Flex has published reference designs for the NVIDIA Omniverse DSX Blueprint aimed at speeding “giga-scale” AI factory builds with prefabricated, modular blocks that package power, high-density IT racks, and cooling into factory-integrated systems. The company says the approach can cut deployment timelines by up to 30% versus traditional construction.
The designs sit under Flex’s AI Infrastructure Platform and are built around hybrid power architectures that support phased migration from conventional AC environments to 800 VDC power architectures. For data center teams trying to bring AI capacity online quickly, the practical implication is straightforward: more of the integration work shifts off the jobsite and into repeatable factory builds, which can reduce commissioning complexity and schedule risk—if the site can support the chosen architecture.
What’s in the reference designs
Flex’s reference designs include an “800 VDC Power Rack” developed in collaboration with NVIDIA. Flex describes it as a disaggregated architecture that uses Flex’s power shelf for the NVIDIA Vera Rubin platform. The concept is to move power components outside the IT rack to free up rack space for compute, with Flex citing higher GPU density, improved communication, and greater performance per rack as the intent.
On the thermal side, the designs incorporate advanced liquid cooling with secondary fluid networks and centralized cooling distribution units (CDUs) to remove heat for power-intensive AI workloads. The modular build also includes integrated IT racks described as high-density and liquid-cooled, housing AI servers plus integrated networking and high-speed interconnects for low-latency, high-bandwidth communication.
For facility power distribution, Flex lists critical power infrastructure elements including dedicated high-capacity power feeds, redundant busway systems with rack-level tap-offs, and integrated power distribution cabinets and units.
Manufacturing and deployment context
Flex says it is already building prefabricated modular solutions for hyperscalers, including at a 400,000-square-foot Dallas facility purpose-built for data center infrastructure. The company also points to more than 18 million square feet across 35+ locations in the Americas, including Austin, Texas, and Guadalajara, Mexico, as part of its manufacturing footprint for serving US AI infrastructure demand.
“By engineering the full infrastructure stack as a unified system, we help customers overcome the power, heat and scale challenges of the AI era and bring capacity online faster,” said Michael Hartung, president and chief commercial officer at Flex.
Source: Flex
