Amazon Doubles Down on Nvidia GPUs Amid Ongoing Investment in Trainium Processors

Amazon's reliance on Nvidia hardware for AI infrastructure has been reaffirmed as it plans to deploy an additional 2 million Nvidia GPUs between 2027 and 2028, doubling its previous commitment of 1 million chips in 2026. This move shows that despite years of investment in its own Trainium processors through Annapurna Labs, Amazon still heavily depends on Nvidia for AI infrastructure.

The extra Nvidia hardware will be part of various projects, including dedicated AI factories for the U.S. government using 100,000 GPUs and the addition of Nvidia's Nemotron models to Amazon Bedrock and SageMaker. Amazon Robotics will also use Nvidia's physical AI platform for warehouse automation.

AWS and Nvidia are expanding their partnership beyond just GPUs. Annapurna Labs, responsible for developing Amazon's Trainium processors, will work with Nvidia to extend NVLink Fusion with high-bandwidth memory for Trainium chips. This collaboration suggests that Amazon is seeking to integrate its internal chip strategy with Nvidia's technology.

The move underscores that Amazon wants more control over its AI stack but acknowledges the importance of Nvidia in meeting current demand levels.