Overview
- Envision says its integrated system is now powering live AI workloads at the Galaxy AI campus in Ulanqab, China, converting 10 kV medium-voltage input directly to an 800V DC bus for server racks.
- The architecture pairs solid-state transformers (SST), battery energy storage (BESS) and the Dubhe AI controller to match supply to sudden GPU load swings with millisecond coordination and second-to-minute battery support.
- Envision claims the single-step conversion reduces conversion losses, cabling and equipment footprint, which it says enables higher rack density and lower infrastructure costs at the site.
- The company frames Galaxy as an operational reference for its Mission Gobi plan to scale this model to 5 GW of green AI capacity in desert regions by 2030.
- All technical and performance figures in this report come from Envision’s statements and press release and have not been independently verified, while broader industry context notes rapidly rising data-center power needs through 2027–2030.