Throughout the GTC Washington, D.C., keynote right this moment, NVIDIA founder and CEO Jensen Huang launched NVIDIA Omniverse DSX, a complete, open blueprint for designing and working gigawatt-scale AI factories — validated on the new AI Manufacturing facility Analysis Middle at Digital Realty’s web site in Manassas, Virginia.
The blueprint brings collectively ecosystem companions throughout the trade which are tapping into NVIDIA Omniverse libraries and OpenUSD to set a brand new commonplace for constructing and working gigascale AI factories. For the primary time, the constructing, energy and cooling will be codesigned with NVIDIA’s AI Infrastructure stack — and the method of the Virginia analysis heart construct out is captured as a scalable, repeatable recipe validated in a real-world deployment.
OpenUSD and SimReady Asset Adoption Speed up the Design Course of
It begins with the digital twin, the place engineering and building accomplice Jacobs can design the constructing, then optimize compute density and general structure to maximise token throughput in accordance with energy constraints.
Jacobs engineers mixture SimReady OpenUSD property for energy, electrical and cooling gear from companions together with Siemens, Schneider Electrical, Trane Applied sciences and Vertiv into PTC’s product lifecycle administration system — permitting cross-functional groups to collaborate in actual time on photorealistic 3D fashions, speed up design selections and cut back improvement danger with shared, version-controlled product information.
The development of SimReady property from {hardware} companions and rising adoption of OpenUSD throughout industries performs an important function within the acceleration of constructing out these extremely advanced designs.
The Cadence Actuality Digital Twin platform has been purpose-built on high of the blueprint, with built-in Omniverse libraries and accelerated by NVIDIA CUDA. This enables engineers to simulate thermals and electricals in a high-fidelity, bodily correct 3D atmosphere to check designs, predict failures and optimize operations within the digital twin earlier than building.
Bodily Construct-Out and Operations Optimized With Digital Twins
As soon as the AI manufacturing unit has been designed just about, NVIDIA companions like Bechtel and Vertiv ship prefabricated modules which are manufacturing unit constructed, examined and able to plug in. This considerably reduces construct time and permits for modular scalability, serving to obtain sooner time to income.
The structure of the DSX blueprint means as soon as the bodily AI manufacturing unit comes on-line, the digital twin can be utilized as an working system to observe, examine and additional optimize processes.
For instance, AI brokers from companions together with Phaidra and Emerald AI are skilled within the digital twin, after which permit engineers to immediate and optimize energy consumption to scale back pressure on each the AI manufacturing unit and the facility grid. Phaidra AI brokers constantly optimize energy, cooling and workloads, turning the AI manufacturing unit digital twins right into a self-learning system that reinforces grid flexibility, resiliency and vitality effectivity.
Emerald AI integrates its reference design in DSX to allow utilities to unlock underutilized grid capability.
What DSX Delivers
DSX maximizes GPU productiveness and vitality effectivity throughout the compute cloth with a single digital twin atmosphere that may be simulated, optimized and used for operations. Prefabricated modules from infrastructure companions show how AI factories will be designed, assembled, validated and expanded with velocity and consistency.
Three Pillars of DSX:
- DSX Flex – Dynamic Grid Collaboration: Balances vitality demand with real-time grid circumstances utilizing Omniverse telemetry and AI brokers, enhancing effectivity and grid cooperation to assist faucet among the 100 gigawatts of underutilized grid capability.
- DSX Enhance – Efficiency-Per-Watt Optimization: Applies NVIDIA’s power-optimization applied sciences on the information heart stage, working workloads at Max-Q effectivity — the purpose of optimum efficiency per watt — for as much as 30% greater GPU throughput inside the similar energy envelope.
- DSX Trade – Unified IT/OT Integration: Acts because the safe information and management cloth linking operational methods for energy, cooling and security with NVIDIA’s software program stack and Omniverse digital twins via real-time utility programming interfaces and accomplice integrations.
Why It Issues
- Scalability: Frequent structure for AI factories from 100 megawatts to multi-gigawatts.
- Effectivity: DSX Enhance maximizes token throughput per megawatt.
- Sustainability: DSX Flex allows renewable technology and adaptive grid steadiness.
- Integration: DSX Trade unifies digital twins, facility methods and operations.
Along with vitality and gear specialists — Eaton, GE Vernova, Hitachi, Schneider Electrical, Siemens Vitality, Trane Applied sciences and Vertiv — DSX combines experience in energy, cooling and grid interplay to ship resilient, environment friendly operations optimized for the NVIDIA Grace Blackwell and Vera Rubin platforms.
This ecosystem extends past {hardware}, to software program and AI answer suppliers, with Cadence, Emerald AI, Phaidra, PTC, Schneider Electrical ETAP, Siemens and Change collaborating to construct agentic digital twin options. Their adoption of OpenUSD and Omniverse libraries assist to energy dynamic simulation, autonomous management and end-to-end optimization for each stage of the AI manufacturing unit lifecycle.
Designed to scale to multi-gigawatt capability, DSX accelerates design cycles and validates infrastructure earlier than deployment.
Watch the Omiverse DSX demo and learn the complete press launch to be taught extra about how NVIDIA is shaping the period of gigascale AI factories.
Be taught extra about how NVIDIA and companions are advancing AI innovation within the U.S. by watching the NVIDIA GTC Washington, D.C., keynote by Huang.
See discover relating to software program product data.