AWS partners with Nvidia to use NVLink in AI chips (techoreon.com)

Amazon Web Services (AWS) has announced a partnership with Nvidia to integrate the NVLink Fusion interconnect into the next generation of its Trainium AI accelerators. This collaboration, revealed at the re:Invent conference, enables AWS to build larger AI training clusters where thousands of chips can operate cohesively as a single system. Although a release date for the updated Trainium devices has not been disclosed, the partnership is viewed as pivotal by Nvidia CEO Jensen Huang, who emphasized its potential to drive the AI industrial revolution. The integration of NVLink will enhance AWS's offerings by allowing customers access to specialized "AI Factories," on-premise setups that combine Trainium processors with Nvidia GPUs and include AWS's Bedrock and SageMaker services. This move is strategically significant as it aims to improve AWS's competitive edge in AI performance while potentially reducing costs compared to rivals. Additionally, AWS announced updates to its Trainium3 servers, boasting quadrupled training throughput and reduced energy consumption, alongside new foundation models such as Nova 2 and Sonic for various AI tasks. These developments signal AWS's continuous push to lead in the AI/ML space amidst increasing competition.