As competition to build AI data centers intensifies, power supply and efficiency are emerging as key variables for infrastructure expansion. With the proliferation of generative and inference AI increasing power demand per server rack, it is becoming crucial for the semiconductor industry to respond with technologies that enhance power conversion and distribution efficiency, in addition to computational performance. ON Semi is responding to this demand by expanding support for power solutions within the NVIDIA MGX ecosystem.

Onsemi announced on the 4th that it is expanding the scope of its power solution supply to support the construction of NVIDIA MGX-based AI data centers and accelerated computing platforms. This support covers the entire supply chain that constitutes the AI server power infrastructure, including power supply units (PSUs), battery backup units (BBUs), and power distribution boards (PDBs).

NVIDIA MGX is a modular architecture designed to enable server manufacturers and system builders to configure AI and high-performance computing systems by combining CPUs, GPUs, networks, storage, and more. As AI servers expand from individual equipment units to rack units and data center units, standardized platforms are being utilized as a means to reduce design complexity and deployment time.

Power demand for AI infrastructure is rapidly increasing due to the proliferation of large-scale AI models and inference workloads. The industry is raising the possibility that power demand for next-generation AI racks could rise to over 1 MW, and it is pointed out that existing low-voltage power structures have limitations in reducing power loss, wiring complexity, and cooling burdens.

ON Semi provides power FETs, multiphase power solutions, SiC JFETs, and GaN-based products for the power conversion stage within the MGX ecosystem. The company supports the improvement of conversion and distribution efficiency in AI server power grids through its power semiconductor product family, which includes silicon, silicon carbide, and gallium nitride technologies.

This announcement also included support for 800VDC power architectures. High-voltage power structures such as 800VDC are being discussed as an alternative to increase power density in AI racks and reduce losses during power distribution. However, the actual scope of application and the speed of adoption may vary depending on data center design methods, power infrastructure investment, and the level of cooperation in the system supply chain.

Sudhir Gopalswami, President of ON Semi Intelligent Sensing and Analog & Mixed Signal Group, explained that the demand for power supply, cooling, and infrastructure efficiency is rapidly growing due to the adoption of AI and the increasing demand for inference, and that high-voltage architectures such as 800VDC are a necessary technological element for expanding AI infrastructure.

Power efficiency is directly linked to the operating costs of AI data centers. As GPU clusters become larger, power conversion losses and cooling burdens increase; consequently, competition in AI infrastructure is expanding beyond computing chips to include power semiconductors and power distribution technologies.

As the construction of next-generation AI factories increases, the importance of data center power architecture is expected to grow. ON Semi's expanded support for the MGX ecosystem demonstrates that the role of power semiconductor companies in the AI infrastructure market is expanding.