Amazon-Nvidia cooperation just got bigger. A fresh agreement, unveiled Wednesday on Nvidia’s earnings call, sends 2 million more GPUs into AWS data centers, with Blackwell Ultra, Rubin, and Rubin Ultra models arriving across 2027 and 2028.
The agreement follows a May commitment to deploy more than 1 million GPUs starting this year. Nvidia says demand has since exceeded those expectations. Financial terms went undisclosed, though unit counts put the value in the tens of billions.
Integration runs deeper than silicon. Nvidia’s networking gear, open models, CPUs, data processing software, and robotics platform will spread across AWS, and an unspecified number of Vera CPUs will ship there too, some paired with Rubin and some standalone.
Amazon is also building its own chips to loosen Nvidia’s grip. Trainium accelerators and Graviton CPUs anchor a custom silicon business past a $25B annualized revenue run rate. Nvidia chief Jensen Huang still calls Vera a $200B new market, with Oracle and SpaceXAI signed on as lead customers.
