Exascale Labs and Compal Plan U.S. GPU Power Validation Platform

Exascale Labs and Compal Electronics have signed a non-binding letter of intent to develop a U.S.-based validation platform for GPU infrastructure. The project focuses on testing 800-volt direct current (VDC) architecture and solid-state transformer (SST) technology to address the power demands of AI data centers.

"The goal is to validate native 800 VDC across a complete data center system," said Dr. Hoansoo Lee, Chief Executive Officer of Exascale, in a statement regarding the partnership.

Technical Scope and Infrastructure

The proposed platform aims to move beyond component-level testing by integrating power distribution, modular data center facilities, and cooling systems into a unified environment. By utilizing 800 VDC, the companies aim to reduce energy conversion losses, which are inefficiencies in current data center designs that rely on multiple stages of AC-to-DC conversion.

"The integration of GPU servers and cooling systems into the development process would allow validation of complete infrastructure under real operating conditions," said Alan Chang, VP and GM of ISBG of Compal.

Commercial Structure and Potential Acquisition

The partnership is currently in the initial non-binding stage. Under the terms of the letter of intent, Compal intends to acquire the portion of the platform developed by Exascale, including the SST systems, power distribution equipment, and associated software licenses, following successful completion and acceptance testing.

The project remains subject to the execution of a definitive agreement. According to a company filing, there is no assurance that the parties will reach a final deal or that the validation platform will reach full operational status.

Why 800 VDC Matters for AI

As GPU clusters grow in density, traditional power delivery methods face thermal and efficiency constraints. Standard data center power architectures involve voltage stepping, which generates heat and requires additional cooling infrastructure. By validating a native 800 VDC system, the companies are testing a mechanism designed to deliver power directly to the rack-level, potentially lowering the total cost of ownership for AI compute environments.