Chinese AI campus runs live workloads on 800-volt DC power

Envision says its Ulanqab data center is the first commercial site to feed computing hardware from an integrated 800 V DC system; the claim is unverified.

Envision Energy says it has put the first commercial, large-scale integrated 800 V DC power system into service at its Galaxy AI data center campus in Ulanqab, China. Kevin Huang, who leads the company's energy storage and AI data center power business, said it is powering live computing workloads. The campus began operating in August 2026.

The design links solid-state transformers, battery storage and equipment that adjusts DC voltage for computing hardware. The transformers step power down from 10 kV medium voltage to an 800 V DC bus. Batteries add secondary support and minute-scale backup. Envision's Dubhe Controller, driven by its Dubhe energy AI model, coordinates the two in milliseconds to absorb sudden GPU load swings. Typical data centers run AC right up to the rack before converting it. Keeping higher-voltage DC deeper into the facility should allow more compute per rack. In July, Envision showed a 2.5 MW transformer with 98.5% conversion efficiency in Munich.

Why it matters: power conversion and floor space are two of the quieter costs of AI compute. Fewer conversion steps waste less electricity, and denser racks need less building per unit of computing. Pairing that with a campus expected to draw over 80% of its power directly from renewables points toward cheaper, cleaner compute, which feeds access to knowledge tools.

Watch for: Mission Gobi, Envision's plan for 5 GW of green AI data center capacity in deserts by 2030. Caveats: the 'world first' claim, efficiency figures and renewable share come from Envision and are not independently verified. The 3 GW capacity and 80% renewables share are planned or expected figures, not confirmed operating results.

Source: pv magazine