Amazon’s Nvidia GPU Bet Swells to 3 Million as AI Demand Outruns Expectations

Amazon’s Nvidia GPU Bet Swells to 3 Million as AI Demand Outruns Expectations
Amazon is committing to a far bigger Nvidia buildout just months after its first mega-order, a sign that the race for AI computing capacity is accelerating faster than cloud providers can plan for it.

The latest agreement adds 2 million Nvidia GPUs to AWS data centers in 2027 and 2028, on top of more than 1 million GPUs Amazon had already agreed to deploy beginning this year. Nvidia said demand had “exceeded those expectations,” driven by startups, enterprises, AI labs and governments seeking capacity to train and run models.

The scale matters because it comes as Amazon tries to reduce its reliance on the chipmaker. AWS has been developing its own Trainium AI accelerators and Arm-based Graviton server CPUs, and has said its custom-chip business reached a $25 billion annualized revenue run rate. Yet the expanded deal makes clear that Amazon still needs Nvidia’s hardware at industrial scale to satisfy customers now.

This is not simply a chip purchase. AWS will also bring Nvidia’s Vera CPUs into its cloud, while integrating Nvidia networking, data-processing software and open models into services including Bedrock and SageMaker. Amazon plans to use Nvidia’s physical-AI stack—Omniverse, Cosmos, Isaac and Jetson—for warehouse robotics as well.

A second account of the announcement framed the move as a doubling-down by both companies, landing alongside another quarter of record Nvidia revenue and Amazon’s plan to offer Vera CPUs through AWS. Nvidia reported $96.2 billion in second-quarter sales, including $89 billion in data-center revenue, and forecast $108 billion for the third quarter.

Nvidia chief executive Jensen Huang argued the spending reflects a new commercial stage for AI: “AI is now doing productive and useful work.” The unresolved question is whether the enormous supply of new compute will yield the profits both companies expect—or merely deepen the industry’s infrastructure arms race.

Continue reading https://foxvector.com/stories/01a0410d-369c-0fde-716b-29eefa31b82c