Mitsubishi Electric announced Chip-to-Grid DSX reference designs for U.S. data centers deploying NVIDIA Vera Rubin NVL72 and future AI infrastructure. Released through Mitsubishi Electric Power Products, the designs coordinate grid connection, on-site power, chip-level power supply and cooling, with support for data centers scaling toward gigawatt-class electrical demand.
The architecture combines power transmission and distribution equipment with data center power supply and cooling systems. It supports both conventional 415/480VAC distribution and 800VDC distribution inside the data center, providing configurations for current infrastructure as well as higher-voltage DC architectures associated with dense AI computing installations.
Sites using the reference designs can incorporate on-site power generation and battery energy storage systems. During normal operation, these resources are intended to reduce reliance on the utility grid; during a grid outage, the facility can switch to autonomous operation through its own microgrid. The configuration is intended to maintain power stability while supporting the electrical requirements of large AI data centers.
Cooling is based on a dual-loop architecture combining direct liquid cooling for chips with air cooling for other equipment. The arrangement targets high-density AI loads and integrates facility cooling with the broader power system rather than treating thermal management as a separate infrastructure layer.
The reference designs are intended to support NVIDIA’s Vera Rubin NVL72 rack-scale AI system, which integrates 72 Rubin GPUs and 36 Vera CPUs, as well as later NVIDIA AI infrastructure. Mitsubishi Electric plans to use the designs as a basis for developing integrated data center offerings that combine power generation, energy storage, transmission and distribution, chip power delivery and cooling.
