NVIDIA and AWS deepen partnership with NVLink Fusion integration into next-gen AI hardware

At AWS re:Invent, NVIDIA and Amazon Web Services announced a strategic expansion, integrating NVIDIA’s NVLink Fusion into AWS’s customised chips to supercharge AI performance, enabling faster deployment, enhanced inference, and new cloud and on-premise AI solutions.

At the recent AWS re:Invent conference, NVIDIA and Amazon Web Services (AWS) announced a significant expansion of their strategic collaboration, deepening integration across several advanced technology domains including interconnect technology, cloud infrastructure, open models, and physical artificial intelligence (AI). Central to this partnership is the integration of NVIDIA’s cutting-edge NVLink Fusion technology into AWS’s customized silicon chips, notably the next-generation Trainium4 AI chip, as well as AWS’s Graviton CPUs and Nitro System virtualization platform.

According to the official announcement, AWS plans to adopt NVIDIA NVLink Fusion, a platform specifically designed for custom AI infrastructure, to enhance the performance and reduce the time to market of its next-generation cloud-scale AI capabilities. This integration combines NVIDIA’s vertical scaling NVLink technology, MGX rack architecture, and AWS’s tailored chips, enabling accelerated deployment of AI models especially for inference and agentic AI training workloads. Specifically, Trainium4 represents the first step in a multi-generational collaboration focused on utilizing NVLink Fusion to create a new class of high-performance AI infrastructure.

The technology integration extends beyond hardware. AWS is embedding NVLink Fusion into broader infrastructure solutions such as the Vera Rubin architecture for networking, which supports AWS Elastic Fabric Adapter and the Nitro System, ensuring full compatibility with AWS cloud services while providing enhanced network options to accelerate AI service delivery. Importantly, AWS has already widely deployed NVIDIA GPU-powered MGX racks and sees NVLink Fusion as a critical enabler for simplifying deployment and systems management across its platforms.

These hardware advances underpin AWS’s broader AI ecosystem expansion, including the launch of AWS AI Factories, a cloud service delivering dedicated AI infrastructure within customers’ own data centres. The AI Factories service allows organisations to leverage cutting-edge AI capabilities like those powered by NVIDIA’s Blackwell GPU architecture, while maintaining strong control over their data and adhering to stringent local regulations. This service is particularly significant for government-level high-performance computing and AI workloads, providing a secure and scalable platform to support sovereign AI needs.

AWS’s investment in the collaboration is also reflected in the recent launch of new servers based on the Trainium3 chip, offering over four times the performance of previous AI infrastructure with 40% reduced power consumption, a development that Reuters reports as part of AWS’s ongoing commitment to enhancing AI compute efficiency.

Beyond hardware, the collaboration expands into software optimisation. NVIDIA’s open models through Nemotron are now integrated with Amazon Bedrock, helping customers build production-ready generative AI and agentic AI applications. This integration allows on-demand access to high-performance NVIDIA models within Amazon’s serverless infrastructure. Leading industry players such as CrowdStrike and BridgeWise have already adopted this combined service to deploy specialised AI agents.

From a developer perspective, AWS and NVIDIA have jointly advanced GPU-accelerated vector indexing through Amazon OpenSearch Service, enabling up to a tenfold speed increase and a 75% reduction in cost for unstructured data processing tasks, marking a paradigm shift where GPU acceleration becomes central in managing AI-driven workloads.

Furthering AI innovation into physical realms, NVIDIA’s Cosmos World Foundation Model (WFM) now runs on AWS’s containerized services, supporting real-time robot control and simulation workloads. This seamlessly integrates with NVIDIA’s Isaac simulation platforms widely used by robotics leaders like Agility Robotics and ANYbotics, who use these tools for large-scale data collection, training, and simulation to accelerate robotic intelligence development.

NVIDIA’s founder and CEO Jensen Huang remarked on this synergy, highlighting how surging GPU computing demands are driving a virtuous cycle of smarter AI applications necessitating more compute power. He emphasised that the NVLink Fusion integration with AWS’s custom chips is crafting the foundation for the AI revolution, making advanced AI accessible globally and accelerating the world’s transition to smarter technologies.

Echoing this vision, AWS CEO Matt Garman underscored the 15-year partnership with NVIDIA as a journey reaching a new milestone. He stressed that the integration of NVLink Fusion across Trainium4, Graviton processors, and the Nitro System will provide customers with unprecedented performance, efficiency, and scalability, empowering faster innovation across myriad industries.

This expanded collaboration signals a strategic alignment in cloud AI infrastructure, blending NVIDIA’s advanced silicon and software ecosystem with AWS’s global cloud platform and bespoke AI hardware developments. Together, they are poised to drive next-generation AI innovations, compelling for enterprises aiming to harness transformative AI capabilities at scale while ensuring compliance, security, and operational efficiency.

📌 Reference Map:

• ^[1] (Newtalk) – Paragraphs 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
• ^[2] (Reuters) – Paragraph 3, 4
• ^[3] (About Amazon) – Paragraph 6
• ^[4] (NVIDIA blog) – Paragraph 2, 3
• ^[5] (NVIDIA developer) – Paragraph 3
• ^[6] (NVIDIA news) – Paragraph 4
• ^[7] (NVIDIA) – Paragraph 3, 5

Source: Fuse Wire Services