Envision announced on November 2, 2026 that an integrated 800V DC power system was supplying live AI workloads at its Galaxy campus in Ulanqab, Inner Mongolia, calling it the world’s first commercial deployment of this kind in a large-scale AI data center. The available reporting establishes that Envision made the announcement, but does not independently establish its “world-first” status or validate the system’s performance claims.
2
3
5
Power architecture and claimed benefits
- Grid-to-computing conversion: Available reporting describes solid-state transformers converting 10 kV grid power directly to an 800V DC distribution bus, with batteries absorbing sudden GPU-training power swings; this is presented as a simplified alternative to a conventional multistage power-conversion chain.
15
16
- Storage and voltage adjustment: The reported system integrates solid-state transformers, battery storage and controls to deliver stable power to AI servers. However, the supplied excerpts do not establish the downstream DC/DC conversion voltages or the detailed interface with computing hardware.
4
14
- Fast response: One retrieved report says the system responds to sudden AI power-demand changes within milliseconds and provides backup for “several minutes.” The available excerpts do not establish the Dubhe controller’s specific control sequence, measured response time or test conditions.
4
- Space, materials and density: Envision claims reduced power-equipment space and copper use, and support for higher-density racks—more computing power within the same physical footprint. The supplied evidence does not quantify cabling reductions, capital-cost savings, footprint reductions or the computing-density improvement.
8
11
13
Campus rollout and commercial ambitions
- Operating versus planned capacity: Envision’s release says Galaxy began operations in August 2026, with planned capacity of 3 GW and a design supporting up to one million AI accelerators; Bloomberg reports an end-of-2028 construction-completion target. Those are campus plans, not evidence that 3 GW or one million accelerators are already operational.
7
1
- Mission Gobi: Retrieved reporting describes Galaxy as the initiative’s first flagship project and gives Mission Gobi a goal of 5 GW of renewable-powered AI infrastructure in desert or arid regions worldwide by 2030.
12
13
- Clean-power and pricing terms: Insufficient evidence in the supplied excerpts to independently verify the claimed greater-than-80% direct clean-energy supply or electricity prices 20% below grid rates for ten years.
What remains unresolved
- “World first”: This should remain explicitly attributed to Envision; repeated publication of a company announcement is not independent verification.
3
5
9
- Battery specifications: I don’t know the installed battery capacity, chemistry, power rating or usable energy from the available evidence.
- Backup and efficiency: “Several minutes” is a reported description, not a disclosed, independently tested backup guarantee at a specified load; no measured conversion efficiency or end-to-end efficiency is established by the supplied excerpts.
4
5
- Deployment scale: The excerpts do not disclose how much of the campus’s computing load actually uses the new 800V system, so the live installation should not be equated with the campus’s ultimate 3 GW capacity.
9
1
Connection to China’s energy and computing strategy
The project’s stated approach is to place large computing loads alongside renewable generation and integrate storage and fast power controls, connecting renewable-energy infrastructure with rising AI demand.
12
13
14
- The available evidence is insufficient to verify the question’s specific national figures: computing-center consumption rising from approximately 170 TWh in 2025 to 800 TWh in 2030, Inner Mongolia holding 57% of China’s wind resources, and a greater-than-80% renewable-electricity requirement for new data centers in major computing hubs.
- It also does not independently establish Envision’s ranking as China’s second-largest wind-turbine supplier or the precise relationship to Beijing’s westward data-center policy.
- Energy-transition implication: The proposed model could support the transition if computing demand is matched by additional clean generation and adequate storage. But millisecond load buffering and minutes of backup do not, by themselves, demonstrate reliable renewable supply across prolonged wind or solar shortfalls; judging that requires generation, storage, grid-import and emissions data that are not available here.