AWS and NVIDIA to add 2 million GPUs for agentic and physical AI

Amazon and NVIDIA announced plans to deploy two million additional NVIDIA GPUs — Blackwell Ultra, Rubin and Rubin Ultra — across Amazon Web Services’ global infrastructure through 2027 and 2028, alongside new infrastructure built on NVIDIA’s forthcoming Vera CPUs. The companies disclosed no total dollar value for the expansion. The announcement came the same day NVIDIA reported record quarterly results and gave its first year-ahead revenue guidance.

Of the new capacity, 100,000 GPUs are earmarked for a set of AWS-built “AI factories” serving US federal and national-security customers, designed to run workloads classified at Impact Level 6 and above — the tier covering the most sensitive government data. AWS said it would also expand its own Trainium chips within the same infrastructure, using NVIDIA’s NVLink Fusion interconnect to let Trainium and NVIDIA GPUs share a common rack-scale architecture, rather than positioning the two as competing options for customers to choose between.

NVIDIA chief executive Jensen Huang said the companies had “built one of the great growth engines of the AI era, and demand is running ahead of every forecast.” Amazon Web Services chief executive Matt Garman framed the deal around customer choice: “Customers want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together.”

The announcement extended a run of large capacity commitments NVIDIA had made over the preceding weeks, including a roughly $500 billion financing platform with six Wall Street firms aimed at funding AI-infrastructure build-out more broadly. As with those deals, the figures describe planned capacity and partnership scope rather than committed spending, and neither company published a schedule for when the two million GPUs would actually ship.