NVIDIA and Amazon Web Services are expanding their collaboration by integrating NVIDIA NVLink Fusion into AWS custom silicon, including the next-generation Trainium4 chip, Graviton CPUs, and the AWS Nitro System. This expansion supports NVIDIA NVLink Fusion—a platform for custom AI infrastructure—by combining NVLink scale-up interconnect and the MGX rack architecture with AWS’s silicon to enhance performance and accelerate cloud-scale AI capabilities. The integration of NVLink Fusion aims to simplify deployment and systems management, while also leveraging a comprehensive supplier ecosystem for full rack-scale deployment, ultimately driving advancements in AI infrastructure and service rollout.
Expanded NVIDIA and AWS Technology Integration
NVIDIA and AWS are expanding their full-stack partnership to deliver high-performance compute for AI innovation. A key component is the integration of NVIDIA NVLink Fusion into AWS custom silicon, including the next-generation Trainium4 chips, Graviton CPUs, and the Nitro System. This combination of NVIDIA’s NVLink scale-up interconnect and MGX rack architecture with AWS’s silicon aims to increase performance and accelerate time to market for cloud-scale AI capabilities, with AWS already deploying MGX racks at scale.
The collaboration extends to accelerated computing with the NVIDIA Blackwell architecture, offering customers immediate access to advanced GPUs like the HGX B300 and GB300 NVL72. AWS AI Factories will utilize these GPUs, providing secure, sovereign AI infrastructure globally, and allowing organizations to maintain control over their data while meeting regulatory requirements. This will transform federal supercomputing and AI landscapes by seamlessly integrating AWS infrastructure with the full-stack NVIDIA accelerated computing platform.
Beyond hardware, NVIDIA and AWS are integrating software. NVIDIA Nemotron open models are now available with Amazon Bedrock, allowing developers to build generative AI applications at scale. Amazon OpenSearch Service now offers serverless GPU acceleration for vector indexing, powered by NVIDIA cuVS, achieving up to 10x faster indexing at a quarter of the cost. These advancements, alongside tools like Strands Agents and the NVIDIA NeMo Agent Toolkit, aim to accelerate the path from prototype to production for AI agents.
Convergence of Scale and Sovereignty with AI Factories
The convergence of scale and sovereignty is being addressed through AWS AI Factories, a new offering providing dedicated infrastructure for advanced AI services. These factories will be operated by AWS, but allow customers to maintain control of their data and comply with local regulations. NVIDIA and AWS are committing to globally deploy these sovereign AI clouds, delivering unprecedented computing capabilities while meeting increasingly rigorous sovereign AI requirements, especially for public sector organizations.
AWS is integrating NVIDIA’s NVLink Fusion with its custom silicon—Trainium4 chips, Graviton CPUs, and the Nitro System—to increase performance and accelerate cloud-scale AI capabilities. This collaboration unifies NVIDIA’s scale-up architecture with AWS’s custom silicon. AWS has already deployed MGX racks with NVIDIA GPUs and is expanding accelerated computing with NVIDIA Blackwell architecture, including HGX B300 and GB300 NVL72 GPUs, giving customers access to the industry’s most advanced GPUs.
Beyond hardware, the partnership extends to software integrations. NVIDIA Nemotron open models are now integrated with Amazon Bedrock, allowing customers to build generative AI applications at production scale. Furthermore, Amazon OpenSearch Service now offers serverless GPU acceleration for vector index building, powered by NVIDIA cuVS, achieving up to 10x faster vector indexing at a quarter of the cost. This co-engineering effort aims to accelerate the data backbone of every enterprise.
AWS and NVIDIA have worked side by side for more than 15 years, and today marks a new milestone in that journey.
Matt Garman, CEO of AWS
NVIDIA Software Optimizations Within the AWS Ecosystem
NVIDIA and AWS are expanding their collaboration to integrate NVIDIA NVLink Fusion into AWS custom silicon, including next-generation Trainium4 chips, Graviton CPUs, and the Nitro System. This integration combines NVIDIA’s NVLink scale-up interconnect and MGX rack architecture with AWS’s hardware, aiming to increase performance and accelerate time to market for cloud-scale AI. AWS has already deployed MGX racks and sees NVLink Fusion simplifying deployment and systems management across its platforms while leveraging a full supplier ecosystem.
The partnership extends beyond hardware with NVIDIA’s software now integrated into the AWS AI ecosystem. NVIDIA Nemotron open models are integrated with Amazon Bedrock, allowing customers to build generative AI applications at scale. Additionally, Amazon OpenSearch Service now offers serverless GPU acceleration for vector index building powered by NVIDIA cuVS, resulting in up to 10x faster indexing at a quarter of the cost.
To further accelerate AI development, NVIDIA and AWS are co-engineering at the software layer. Tools like Strands Agents, the NVIDIA NeMo Agent Toolkit, and Amazon Bedrock AgentCore provide a complete path from prototype to production for AI agents. Furthermore, NVIDIA Cosmos world foundation models are available on Amazon EKS and AWS Batch, supporting both real-time robotics control and large-scale synthetic data generation.
Accelerating Physical AI and Robotics Development
Accelerating physical AI and robotics development is a key focus of the expanded NVIDIA and AWS partnership. NVIDIA Cosmos world foundation models are now available on both Amazon EKS and AWS Batch, enabling real-time robotics control and large-scale synthetic data generation. This allows for training and validation of robot models using simulation before real-world deployment, addressing the need for high-quality datasets critical for physical AI.
AWS and NVIDIA are collaborating to build a compute fabric for the “AI industrial revolution,” with NVIDIA NVLink Fusion integrated into AWS Trainium4 chips, Graviton CPUs, and the Nitro System. This integration aims to increase performance and accelerate time to market for cloud-scale AI capabilities. By combining NVIDIA’s NVLink scale-up interconnect and MGX rack architecture with AWS custom silicon, the companies are creating a new generation of accelerated platforms.
The partnership also extends to software optimization. Amazon OpenSearch Service now offers serverless GPU acceleration for vector index building, powered by NVIDIA cuVS, achieving up to 10x faster indexing at a quarter of the cost. This advancement, alongside tools like Strands Agents and the NVIDIA NeMo Agent Toolkit, provides a complete path from prototyping to production-ready AI agents, crucial for dynamic AI techniques.
