NVIDIA and Amazon Web Services have deepened their strategic partnership at AWS re:Invent, unveiling a series of new integrations designed to deliver high-performance, secure, and scalable compute infrastructure for the next era of AI innovation. The collaboration spans hardware, software, and cloud infrastructure, with a focus on accelerating AI development across industries. A key component of the expansion is AWS’s support for NVIDIA NVLink Fusion, a platform enabling custom AI infrastructure that combines NVIDIA’s scale-up interconnect technology with AWS’s custom silicon. This includes the upcoming Trainium4 chips for inference and agentic AI training, Graviton CPUs for diverse workloads, and the Nitro System virtualization infrastructure. By integrating NVLink Fusion, AWS will enhance performance, simplify deployment, and speed up time to market for next-generation AI capabilities. AWS has already deployed NVIDIA MGX racks at scale with GPUs and is now extending this architecture to support Trainium4, marking the first step in a multi-generational collaboration. The integration allows AWS to leverage the full NVLink Fusion ecosystem—covering racks, chassis, power delivery, and cooling—enabling end-to-end, rack-scale AI deployment. Additionally, the NVIDIA Vera Rubin architecture on AWS will offer customers advanced networking options through the Elastic Fabric Adapter, while maintaining full compatibility with AWS’s cloud infrastructure. The partnership also brings NVIDIA’s Blackwell architecture to AWS, including the HGX B300 and GB300 NVL72 GPUs, providing customers with access to the most powerful GPUs for training and inference. The RTX PRO 6000 Blackwell Server Edition GPUs, designed for visual AI applications, will be available on AWS in the coming weeks. A major new offering, AWS AI Factories, delivers dedicated, sovereign AI infrastructure that enables organizations to run advanced AI services in their own data centers managed by AWS. This ensures data control and compliance with local regulations, making it ideal for public sector and highly regulated industries. The platform combines AWS’s cloud infrastructure with NVIDIA Blackwell GPUs and the full-stack NVIDIA accelerated computing platform, including Spectrum-X Ethernet switches. NVIDIA’s Nemotron open models are now integrated with Amazon Bedrock, allowing developers to build and deploy generative AI applications and agents at scale. With access to Nemotron Nano 2 and Nemotron Nano 2 VL, customers can create AI agents that process text, code, images, and video with high accuracy. Early adopters like CrowdStrike and BridgeWise are already using the service to power specialized AI agents. On the software side, Amazon OpenSearch Service now offers serverless GPU acceleration for vector index building via NVIDIA cuVS, an open-source library. This enables up to 10x faster vector indexing at a quarter of the cost, significantly reducing latency and accelerating AI workflows like retrieval-augmented generation. The partnership also enhances agent development with tools like Strands Agents, NVIDIA NeMo Agent Toolkit, and Amazon Bedrock AgentCore, providing a complete path from prototype to production. This builds on existing integrations with NVIDIA NIM microservices, Riva, BioNeMo, and SageMaker. For physical AI, NVIDIA Cosmos world foundation models are now available as NIM microservices on Amazon EKS for real-time robotics control and simulation. For batch processing and synthetic data generation, Cosmos WFMs run on AWS Batch. Robotics companies including Agility Robotics, ANYbotics, and Skild AI are using this stack for data processing, training, and simulation with NVIDIA Isaac Sim and Isaac Lab. The collaboration has been recognized with the AWS Global GenAI Infrastructure and Data Partner of the Year award, highlighting NVIDIA’s role in advancing generative AI infrastructure. The partnership continues to evolve, with sessions and demonstrations available at AWS re:Invent in Las Vegas through December 5.