Nvidia pitches Omniverse DSX as model for gigawatt-scale AI factories
GPU giant teams with partners to create digital twin blueprint for next-gen datacenters
Nvidia unveiled Omniverse DSX at its GTC event in Washington DC — a blueprint for designing and operating gigawatt-scale AI datacenters using digital twin technology.
The GPU giant also confirmed the build-out of an AI Factory Research Center at Digital Realty's Manassas, Virginia site to develop the platform, which combines Nvidia's Omniverse simulation environment with open source Universal Scene Description (OpenUSD) technology.
"Nvidia started out by designing chips, and then we started to design systems, and we designed AI supercomputers. Now we're designing entire AI factories," said CEO Jensen Huang during his GTC keynote.
The platform creates a digital twin of the physical datacenter, including building, power, and cooling infrastructure that acts as an operating system once the facility goes live. Nvidia claims this enables continuous optimization for performance, energy efficiency, and sustainability.
Datacenter infrastructure provider Schneider Electric is partnering on the project, having announced its own Omniverse-based digital twin platform at GTC in March.
Huang has repeatedly positioned these GPU-dense facilities as distinct from traditional datacenters.
"It's an AI factory because this factory produces one thing," he said on stage. "It runs AI. And its purpose is designed to produce tokens that are as valuable as possible, meaning they have to be smart. And you want to produce these tokens at incredible rates because when you ask an AI for something, you would like it to respond."
"This AI factory is what we're building for Vera Rubin, and we created a technology that makes it possible for all of our partners to integrate into this factory digitally," he added.
Huang previously claimed that AI factories will become "the bedrock of modern economies" around the world.
Nvidia also unveiled BlueField-4 at GTC, its next-genetation data processing unit (DPU) for accelerating infrastructure operations inside AI datacenters.
Essentially designed to offload network and storage functions, BlueField-4 combines a Grace CPU featuring 64 Arm Neoverse V2 cores with ConnectX-9 networking and supporting Ethernet and InfiniBand at up to 800 Gbps per port. It claims the DPU has 6x the compute power of BlueField-3.
The new silicon is purpose-built as an end-to-end engine for a new class of AI storage platform, serving as the foundation of high-performance data pipelines for AI factories.
Early availability of BlueField-4 is expected to come with the launch of the Vera Rubin platform in 2026. ®