Commercial deployment in Inner Mongolia
Chinese energy company Envision said it has commercially deployed an integrated 800V direct-current (DC) power system at an artificial intelligence (AI) data center in Inner Mongolia.
According to technology publication TechRadar on November 9, with dates given in local time, the system is supplying electricity for actual AI computing at the Galaxy AI Campus in Ulanqab. Envision claimed it was the world's first commercial deployment of this power architecture at a large-scale AI data center.
Envision has been testing its Dubhe controller and energy model at the site since August 2026. However, it has not yet disclosed independently verified figures demonstrating the system's performance.
Power conversion and integrated battery storage
The key is the power conversion architecture. Conventional data centers go through multiple conversion stages before electricity reaches servers and AI accelerators. This can increase the amount of associated equipment and cause power losses. Envision, by contrast, uses solid-state transformers to convert 10kV medium-voltage alternating current (AC) into an 800V DC distribution network.
The company said this architecture could improve efficiency by reducing power conversion stages and significantly reduce the space occupied by power equipment inside a data center. It also claims that raising the distribution voltage can reduce the amount of copper needed for cables, easing material requirements for large-scale computing facilities.
The system also integrates battery storage. The batteries respond rapidly to sudden changes in processor power demand and provide temporary backup power during unexpected outages. Envision said the batteries support power reinforcement lasting seconds and backup lasting minutes, while the Dubhe controller controls the transformers and batteries on a millisecond timescale to respond to rapid changes in graphics processing unit (GPU) power demand.
Stable electricity supply is important because AI computing power demand can fluctuate substantially with the workload. Envision said its system supports high-density server racks, allowing more computing equipment to be installed in the same space.
Expansion plans and verification still needed
The company also outlined expansion plans. The Galaxy campus ultimately targets a capacity of 3GW and is designed to support 1,000,000 AI accelerators upon completion. Envision also plans to apply the technology to Mission Gobi, a project to build 5GW of renewable energy-powered AI data center capacity in desert regions by 2030.
However, independent measurements demonstrating the extent of efficiency improvements, installation costs and savings compared with conventional power systems have not yet been disclosed. It has also not been established whether the system can deliver the same performance under prolonged maximum loads or at data centers with different hardware configurations. The deployment represents a commercial application of an 800V DC power system, but its actual operational benefits require further verification.
Original Korean article: digitaltoday.co.kr
English brief by Korea Wind Intelligence, koreawindintel.com
