LITEON Targets AI Factory Efficiency with New 800 VDC Power Architecture

The company’s latest exhibition focuses on a comprehensive Grid-to-Chip energy strategy, bridging the gap between high-voltage rack distribution and chip-level power delivery. A centerpiece of the display is an 800 VDC Power Shelf integrated with a battery backup unit, designed to minimize transmission losses while maximizing space for compute-heavy deployments. To manage the resulting thermal load, LITEON introduced a 2.6MW Enterprise Cooling Distribution Unit, developed alongside DCX Liquid Cooling Systems. This unit features an N+1 redundant pump design and an approach temperature of 3°C, specifically engineered to handle the extreme heat generated by modern high-density AI and HPC hardware.

For systems utilizing NVIDIA MGX and OCP ORv3 architectures, LITEON is showcasing integrated solutions that combine power conversion with advanced thermal management. By incorporating 95kW DC/DC power shelves and liquid-to-liquid in-rack cooling, the firm aims to simplify the infrastructure footprint of GPU clusters. John Chang, General Manager of LITEON’s Cloud Infrastructure Platform, emphasized that the competitiveness of AI data centers now relies on maximizing tokens per watt. To meet this demand, the company is rolling out a full range of power components, including 800V power bricks, intermediate bus converters, and voltage regulator modules. These components are designed to stabilize power delivery from the rack down to the individual GPU, providing the granular control required for the next generation of AI factories.