Details
- NVIDIA announced the Vera Rubin DSX AI Factory reference design, a guide for codesigned AI infrastructure maximizing tokens per watt, and general availability of the Omniverse DSX Blueprint for physically accurate digital twins of AI factories.
- Involved parties include NVIDIA, industry leaders like Cadence, Dassault Systèmes, Eaton, Jacobs, NScale, Phaidra, Procore, PTC, Schneider Electric, Siemens, Switch, Trane Technologies, Vertiv, and energy firms Emerald AI, GE Vernova, Hitachi, Siemens Energy.
- New features include open DSX software stack with DSX Max-Q for compute optimization, DSX Flex for grid power management, DSX Exchange for system integration, and DSX Sim for high-fidelity simulations using Omniverse libraries to model power, cooling, networking before deployment.
- Compares to traditional methods by enabling pre-construction validation via digital twins, addressing limitations in modeling full systems, power flexibility, unlike prior designs lacking integrated simulation for gigawatt-scale factories.
- Web searches confirm broad adoption, including Vertiv's OneCore Rubin DSX for prefabricated infrastructure scaling to 1 GW and DDN/Aleria's sovereign AI factory for regulated sectors, highlighting ecosystem scale for faster deployment and efficiency.
Impact
NVIDIA's DSX tools address critical power and cooling bottlenecks in AI infrastructure, enabling gigawatt-scale factories with optimized tokens-per-watt amid $300B equipment backlogs and 200 GW U.S. queues. Broad partner integration accelerates time-to-revenue, strengthens grid stability, and positions NVIDIA to dominate AI data center design against competitors like custom hyperscaler builds.
