Envision Launches Commercial 800V DC Power Architecture for AI Data Centers

Envision Launches Commercial 800V DC Power Architecture for AI Data Centers

Envision has placed an 800V DC power architecture into commercial service at its Galaxy AI campus in Ulanqab, Inner Mongolia, describing the installation as the first commercial deployment of its kind. The company says trials of its Dubhe controller and associated energy model started at the site in August 2026, though it has not publicly released independently verified performance figures.

A shorter route from the grid to the rack

Article image

Most data centers move incoming electricity through multiple conversion stages before it reaches servers and accelerators. Each additional stage brings more equipment and more opportunity for losses. Envision’s approach collapses that chain: a solid-state transformer takes 10-kilovolt medium-voltage alternating current and turns it directly into an 800-volt direct-current distribution line.

According to the company, that design trims conversion steps, raises electrical efficiency and sharply reduces the floor space needed for power equipment. Higher-voltage distribution can also cut the volume of copper required for cabling, a material consideration as computing installations grow larger and denser.

Batteries and a controller built for GPU swings

The architecture pairs the transformer with battery storage. Envision says the batteries respond quickly when processor demand shifts and provide temporary backup during unexpected interruptions, offering immediate support over seconds and backup capability across periods measured in minutes.

A controller coordinates the transformer and the batteries on a millisecond timescale. That coordination is meant to absorb the abrupt consumption changes that GPUs can produce, which place extra strain on the electrical infrastructure feeding tightly packed computing equipment. Envision also says the arrangement can support higher-density server racks, letting operators fit more computing capacity into a given floor area.

Plans reach well beyond the first site

The Galaxy facility is expected to expand to 3 gigawatts, and Envision says the eventual capacity could serve one million AI accelerators. The company also plans to introduce the technology through Mission Gobi, an initiative targeting 5 gigawatts of renewable-powered computing capacity by 2030, concentrated in desert locations where renewable generation and energy storage are expected to support large computing developments.

The system is designed to work alongside renewable electricity sources, with its batteries intended to balance supply against changing computing requirements.

What remains unproven

Envision has not provided independently verified measurements quantifying efficiency improvements, installation costs or savings relative to conventional power systems. Nor does the announcement establish how the architecture performs under sustained full-load conditions or across facilities with substantially different hardware configurations.

The deployment is therefore a commercial milestone for Envision. The size of the operational advantages it claims still awaits detailed independent evidence.

Leave a comment

Your email address will not be published. Required fields are marked *