NVIDIA and AWS deepen their alliance with 2 million more GPUs, Vera CPU infrastructure, and a government AI factory.
The AMW Read
Extends the known AWS-NVIDIA compute buildout with an explicit 2M-GPU hyperscaler-scale commitment, new Vera CPU/NVLink Fusion/NVHBM silicon integration, and a dedicated federal AI factory spanning infrastructure, data, and robotics workloads.
NVIDIA and AWS deepen their alliance with 2 million more GPUs, Vera CPU infrastructure, and a government AI factory.
AWS and NVIDIA announced an expansion of their 16-year partnership, committing 2 million additional NVIDIA GPUs — including Blackwell Ultra, Rubin, and Rubin Ultra — across AWS's global infrastructure between 2027 and 2028, on top of the 1 million-plus GPUs pledged at GTC 2026. The expansion brings NVIDIA Vera CPU-based infrastructure to AWS, extends NVLink Fusion with NVIDIA's custom high-bandwidth memory (NVHBM) for AWS's next-generation Trainium chips, and builds a secure AI factory with 100,000 GPUs for U.S. federal and national-security workloads rated for Impact Level 6 classification. AWS will be the first major cloud provider to offer NVIDIA's RTX PRO 4500 Blackwell Server Edition GPUs via new EC2 G7 instances, and will keep supporting NVIDIA's open Nemotron models on Bedrock and SageMaker alongside GPU-accelerated data processing on EMR and OpenSearch.
The scale signals that AI compute demand is still outrunning hyperscaler capacity plans announced only months earlier — AWS says demand has already exceeded what it projected at GTC 2026. Weaving Vera CPUs, custom NVHBM memory, and NVLink Fusion into AWS's stack ties Amazon's own Trainium silicon more tightly to NVIDIA's platform rather than positioning it as a standalone alternative, reinforcing NVIDIA as the default substrate for agentic and physical AI workloads even at a hyperscaler with in-house chips. The government AI factory also places NVIDIA and AWS at the center of federal AI procurement, a segment where security-clearance requirements have historically slowed vendor entry.
For builders, RTX PRO 4500-equipped G7 instances (NVIDIA cites 4.6x inference and 2.1x graphics performance over G6) and broader Nemotron availability on Bedrock lower the barrier to running agentic and robotics workloads inside AWS's managed environment — Amazon Robotics is adopting NVIDIA's physical AI platform for warehouse automation. For investors, the size of this commitment, even without a disclosed dollar figure, underscores that GPU capacity remains the binding constraint hyperscalers are racing to relax through 2028.
