A new kind of power system now runs live artificial intelligence work at a huge data center campus in China. The Envision 800V power system delivers power end to end, from grid input to computing loads at Galaxy in Ulanqab. AAP carries the company announcement, and Envision calls the deployment the first commercial use of its kind. The claim comes from the company itself, and readers should still weigh it as a company claim, not as an independent test result.
The 800V system combines solid state transformers, battery storage, and equipment that adjusts direct current voltage for computing hardware. Those transformers convert 10 kV alternating current into an 800V direct current bus. Batteries add power support on a scale of seconds, and backup on a scale of minutes. A controller called Dubhe coordinates the transformers and batteries in real time, and it matches power supply with computing demand as loads swing. Financing for that buildout is growing too, while a recent hardware loan facility shows how much debt now funds AI hardware.
AI work creates sharp swings, because graphics processors draw heavily, and the 800V system must answer those bursts. A power system that lags behind those bursts can still waste energy, or strain the hardware it feeds. Envision, however, says millisecond level coordination between transformers and storage absorbs those swings close to the load. The goal is still steadier power, less waste, and a smaller equipment footprint inside the building. Less cabling and less copper also follow from the simpler design, according to the company.

The Galaxy campus, however, began operations in August 2026, and its planned capacity reaches 3GW at full build out. At that scale, Envision says the campus could still support up to one million AI accelerators in parallel. Envision, however, plans more than 80 percent of its power to come from direct renewable electricity connections. That figure is a plan, not a present day reading. The present milestone is narrower: live workloads now run on the new power system.
Energy use has become a question, because large models draw power, and the Envision 800V power system enters that debate. A European network of AI factories shows how fast new capacity is rising. Power grids and land limit where new centers can grow. Thus a system that uses each watt better can ease those limits. The live site offers a reference point, not only a laboratory demo.
Renewable power adds a second challenge, with wind and solar output rising and falling with weather. Batteries help bridge short gaps, and smart controls help match demand to supply. Envision pairs its power system with storage, and it plans heavy use of direct renewable connections at Galaxy. Readers should look for those readings, and for uptime data, before judging the system.

Scale gives the Envision 800V power system story its weight, and scale also gives it its risk. On the other hand one million accelerators would form one of the largest AI sites on earth, if the plan ever arrives. A 3GW campus needs grid capacity and supply chains to match its ambition. Mission Gobi, the Envision initiative, targets 5GW of green AI data center capacity in desert regions by 2030. That target will need years of building, funding, and permits to reach.
Power companies face a similar choice in every new project, with demand growing faster than many grids can follow. A data center debt deal shows how much money flows into AI sites. The Envision 800V power system can lower running costs with efficient design, and it can also lower strain on local grids. Communities near large sites will ask hard questions about water and grid prices.
Power is the quiet partner of every bright AI demo, and a steadier the 800V system lets good work run longer and cleaner. The Galaxy deployment is a first step, not a final verdict, and its true value will show in months of steady running. Instead if it holds up, others will follow. If it falters, the industry will learn from that too. Either way, the lights stay on only when power earns its place. The Galaxy deployment will be judged on months of steady running at full load, and the results over time will matter more than the launch claims in daily use.
