NVIDIA and Amazon Web Services unveiled a groundbreaking expansion of their full-stack partnership at AWS re:Invent 2025 in Las Vegas on December 2, 2025. The collaboration integrates NVIDIA’s NVLink Fusion platform into AWS’s custom silicon, including next-generation Trainium4 AI chips, Graviton CPUs, and Nitro System, to deliver unprecedented performance for cloud-scale AI workloads. NVIDIA CEO Jensen Huang described it as the “compute fabric for the AI industrial revolution,” while AWS CEO Matt Garman emphasized over 15 years of joint innovation for superior efficiency and scalability.
Hardware Breakthroughs
AWS will leverage NVLink Fusion and NVIDIA’s MGX rack architecture to accelerate Trainium4 deployments, enabling rapid chip-to-chip connectivity and simplified rack-scale systems. New Trainium3 servers, available now, feature 144 chips per rack with 4x compute power and 40% lower energy use than predecessors. NVIDIA Blackwell GPUs, such as HGX B300 and GB300 NVL72, power AWS AI Factories—dedicated infrastructures for enterprises to run sovereign AI in their data centers while ensuring data control and regulatory compliance.
Software and Physical AI Advances
NVIDIA Nemotron open models now integrate with Amazon Bedrock, allowing developers to build efficient agentic AI apps for text, code, images, and video; early adopters include CrowdStrike and BridgeWise. Amazon OpenSearch Service gains serverless GPU acceleration via NVIDIA cuVS, delivering up to 10x faster vector indexing at quarter the cost. For physical AI, NVIDIA Cosmos world models launch as NIM microservices on Amazon EKS and Batch, supporting robotics firms like Agility Robotics in simulation and data generation with Isaac Sim and Lab.
This alliance positions NVIDIA and AWS to dominate AI infrastructure, enabling global sovereign clouds and transforming federal supercomputing for public sector needs. NVIDIA also received AWS’s Global GenAI Infrastructure Partner of the Year award, capping a week of re:Invent announcements through December 5.
