Envision Builds 800MW AI Data Center Campus in Inner Mongolia With On-Site Wind Power and 800V DC

Envision is developing a large-scale artificial intelligence data center campus in Ulanqab, Inner Mongolia, that is designed to reach up to 800MW of IT capacity by the end of 2027, according to a report by SemiAnalysis based on a site visit in September. The project, named Galaxy, is being built in four phases and will serve two unnamed internet companies, marking one of China's most ambitious efforts to pair AI computing infrastructure with dedicated renewable power generation.

The first 120MW facility was completed in under a year, and researchers estimate its cost at roughly one-fifth that of a comparable U.S. project. Once fully built out, Galaxy's planned 800MW of IT capacity would approach the approximately 820MW of critical IT capacity at the Stargate campus in Abilene, Texas, underscoring the scale of China's push to expand domestic AI infrastructure despite ongoing restrictions on access to advanced accelerators.

Rather than relying entirely on the local grid, Envision is building wind farms, battery storage and dedicated power infrastructure alongside the data center. The company, which has installed more than 100GW of wind turbines globally, aims for its wind farms to supply more than 60% of the power required by Galaxy's first two phases. The wind farms, located about 80 kilometers away with a combined installed capacity of 420MW, will support phases with an expected load of 240MW. Because turbines rarely operate at full power, Envision is pairing them with battery facilities capable of storing four hours of energy; wind capacity factors in Ulanqab range between 35% and 40%, according to the report.

The power strategy stands in contrast to the U.S. market, where some data center builders are struggling to secure electrical equipment. Turbine shortages have pushed delivery timelines to as late as 2030, and SpaceX has said it plans to manufacture its own turbine blades for gas-powered AI facilities. Envision's approach of deploying wind generation and battery storage next to its servers offers a potential template for bypassing grid interconnection bottlenecks.

SemiAnalysis researchers visited a 5MW server hall where solid-state transformers convert 10kV AC power directly into 800V DC for server racks. "In a 5MW hall, we saw the 800V system running," the researchers wrote, explaining that the solid-state transformer "converts the grid's 10kV AC directly into 800V DC for the racks." The architecture eliminates several intermediate components used in conventional data center power distribution, including low-voltage transformers, uninterruptible power supplies and rectifiers. Fewer conversion stages can reduce equipment requirements and energy losses, though SemiAnalysis did not report measured efficiency gains from the Galaxy pilot. The 5MW facility represents only a fraction of the campus, with Envision planning to scale the deployment in later construction phases. The approach aligns with Nvidia's push toward 800V DC distribution for future high-density AI systems.

Galaxy also employs a cooling design in which liquid-cooled and air-cooled servers share a single water loop, eliminating the need for traditional central chillers. Removing central chillers can simplify infrastructure, but the design depends on local climate and the facility's ability to reject heat through the water loop. China has tested other unconventional cooling approaches, including a wind-powered underwater data center near Shanghai that uses surrounding seawater to help cool servers.

Galaxy is the first major project in Envision's broader data center plans in Inner Mongolia, which SemiAnalysis says total 3GW. The rest of the campus remains under construction. A local report cited by the researchers said nearly 30,000 construction workers are spread across 22 data center projects in the expanding Chahar high-tech zone, where companies including Alibaba, Huawei and ByteDance are expanding their infrastructure. Local electricity prices are relatively low, land is abundant and wind resources are strong. But access to power and data center buildings does not by itself translate into AI compute capacity: China still faces a separate constraint in obtaining advanced accelerators, particularly as U.S. export restrictions have limited its ability to purchase Nvidia's most powerful chips.