NVIDIA's latest announcement introduces the Omniverse DSX blueprint, a comprehensive, open framework designed to build and operate gigawatt-scale AI factories. This initiative directly addresses the escalating challenge of AI energy consumption, offering a standardized approach to designing and managing the massive infrastructure required for advanced AI. Validated at Digital Realty’s new AI Factory Research Center, DSX signals a critical industry shift towards more efficient and sustainable AI deployment.
The blueprint's foundation lies in the digital twin, leveraging NVIDIA Omniverse libraries and OpenUSD to co-design buildings, power, and cooling systems with the AI infrastructure stack. Engineering and construction partners like Jacobs can optimize compute density and layout virtually, maximizing token throughput within strict power constraints. This pre-construction simulation, incorporating SimReady assets from equipment partners such as Siemens and Schneider Electric, is crucial for mitigating risks and accelerating design decisions before any physical build begins.
Once designed, the physical AI factory benefits from prefabricated modules delivered by partners like Bechtel and Vertiv, significantly reducing construction time and enabling modular scalability. Critically, the digital twin then transitions into an operational system, continuously monitoring and optimizing processes in real-time. This continuous feedback loop is essential for managing the complex demands of large-scale AI operations.
The Gigawatt Challenge: Optimizing AI Energy Consumption
AI agents are central to DSX's operational efficiency, trained within the digital twin to optimize power consumption and reduce strain on both the AI factory and the broader power grid. Phaidra AI agents, for instance, continuously fine-tune power, cooling, and workloads, transforming the digital twin into a self-learning system that enhances grid flexibility and energy efficiency. Emerald AI further integrates its reference design to help utilities unlock underutilized grid capacity, a vital step in accommodating future AI growth.
DSX delivers on three key pillars: DSX Flex, DSX Boost, and DSX Exchange, all directly impacting AI energy consumption. DSX Flex enables dynamic grid collaboration, balancing energy demand with real-time grid conditions. DSX Boost applies NVIDIA’s power-optimization technologies to achieve Max-Q efficiency, delivering up to 30% higher GPU throughput within the same power envelope. DSX Exchange unifies IT and operational technology systems, providing a secure data and control fabric for comprehensive optimization.