Quick Facts

  • Envision deployed the world's first commercial 800V DC power system at its Galaxy AI campus in Ulanqab, China, announced November 2, 2026.
  • The system uses solid-state transformers to convert 10kV grid power directly to 800V DC in a single stage, achieving 98.5% conversion efficiency.
  • The Galaxy campus is planned at 120,000 square meters with 3 GW of capacity designed to support up to one million AI accelerators.

Chinese green technology company Envision has switched on what it calls the world's first commercial 800-volt DC power system inside a large-scale AI data center, with the system now powering live computing workloads at its Galaxy AI campus in Ulanqab, Inner Mongolia.

The deployment addresses a problem growing more urgent across the industry. Conventional data centers pass electricity through multiple conversion stages before it reaches a server rack, losing energy at each step. As AI cluster rack power climbs toward 350 to 500 kilowatts, that chain is becoming a bottleneck.

How the System Works

Envision's approach cuts that chain to a single conversion. Its solid-state transformers, built on silicon-carbide power electronics, step 10,000-volt AC grid power down directly to an 800-volt DC bus. The company brought a 2.5 MW SST model to a recent industry event, listing specs of a 10/13.8 kV AC input, 800V DC output, and 98.5% conversion efficiency. By bypassing intermediate AC-to-AC and step-down stages, the architecture also cuts equipment volume by half compared with legacy setups.

Three hardware layers make up the full system: solid-state transformers, battery energy storage, and DC voltage adjustment equipment for computing hardware. Coordinating them is Envision's Dubhe Controller, which runs the company's Dubhe Energy Foundation Model. The controller adjusts transformers and batteries in milliseconds to track abrupt GPU load swings and absorb them locally rather than passing instability upstream to the grid.

Battery storage serves two roles: second-scale power buffering for rapid load changes and minute-scale backup during grid interruptions. Testing began in August 2026 before the November commercial launch.

Scale of the Galaxy Campus

The Galaxy AI campus spans 120,000 square meters, roughly equivalent to 20 football pitches. At full buildout, the facility is planned for 3 GW of capacity and is designed to support up to one million AI accelerators while drawing more than 80% of its power from direct renewable energy connections.

Envision says the campus is designed to generate up to 10 times more compute output per square meter than conventional data centers. The company has also set a broader target under its Mission Gobi initiative: 5 GW of AI data center capacity by 2030.

Kevin Huang, senior vice president of Envision Energy and president of its energy storage and AI data center power systems division, said the deployment proves the concept at operational scale. "Our integrated 800V DC system is now powering live computing workloads, demonstrating an end-to-end architecture that connects renewable energy supply with computing demand," Huang said.

Industry Momentum Behind 800V DC

Envision is not moving alone. Nvidia has been openly advocating for 800V DC architectures, arguing that legacy 54V DC systems are nearing physical scaling limits for next-generation AI hardware. Nvidia, Google, and Microsoft are developing an 800V DC standard together through the Open Compute Project, with Nvidia planning to push the standard starting in 2027.

The urgency is driven by hardware roadmaps. Nvidia's upcoming Rubin Ultra and Feynman platforms are projected to require between 450 kilowatts and one megawatt per rack. Nvidia states that moving a data center backbone from 415 or 480V AC to 800V DC transmits more than 150% more power through the same copper, removing the need for roughly 200-kilogram copper busbars that a single legacy 1 MW rack requires.

Envision's commercial deployment puts a working reference point on the table ahead of that 2027 industry push. The company's claim to a world first remains self-reported, but no competing operator has announced a comparable commercial installation at this voltage and scale.

This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.