Amazon is dramatically expanding its AI footprint, announcing plans to deploy 2 million more Nvidia GPUs across AWS’s global infrastructure between 2027 and 2028. This expansion builds on a previous commitment to roll out over 1 million Nvidia chips starting this year, signaling Amazon’s belief that AI workloads — from agentic systems to robotics — will continue scaling far beyond current forecasts.
Image Courtesy : nvidianews.nvidia.com
The GPUs involved include Nvidia’s most advanced architectures: Blackwell Ultra, Rubin, and Rubin Ultra, processors designed for frontier‑model training, massive inference workloads, and high‑bandwidth data processing. These chips typically cost tens of thousands of dollars each, placing the total value of the deal in the tens of billions — though neither company has disclosed exact financial terms.
Amazon’s move isn’t just about adding GPUs. The expanded partnership deepens integration across the entire AI stack. AWS will incorporate Nvidia Vera CPUs, advanced networking technologies, and Nvidia’s physical‑AI robotics platforms into its infrastructure. This includes building AI factories — specialized clusters optimized for large‑scale model development — including a 100,000‑GPU facility dedicated to U.S. government workloads.
The collaboration also extends into software. AWS will offer Nvidia’s Nemotron open‑model family through Amazon Bedrock and SageMaker, giving customers more flexibility in choosing foundation models. Meanwhile, Amazon Robotics will adopt Nvidia’s physical‑AI stack — including Omniverse, Cosmos, Isaac, and Jetson — to accelerate warehouse automation.
This massive GPU expansion comes as AWS continues developing its own silicon, including Trainium for AI training and Graviton CPUs for general‑purpose workloads. Amazon recently reported that its custom‑chip business surpassed a $25 billion annualized revenue run rate, showing that the company is balancing both in‑house innovation and deep reliance on Nvidia hardware.
Nvidia, for its part, is experiencing record demand. The company recently posted $96.2 billion in quarterly revenue, with data‑center sales up 117% year‑over‑year — momentum fueled directly by hyperscalers like AWS scaling GPU fleets at unprecedented speed.
Amazon’s bet is clear: AI demand isn’t slowing. Enterprises are moving from pilot projects to full‑scale production, agentic AI systems are exploding in complexity, and robotics workloads are becoming more compute‑intensive. By securing 2 million additional GPUs, Amazon is positioning AWS as the backbone of global AI development for the next decade.