17:48Machine-translated from Azerbaijani by AIRead in Azerbaijani
Envision Builds 800 MW AI Data Center in China
Envision is building the 800 MW Galaxy artificial intelligence (AI) data center campus in Inner Mongolia. The project in Ulanqab is expected to be completed by the end of 2027. In scale, the campus will be comparable to the approximately 820 MW Stargate campus in the US. The first 120 MW section of the project was built in less than 1 year. According to estimates, this section cost approximately 20 percent as much as comparable projects in the US.
The campus is being built in 4 phases and is designed for 2 unnamed internet companies. Envision is not relying solely on the local grid for electricity. Wind farms, battery systems for energy storage, and dedicated power infrastructure are being built next to the facility. US projects have opted for gas turbines because of power grid constraints. Galaxy, however, will run primarily on wind energy. Envision has installed wind turbines with a total capacity of more than 100 GW to date. Galaxy will be served by wind farms located approximately 80 km apart. Their total capacity is 420 MW. These farms are expected to meet more than 60 percent of the first 2 phases’ 240 MW power demand. Wind turbines in the region operate at an average of 35 percent to 40 percent of their capacity. Therefore, battery storage capable of storing energy for 4 hours is also being added to the system. A new electrical system is being tested in Galaxy’s 5 MW hall. Solid-state transformers directly convert 10 kV alternating current into the 800 V direct current required by server racks. In conventional data centers, electricity must pass through several additional devices. Here, that equipment, including low-voltage transformers, uninterruptible power supplies, and rectifiers, is no longer needed. This could reduce the amount of equipment and energy losses. However, the efficiency figures achieved during testing have not been disclosed. At Galaxy, liquid- and air-cooled servers are connected to the same water loop. This approach eliminates the need for conventional central cooling units and makes it easier to manage the high heat generated by AI servers. However, the system’s efficient operation depends on the local climate and how effectively the water loop transfers heat outside. This power unit, currently rated at 5 MW, is planned to operate on a larger scale in subsequent phases. Galaxy is the first major project among the 3 GW of data centers that Envision plans to build in Inner Mongolia. In the nearby Chahar High-Tech Zone, 22 data center projects are under construction, employing approximately 30,000 construction workers. Alibaba, Huawei, and ByteDance are also expanding their infrastructure in the region. However, US export restrictions make it difficult for China to purchase Nvidia’s most powerful AI chips. This prevents the buildings under construction from operating at full computing capacity.
The campus is being built in 4 phases and is designed for 2 unnamed internet companies. Envision is not relying solely on the local grid for electricity. Wind farms, battery systems for energy storage, and dedicated power infrastructure are being built next to the facility. US projects have opted for gas turbines because of power grid constraints. Galaxy, however, will run primarily on wind energy. Envision has installed wind turbines with a total capacity of more than 100 GW to date. Galaxy will be served by wind farms located approximately 80 km apart. Their total capacity is 420 MW. These farms are expected to meet more than 60 percent of the first 2 phases’ 240 MW power demand. Wind turbines in the region operate at an average of 35 percent to 40 percent of their capacity. Therefore, battery storage capable of storing energy for 4 hours is also being added to the system. A new electrical system is being tested in Galaxy’s 5 MW hall. Solid-state transformers directly convert 10 kV alternating current into the 800 V direct current required by server racks. In conventional data centers, electricity must pass through several additional devices. Here, that equipment, including low-voltage transformers, uninterruptible power supplies, and rectifiers, is no longer needed. This could reduce the amount of equipment and energy losses. However, the efficiency figures achieved during testing have not been disclosed. At Galaxy, liquid- and air-cooled servers are connected to the same water loop. This approach eliminates the need for conventional central cooling units and makes it easier to manage the high heat generated by AI servers. However, the system’s efficient operation depends on the local climate and how effectively the water loop transfers heat outside. This power unit, currently rated at 5 MW, is planned to operate on a larger scale in subsequent phases. Galaxy is the first major project among the 3 GW of data centers that Envision plans to build in Inner Mongolia. In the nearby Chahar High-Tech Zone, 22 data center projects are under construction, employing approximately 30,000 construction workers. Alibaba, Huawei, and ByteDance are also expanding their infrastructure in the region. However, US export restrictions make it difficult for China to purchase Nvidia’s most powerful AI chips. This prevents the buildings under construction from operating at full computing capacity.