Beating AI News Flash: U.S. semiconductor research firm SemiAnalysis conducted an on-site visit to the "Envision Galaxy Base" in Ulanqab, Inner Mongolia, and found that the construction cost of this Chinese AI data center park may be only one-fifth that of comparable U.S. projects.
The first 120MW of capacity at the Galaxy Base was built and delivered in less than a year. On August 6 this year, the Phase I infrastructure was completed, and Envision announced the official commissioning of the base. After nearly three months of mechanical and electrical commissioning and computing equipment deployment, the approximately 200MW Phase I project was fully delivered to customers by the end of October. The entire park plans to reach 800MW of IT power supply capacity by the end of 2027, comparable to the approximately 820MW planned for the core Stargate park in Abilene, USA. The Abilene core park is expected to reach that scale by the end of 2026.
OpenAI's Stargate in the United States is a data center construction plan spanning multiple states. The initial goal was to deploy 10GW of AI infrastructure capacity in the United States by 2029. In April this year, OpenAI announced it had secured more than 10GW of planned capacity ahead of schedule, but these data centers have not all been completed and put into operation.
The most distinctive aspect of the Galaxy Base is that its builder, Envision Technology, was originally a wind power and energy storage company. It directly integrates power generation, energy storage, power supply, and data center construction.
Envision built two wind farms with a total installed capacity of 420MW, along with battery storage, aiming to make more than 60% of the electricity used by the first two phases of the data center come from its own wind power. By contrast, the U.S. Stargate Abilene park mainly relies on the power grid and on-site gas-fired generation facilities for power supply.
There have also been breakthroughs in power supply technology. Envision has actually operated an 800V high-voltage direct current system in a 5MW machine room, converting 10kV alternating current directly into the direct current needed by server racks, eliminating the multi-stage conversion equipment in traditional power supply architectures. This is also the next-generation high-density AI rack power supply direction that Nvidia is promoting.
The park also adopts a cooling design that does not require central chillers, allowing liquid-cooled and air-cooled servers to share one water circulation system, further reducing equipment investment.
