China Launches World's First 800V DC Power System for AI Data Center in Inner Mongolia

Envision Energy Deploys Groundbreaking Power Architecture at Galaxy AI Campus

Envision Energy has put the world's first commercial 800V DC power system into operation at its Galaxy AI campus in Ulanqab, Inner Mongolia, China1

1

2

2

. The China AI data center began testing the Dubhe Controller and its supporting Dubhe Energy Foundation Model in August 2024, marking a departure from conventional power distribution methods that rely on multiple conversion stages2

2

. This AI data center deployment addresses a critical challenge as global data center power demand could surge by 160 percent by 2030, with facilities potentially requiring 68 gigawatts of capacity worldwide by 20273

3

.

Direct Conversion Eliminates Traditional Power Infrastructure Bottlenecks

The 800V DC power system uses solid-state transformers to convert electricity directly from 10kV medium-voltage AC to an 800V DC bus, eliminating intermediate components found in conventional data centers1

1

. Traditional facilities require low-voltage transformers, uninterruptible power supplies, and rectifiers—all of which the new architecture removes1

1

. Kevin Huang, Envision Energy's senior vice president, explained that AI data centers need power systems coordinating renewable energy, battery storage, and the grid to meet computing demand3

3

. The high-efficiency power infrastructure reduces equipment requirements and energy losses while supporting higher computing density within the same physical footprint2

2

4

4

.

Battery Storage Responds to AI Workload Power Demands in Milliseconds

The Inner Mongolia data center integrates battery storage that provides both second-scale power support and minute-scale backup capability2

2

. The Dubhe Controller coordinates solid-state transformers and batteries with millisecond-level precision to absorb GPU load fluctuations locally2

2

3

3

. This matters because AI workload power demands can cause substantial swings in electricity consumption, placing additional pressure on electrical infrastructure supplying densely packed computing equipment4

4

. The batteries respond to sudden changes in power demand within seconds while providing backup power for several minutes during interruptions3

3

.

Galaxy AI Campus Scales to 800MW with Renewable-Powered Computing Capacity

Envision built the first 120 MW of the Galaxy AI campus in less than a year at one-fifth the cost of comparable US projects1

1

. The facility is planned to provide up to 800MW of IT capacity by the end of 2027, approaching the approximately 820MW of critical IT capacity at the Stargate project in Abilene, Texas1

1

. Current plans show the campus scaling to 3GW capacity, eventually supporting up to one million AI accelerators operating simultaneously2

2

3

3

. Envision has installed wind farms with a combined capacity of 420MW located about 80 kilometers apart, designed to provide over 60 percent of electricity needed by the campus's first two phases1

1

. The company aims to source more than 80 percent of the campus's electricity through direct renewable energy connections in the future3

3

.

Mission Gobi Initiative Targets 5GW of Desert Data Center Capacity

Envision plans to expand the 800V DC technology through its Mission Gobi initiative, targeting 5GW of green AI data center capacity in desert regions by 20302

2

3

3

4

4

. The programme focuses on desert locations where renewable generation and energy storage can support large computing developments4

4

. The surrounding Chahar High-tech Zone in Ulanqab is expanding rapidly, with nearly 30,000 construction workers spread across 22 data center projects1

1

. Major technology companies including Alibaba, Huawei, and ByteDance are expanding infrastructure in the region, which benefits from relatively cheap electricity, available land, and abundant wind resources1

1

. The Galaxy facility also uses hybrid liquid-air cooling where liquid-cooled and air-cooled servers share a single water loop, eliminating the need for conventional central chiller plants1

1

. This power efficiency advancement aligns with Nvidia's push for 800V DC distribution in future high-density AI systems1

1

.