NVIDIA Unveils Omniverse DSX Blueprint to Accelerate Global Buildout of Gigawatt-Scale AI Factories with Digital Twin-Driven Design and Operations

At the GTC Washington, D.C. keynote, NVIDIA CEO Jensen Huang unveiled Omniverse DSX, a comprehensive open blueprint for designing and operating gigawatt-scale AI factories. The announcement was made at the newly established AI Factory Research Center in Manassas, Virginia, where the DSX framework has been validated in a real-world deployment. Omniverse DSX brings together a broad ecosystem of industry partners to standardize the development of massive AI infrastructure. It enables the co-design of buildings, power systems, and cooling solutions with NVIDIA’s AI infrastructure stack, turning the construction process into a scalable, repeatable blueprint. The entire build-out of the Virginia facility has been captured as a proven, reusable model. The foundation of DSX lies in digital twins powered by OpenUSD and SimReady assets. Jacobs, a key engineering partner, uses these assets—provided by companies like Siemens, Schneider Electric, Trane, and Vertiv—within PTC’s product lifecycle management system. This allows teams across disciplines to collaborate in real time on photorealistic 3D models, make faster decisions, and reduce risk through shared, version-controlled data. The Cadence Reality Digital Twin platform, built specifically on the DSX blueprint and accelerated by NVIDIA CUDA, enables high-fidelity simulations of thermal and electrical behavior. Engineers can test designs virtually, predict failures, and optimize performance before any physical construction begins. Once built, AI factories constructed using DSX leverage prefabricated modules from partners like Bechtel and Vertiv. These modules are factory-assembled, tested, and ready to deploy, drastically cutting build time and enabling rapid scalability. The digital twin then becomes the operational backbone of the facility, continuously monitoring and optimizing performance. AI agents from companies such as Phaidra and Emerald AI are trained in the digital twin environment to autonomously manage power, cooling, and workloads. Phaidra’s agents dynamically adjust operations to reduce strain on both the AI factory and the broader power grid, creating a self-learning system that enhances energy efficiency and grid resilience. Emerald AI uses the DSX framework to help utilities unlock underutilized grid capacity. DSX is designed to maximize GPU productivity and energy efficiency across the compute fabric, all within a unified digital twin environment. It supports end-to-end lifecycle management—from design and simulation to deployment and operations—using a single, consistent model. The DSX ecosystem includes major players in power, cooling, and grid integration such as Eaton, GE Vernova, Hitachi, Schneider Electric, Siemens Energy, Trane, and Vertiv. It also extends to software and AI solution providers like Cadence, PTC, Schneider Electric ETAP, Siemens, and Switch, all using OpenUSD and Omniverse libraries to enable dynamic simulation, autonomous control, and system-wide optimization. Built to scale to multi-gigawatt levels, DSX significantly shortens design cycles and validates infrastructure before real-world deployment. The framework is a key enabler in the race to build the next generation of AI infrastructure, supporting NVIDIA’s Grace Blackwell and Vera Rubin platforms. For more details, viewers can watch the Omniverse DSX demo and read the full press release. The full keynote by Jensen Huang is available to learn how NVIDIA and its partners are advancing AI innovation across the United States.