AWS, Nvidia expand AI partnership with plan for 2 million more GPUs

Amazon Web Services (AWS) and Nvidia are deepening their long-standing collaboration, unveiling an ambitious expansion that will bring more than two million additional GPUs into AWS's infrastructure over the coming years. The move signals a major bet on the continued surge in demand for AI compute capacity, as enterprises and research institutions race to deploy large-scale machine learning workloads.

The expanded partnership, announced Wednesday, will see Nvidia supply AWS with a broad range of its latest accelerators, including next-generation GPUs designed for both training and inference tasks. Beyond the sheer volume of chips, the two companies plan to jointly develop new infrastructure tailored for agentic AI systems and physical AI applications, which require increasingly complex and real-time processing capabilities. These efforts are expected to span custom networking, cooling, and software integration optimized for high-density GPU clusters.

AWS has long been one of Nvidia's largest customers, and this latest commitment reinforces the cloud provider's strategy of positioning itself as the primary destination for AI-heavy workloads. The additional two million GPUs represent a significant scaling of AWS's AI capacity, building on previous multi-billion-dollar investments in Nvidia-based instances. Industry analysts note that the sheer scale of this deployment could tighten global GPU supply further, with Nvidia's production capacity already stretched by demand from hyperscalers and AI startups alike.

The collaboration also underscores a broader industry trend: cloud providers are increasingly locking in long-term access to advanced silicon as AI infrastructure becomes a critical competitive differentiator. For Nvidia, the deal provides a stable, high-volume revenue stream and a reference architecture that can be showcased to other cloud and enterprise customers. For AWS, it ensures the company can meet what it describes as "unprecedented customer demand" for AI services, ranging from foundation model training to edge inference.

Executives from both companies framed the expansion as a response to the rapid evolution of AI use cases, particularly in areas like robotics, autonomous systems, and digital twins, where physical AI is beginning to merge with cloud-based intelligence. The joint roadmap will also focus on improving energy efficiency and total cost of ownership for AI infrastructure, a growing concern as data center power consumption becomes a limiting factor in many regions.