Lite-On Showcases 800 VDC Power and Liquid Cooling Solutions at OCP 2026

Preface

As artificial intelligence workloads push data centers toward higher power density, infrastructure providers face a growing challenge: delivering electricity efficiently while removing heat reliably and leaving room for future expansion. At the OCP Global Summit in San Jose, United States, from 10/12-10/15, power solutions provider Lite-On Technology (光寶科, 2301-TW) presented a portfolio built around 800 VDC distribution, backup power, and liquid cooling. Its exhibits included systems designed for the NVIDIA Vera Rubin platform, solutions based on NVIDIA MGX and the open OCP ORv3 architecture, and power-conversion modules spanning the path from facility supply to the processor. The central theme was Grid-to-Chip integration: coordinating power delivery, energy resilience, and thermal management as one infrastructure system. This article outlines the technologies shown and explains how they address the practical requirements of high-density AI and high-performance computing (HPC) environments.

Lazy bag

Lite-On’s OCP presentation connected 800 VDC distribution, backup power, and liquid cooling in a system-level approach to AI infrastructure. Highlights included an 800 VDC Power Shelf with a battery backup unit (BBU), a 2.6MW enterprise cooling distribution unit (CDU), and integrated power and cooling solutions for NVIDIA Vera Rubin NVL72. The company also presented ORv3-compatible power racks, capacitor backup, in-rack CDUs, and DC-DC conversion modules that carry power from the rack toward the GPU. Together, these designs target efficiency, reliability, cooling capacity, and flexibility as AI platforms demand more power.

Main Body

Lite-On’s exhibition at the OCP Global Summit focused on a shift in data-center design: AI infrastructure is increasingly judged not only by compute capacity, but also by how efficiently and reliably it can supply power and remove heat. The company presented an integrated portfolio centered on 800 VDC architecture, liquid cooling, and power conversion. Its solutions addressed the facility-to-rack and rack-to-chip stages, with products intended for NVIDIA platforms and open hardware architectures including NVIDIA MGX and OCP ORv3.

A key exhibit was an 800 VDC rack-mounted Power Shelf integrated with a high-efficiency battery backup unit (BBU). The design reflects Lite-On’s approach to power delivery for next-generation AI factories, where large and rapidly changing loads require both efficient distribution and resilience. Integrating backup power at the rack level can help address transient power needs and support continuity, while the 800 VDC architecture provides a framework for delivering power to high-density computing equipment.

Thermal management formed another major part of the presentation. Lite-On showed a 2.6MW enterprise cooling distribution unit developed by DCX Liquid Cooling Systems, a liquid-cooling technology company in which Lite-On has made a strategic investment. The CDU supports heat-exchange capacity of up to 2.6MW, a 3°C approach temperature, and an N+1 redundant pump design. These features are intended to meet the cooling demands of high-density AI and HPC environments while supporting system reliability. The redundancy design provides an additional level of resilience, while the stated approach temperature describes a relevant measure of heat-transfer performance.

Lite-On also presented a new-generation human-machine interface (HMI) and integrated software platform for the CDU. These tools provide real-time monitoring and visual management of critical system parameters. In a complex liquid-cooling installation, visibility into operating conditions can help teams understand system status and manage equipment more effectively. The software and interface therefore complement the physical cooling hardware by supporting operational oversight.

For the NVIDIA Vera Rubin NVL72 platform, Lite-On displayed an integrated power and cooling solution that combines a 95kW DC/DC rack-mounted power system with an L2L 280kW in-rack CDU. The solution is based on the NVIDIA MGX architecture and is intended to demonstrate the company’s ability to provide energy infrastructure for high-density systems. By combining 800 VDC power delivery, energy backup, and liquid cooling, the design aims to reduce power-distribution losses and heat generation, improve overall energy efficiency and power density, and simplify the power architecture. A simplified arrangement may also free up rack space for computing equipment and support dense GPU deployment.

Lite-On Cloud Infrastructure Systems and Platform Business Group General Manager Chang Kun-Sung (張坤松) said that competition among data centers in the AI era is no longer only about the scale of computing capacity. He emphasized the importance of turning each watt of electricity into useful computing output. As AI platform power density continues to rise, data centers need to balance energy efficiency, cooling capability, and flexibility for expansion. In his view, efficient power and cooling architectures will be a key foundation for scaling AI factories.

Alongside its NVIDIA-focused systems, Lite-On presented an AI power-rack solution compatible with the OCP ORv3 open architecture. The configuration combines a 180kW rack-mounted Power Shelf with a 108kW rack-mounted capacitor backup unit (CBU). Centralized power delivery and a supercapacitor-based backup structure are designed to respond to power demands caused by sudden changes in AI workloads. By supporting these fluctuations, the system aims to improve power stability and overall reliability.

ORv3’s system architecture is optimized for liquid-cooling compatibility. In keeping with this approach, Lite-On displayed 120kW and 220kW in-rack CDUs as key thermal-management modules. These units can be integrated with newer cooling designs such as Direct-to-Chip cooling. The modular approach is intended to support higher-density GPUs and the integration of liquid-cooling and energy systems, addressing the combined power and thermal requirements of next-generation AI and HPC platforms. Lite-On’s broader proposition is that coordinated power, backup, and cooling systems can help customers build infrastructure that is efficient, reliable, and expandable.

The exhibits also addressed power conversion along the route from the rack to the GPU. As AI chip power consumption continues to increase, moving electricity efficiently from a rack to processors has become a central infrastructure concern. Lite-On presented a range of DC-DC conversion technologies, including 800V Power Brick, 50V Power Brick, IBC modules, and VRMs. Together, these components illustrate a chain that starts with 800 VDC distribution, continues through server-level conversion, and ends with power regulation at the chip.

According to Lite-On, the 800V DC/DC Power Brick can be used with a high-voltage direct-current (HVDC) power distribution board, or Power Distribution Board (PDB), for subsequent distribution and voltage reduction. The design can be adapted to different power requirements through the parallel use of multiple units. Active current sharing is intended to improve the stability and reliability of the power supply when units operate together.

The IBC, or Intermediate Bus Converter, performs an intermediate voltage-conversion role. It can improve overall power efficiency and reduce line losses, while helping optimize circuit-board space and system cost. Lite-On also described the IBC as supporting power stability in systems with multiple output paths. At the final stage, the VRM, or Voltage Regulator Module, provides the GPU with a precise and stable operating voltage. This completes the last stage of power delivery from the rack to the chip.

Viewed as a whole, Lite-On’s OCP portfolio links facility-level distribution with rack power, backup systems, liquid cooling, and chip-level voltage regulation. The company’s approach reflects the growing need to manage energy as an integrated system rather than as a collection of separate components. Higher power density increases the importance of reducing conversion and distribution losses, while also requiring cooling systems capable of handling substantial heat loads. Backup systems and monitoring tools add further support for stability and operational management.

As AI infrastructure continues to move toward higher power density and energy efficiency, Lite-On is expanding its capabilities across power delivery and thermal management. Its Grid-to-Chip strategy brings these capabilities together in a system-level energy-management approach, with the aim of providing a complete infrastructure foundation for future AI platforms.

Key Insights Table