ROHM has presented a high-efficiency power supply solution for realizing gigawatt-class AI factories, heralding a new paradigm in global data center design.

ROHM has published a white paper outlining its 800VDC architecture-based power solution for next-generation AI data centers.

This document, part of a collaborative project with NVIDIA announced in June 2025, presents a highly efficient power supply method for realizing a gigawatt-scale AI factory.

As power consumption per rack in AI data centers rapidly increases, the existing 48V/12V DC method has reached its limits.

ROHM emphasizes that the 800VDC architecture can dramatically improve efficiency, power density, and sustainability.

In particular, the PSU that converts AC power to DC power runs in an independent power rack outside the IT rack, and high-efficiency topology design is key.

ROHM offers a wide range of power devices based on silicon (Si), silicon carbide (SiC), and gallium nitride (GaN), and also possesses analog IC technology to optimize these devices.

The EcoSiC™ series contributes to high power density with low ON resistance and top-side heat dissipation modules, while the EcoGaN™ series realizes high-frequency control and stable gate drive through Nano Pulse Control™ technology.

ROHM is working with industry leaders such as NVIDIA and Delta Electronics to commercialize the 800VDC architecture.

In particular, we signed a strategic partnership with Delta Electronics in 2022 to develop a power system.We are jointly developing a power device for the temple.

ROHM is leading the way in building an energy-efficient digital society through wide-bandgap semiconductor technology.

The key takeaways from the white paper are that the 800VDC architecture overcomes the limitations of traditional DC approaches to meet the power demands of AI data centers, with AC-DC conversion performed outside the IT rack and DC-DC conversion performed inside the rack for high-density GPU configurations.

It also includes content on achieving high efficiency, low noise, and miniaturization through a combination of SiC·GaN devices and analog ICs, and building a sustainable AI infrastructure through industry collaboration.