NVIDIA Expand AI Computing Infrastructure

The new GPU deployment adds to AWS’s existing work with NVIDIA hardware. According to the announcement, the companies are building out infrastructure for AI workloads across AWS’s global footprint.

NVIDIA founder and CEO Jensen Huang said that NVIDIA and AWS have built “one of the great growth engines of the AI era,” adding that demand is exceeding forecasts. AWS CEO Matt Garman said the broader collaboration provides frontier labs, enterprises, and governments with more options to build and deploy AI on AWS.

AWS’s planned GPU expansion includes several NVIDIA GPU platforms:

  • NVIDIA Blackwell Ultra
  • NVIDIA Rubin
  • NVIDIA Rubin Ultra

The announcement places the additional 2 million GPUs alongside the more than 1 million NVIDIA GPUs AWS said it would start installing in 2026.

NVIDIA GPUs and AWS Custom AI Chips

AWS described the expanded NVIDIA relationship as complementary to its in-house silicon strategy. Through Annapurna Labs, Amazon develops the Trainium family of custom AI chips.

AWS said it aims to offer customers a broad choice of compute options, including Trainium chips and NVIDIA accelerators, so they can select technology suited to their workloads. The additional NVIDIA GPU deployment indicates that AWS is continuing to use NVIDIA hardware while also developing its own AI chip offerings.

The companies are also expanding their work around NVLink Fusion interconnect technology. NVIDIA and Amazon’s Annapurna Labs plan to work with NVIDIA’s custom high-bandwidth memory so AWS Trainium chips and NVIDIA GPUs can operate in a shared rack-scale architecture.

Vera CPU Infrastructure for AI Workloads

The collaboration extends beyond GPU capacity. AWS and NVIDIA are working to bring NVIDIA’s Vera CPU-based infrastructure to AWS.

The Vera platform is intended to support agentic AI workloads that require high-performance CPU computing alongside GPU acceleration. This work adds a CPU component to the companies’ GPU-focused infrastructure plans.

By combining CPU and GPU resources, AWS and NVIDIA are positioning the infrastructure for workloads that need both types of compute capability.

AI Factories for U.S. Government Workloads

AWS and NVIDIA also plan to build AI factories for the U.S. government. These systems are expected to provide 100,000 GPUs through AWS’s secure infrastructure.

The planned government infrastructure is intended for federal and national-security workloads classified at Impact Level 6 and above. The announcement identifies this effort as part of the wider AWS-NVIDIA partnership expansion.

RTX PRO 4500 Blackwell Server Edition Instances

AWS said it is the first major cloud provider to offer compute instances accelerated by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs.

According to the companies, these instances provide 4.6 times the AI inference performance of previous-generation instances. The offering adds another NVIDIA-accelerated option within AWS’s cloud infrastructure portfolio.

What the AWS NVIDIA GPU Deployment Includes

AWS’s expanded NVIDIA plans cover more than a single hardware purchase. The collaboration includes:

  • Deployment of 2 million additional NVIDIA GPUs in 2027 and 2028
  • GPU platforms based on Blackwell Ultra, Rubin, and Rubin Ultra
  • More than 1 million NVIDIA GPUs AWS said it would start installing in 2026
  • NVIDIA Vera CPU-based infrastructure for agentic AI workloads
  • AI factories for U.S. government workloads with 100,000 GPUs
  • NVLink Fusion work involving AWS Trainium chips and NVIDIA GPUs
  • RTX PRO 4500 Blackwell Server Edition GPU-accelerated instances