NVIDIA Launches Omniverse DSX Blueprint to Build Physically Accurate Digital Twins of Gigawatt-Scale AI Factories
NVIDIA released the Omniverse DSX Blueprint at GTC 2026, a framework for building physically accurate digital twins of AI factories that unifies power, cooling, and network simulation.
As the global buildout of AI data centers accelerates into gigawatt territory, operators face an increasingly complex engineering challenge: optimizing the interplay of power delivery, liquid cooling, network topology, and compute density before a single rack is installed. NVIDIA is betting that the answer lies in simulation, and at GTC 2026 the company released a framework designed to make AI factory digital twins an industry standard.
The Omniverse DSX Blueprint, now generally available, is a reference architecture for building physically accurate digital twins of large-scale AI factories. It unifies simulation of power, cooling, networking, and operations into a single virtual environment, allowing operators to validate designs and optimize performance before construction begins. The framework was announced March 16 at NVIDIA’s GPU Technology Conference alongside the Vera Rubin DSX AI Factory reference design.
“Intelligence tokens are the new currency, and AI factories generate them,” said Jensen Huang, NVIDIA’s CEO, during the announcement. “We provide the foundation for the world’s most productive AI factories.”
A Modular Software Stack for Factory-Scale Simulation
The DSX architecture introduces several software components that address distinct aspects of AI factory operation. DSX Max-Q uses dynamic power provisioning to maximize computing output per watt within fixed power budgets, while DSX Flex connects factories to grid services for dynamic power adjustment, unlocking what NVIDIA describes as stranded energy capacity. DSX Exchange integrates compute, network, energy, and cooling signals into a unified data layer, and DSX Sim validates factory designs using the DSX Air modeling platform.
The blueprint enables operators to simulate airflow patterns, model power utilization, design network topologies, and analyze thermal behavior in a virtual environment, as SiliconANGLE reported. By running these simulations before physical deployment, companies can identify bottlenecks and optimize configurations that would be prohibitively expensive to discover through trial and error in a live facility.
Broad Industry Adoption
The blueprint has attracted integrations from more than a dozen industrial and technology partners. Schneider Electric and Siemens are using it to build physically accurate digital twins of AI factories, while Dassault Systemes and Cadence have incorporated the framework into their respective Systems Engineering and Reality Data Center Digital Twin platforms, according to SiliconANGLE.
AVEVA, the industrial software company owned by Schneider Electric, announced a parallel integration that brings its engineering and operations tools into the Omniverse DSX ecosystem. The collaboration spans the full AI factory lifecycle: AVEVA Unified Engineering converts existing assets into OpenUSD SimReady format for use in Omniverse simulations, AVEVA Process Simulation models advanced liquid-cooling networks, and the AVEVA PI System aggregates IT and operational technology data across the DSX Exchange layer. NVIDIA’s NV-Tesseract model will extend these capabilities with anomaly detection and forecasting.
Other partners contributing to the DSX ecosystem include Eaton, Jacobs, Nscale, Phaidra, Procore, PTC, Switch, Trane Technologies, and Vertiv. Energy partners include GE Vernova, Hitachi, and Siemens Energy, as NVIDIA detailed.
Early Results
Phaidra, a company specializing in autonomous control systems for data centers, has already demonstrated measurable gains from the DSX integration. The company achieved approximately 10 percent additional compute capacity through cooling optimization using the DSX Max-Q framework, according to NVIDIA’s announcement.
The Omniverse DSX Blueprint also fits into a broader pattern visible at GTC 2026, where NVIDIA positioned digital twins not just as design tools but as operational infrastructure. “Factories themselves are now robotic systems,” said Rev Lebaredian, NVIDIA’s vice president of Omniverse and simulation technologies, during the conference, as the NVIDIA blog noted. The implication is that as AI factories grow to gigawatt scale, managing them will require the same kind of continuous simulation and optimization that currently governs autonomous vehicles and industrial robots.
With hyperscalers, colocation providers, and sovereign AI programs all racing to build out compute capacity, the ability to simulate and optimize an entire facility before breaking ground could shave months off deployment timelines and prevent costly design errors. Whether the Omniverse DSX Blueprint becomes the de facto standard for AI factory planning will depend on how quickly its partner ecosystem delivers production-ready integrations — but with the breadth of companies already on board, NVIDIA has staked out a commanding early position.