At a glance
- The campus opened in August, and Envision has not disclosed the size of the 800VDC system.
- A conventional data center takes AC from the grid, converts it to DC for the site's battery or UPS, back to AC for the IT racks and to DC again inside the servers.
- For the firms underwriting these campuses, the claim being tested is power per shell.
Envision says live compute workloads are running on an 800VDC power system at its Galaxy AI data center in Ulanqab, Inner Mongolia, a deployment the company calls the world's first. The campus opened in August, and Envision has not disclosed the size of the 800VDC system.
A conventional data center takes AC from the grid, converts it to DC for the site's battery or UPS, back to AC for the IT racks and to DC again inside the servers. Envision's design makes the AC-to-DC conversion once, at the building. Solid-state transformers take 10kV medium-voltage AC from the grid to an 800VDC busbar that feeds the hardware, with battery storage and equipment that adjusts DC voltage for computing hardware alongside it. A controller called Dubhe, supported by the company's own Dubhe Energy foundation model, coordinates the transformers and batteries in real time to match supply with computing demand and manage sudden load swings.
For the firms underwriting these campuses, the claim being tested is power per shell. Envision's argument is that one conversion gets more power into the server hall, where future AI hardware is expected to need it. Distribution past the busbar is not settled. An Open Compute Project white paper released earlier this year proposed busbar, cables or a mixture of the two, which suggests a landlord specifying an 800VDC layout now is committing to a design the industry has not fixed.
"AI data centers need power systems that coordinate renewable energy, energy storage, and the grid to meet computing demand," said Kevin Huang, senior vice president of Envision Energy and president of energy storage and AIDC Power Systems.
The Galaxy campus is planned at 120,000 sqm (1.2 million sq ft) with 3GW of capacity and is designed to support up to one million AI accelerators.
AI data centers need power systems that coordinate renewable energy, energy storage, and the grid to meet computing demandKevin Huang, senior vice president of Envision Energy and president of energy storage and AIDC Power Systems
Reported by the Data centers & fiber desk from Data Center Dynamics. Published . Corrections