Notícias

Voltar à página de notícias mais recentes
FOR IMMEDIATE RELEASE No. 3941

Mitsubishi Electric Unveils NVIDIA AI Infrastructure-compatible Reference Designs for Next-generation AI Data Centers

Optimizing power supply and cooling from chip-to-grid to meet growing AI demands

TOKYO, September 24, 2026 – Mitsubishi Electric Corporation (TOKYO: 6503) announced today that its U.S. subsidiary Mitsubishi Electric Power Products, Inc. released Chip-to-Grid1 DSX2 Reference Designs3 to support adoption of NVIDIA Corporation’s Vera Rubin NVL724 and future next-generation NVIDIA AI infrastructure in U.S. data centers. Going forward, Mitsubishi Electric will use the reference designs to develop solutions for next-generation AI data centers.

 

Mitsubishi Electric’s Chip-to-Grid DSX Reference Designs encompass and coordinate systems from grid connection to chip power supply, and direct-rack to facility-level cooling. This comprehensive solution integrates Mitsubishi Electric’s proven power transmission and distribution equipment with next-generation power supply that incorporates cutting-edge technologies as well as cooling systems. The designs also offer reliable scalability to support future gigawatt-class data centers.5

Specifically, data center sites will be equipped with on-site power generation and battery energy storage systems (BESSs). During normal operation, this solution will minimize a site’s reliance on the grid, and in the event of a grid failure, the facility will switch to autonomous operation on its own microgrid, stabilizing power and operation. In addition, the designs support both conventional 415/480VAC and 800VDC power distribution within the data center. Furthermore, by combining with a dual-loop cooling system6 suitable for high-density AI loads, the solution maximizes power efficiency.

 

Mitsubishi Electric’s solution, which supports the optimization of data center power operations, is one of the company’s key strategic initiatives. In recent years, the scope of AI utilization has expanded from generative AI, which has been widely adopted, to agent AI and physical AI. Consequently, data centers are expected to expand rapidly as a critical component of societal infrastructure. However, they must balance this demand with high power efficiency, reliability and scalability. Data center operators face major challenges in securing power, managing thermal conditions, and ensuring stable operations to meet the increasing growth of AI.

 

Going forward, Mitsubishi Electric will leverage its new reference designs to address data center challenges and expand its business in the global market by offering high-value solutions that integrate with the company’s diverse range of data center systems. Through this initiative, Mitsubishi Electric will support AI infrastructure essential to future societies, thereby contributing to its own sustainable growth.


  • 1

    Concept for integrated optimization of everything from power delivery to chips such as Graphics Processing Units (GPUs) to power transmission, distribution, reception, storage, and cooling.

  • 2

    A platform and framework for integrating and streamlining the design, simulation, and operation of AI data centers.

  • 3

    Standard configurations and design models that have been verified and organized in advance to ensure reliability and scalability.

  • 4

    NVIDIA’s next-generation rack-scale AI supercomputing system, integrating 72 Rubin GPUs and 36 Vera CPUs.

  • 5

    Large-scale data centers that consume gigawatt-class power.

  • 6

    Cooling architecture that combines direct liquid cooling to directly cool the chip and air cooling to cool other parts.

Inquiry