Overview

Envision has moved its integrated 800-volt direct current power system for large AI data centres into commercial operation at the Galaxy AI data centre campus in Ulanqab, Inner Mongolia, China. According to Envision, this is the world's first commercial deployment of an integrated 800V DC power system in a large data centre. The system is powering live computing workloads, giving operators a commercial reference point for direct current infrastructure. Kevin Huang, Senior Vice President of Envision Energy and President of Energy Storage and AIDC Power Systems, framed the milestone as a step from strategic vision to operational reality for power-compute coordination.

Technical Architecture

The architecture delivers power end to end from a 10 kV medium-voltage AC supply to an 800V DC bus. Engineers use direct solid-state transformer conversion, avoiding conventional AC distribution stages. The system pairs solid-state transformers with battery energy storage and equipment that adjusts DC voltage for computing hardware. Envision coordinates these components with its Dubhe Controller, which matches incoming power supply with live computing demand at millisecond-level resolution. This coordination lets local energy storage absorb sudden load fluctuations generated by high-performance GPUs. On-site battery storage provides second-scale power support and minute-scale backup.

Efficiency and Footprint Advantages

Envision says the 800V architecture improves power conversion and watt-to-bit efficiency. It also requires a smaller equipment footprint and reduced cabling and copper requirements compared with conventional set-ups. These combined architectural advantages lower overall infrastructure costs and enable higher computing density within constrained spaces. The design is not simply a voltage change; it is an integrated power chain that combines conversion, storage, and control. By reducing conversion stages and copper-intensive distribution, the architecture addresses two constraints that increasingly shape AI data centre design: electrical efficiency and physical space.

Campus Scale and Renewable Supply

The Galaxy campus began commercial operations in August 2026 and has a planned total capacity of 3GW. Developers designed the facility to eventually house up to one million AI accelerators operating in parallel. Envision plans to supply more than 80% of the required power through direct connections to local renewable electricity sources. The Ulanqab site therefore combines large-scale compute with direct renewable supply and on-site storage. Envision will expand this development approach globally through its Mission Gobi initiative, which seeks to bring 5GW of green AI data centre capacity online across desert regions globally by 2030.

Market Context

The deployment arrives as data centre operators confront rising AI power densities, grid interconnection constraints, and sustainability commitments. Conventional AC distribution architectures rely on multiple conversion stages between the medium-voltage grid and low-voltage server loads. An 800V DC bus can reduce those stages and provide a common interface for solid-state transformers, storage, and compute power delivery. Envision's commercial operation is significant because it provides an operating reference rather than a laboratory demonstration. The company's partner ecosystem includes Microsoft, Accenture, and Amazon Web Services, reflecting the intersection of AI, cloud infrastructure, and energy management.

Industry Impact

For power infrastructure vendors, the Ulanqab deployment validates solid-state transformers and integrated battery storage as part of a data centre power chain rather than as separate ancillary systems. For operators, it suggests a path to higher computing density without proportional increases in electrical distribution footprint. For utilities and grid planners, the architecture's ability to coordinate local storage with compute demand may ease some interconnection and load-fluctuation concerns. However, broad adoption will depend on standards, supply chains, and operational experience. The significance of Envision's deployment is that it moves 800V DC from concept to a live commercial environment.

Implications for AI Data Centres

AI workloads are not steady. High-performance GPUs can produce rapid load swings that stress conventional power delivery. Envision's Dubhe Controller and battery storage are designed to buffer those swings at millisecond and second scales, while minute-scale backup provides ride-through capability. The integration of renewable energy, storage, and the grid is central to the design. As AI data centres grow toward gigawatt-scale campuses, power-compute coordination becomes an operational requirement rather than a strategic aspiration. The Galaxy campus demonstrates an end-to-end architecture that connects renewable supply with computing demand, according to Huang.

Future Outlook

Envision's next challenge is replication. The Mission Gobi initiative aims to bring 5GW of green AI data centre capacity online across desert regions globally by 2030, extending the Ulanqab model to other geographies. Success will depend on local renewable resources, grid conditions, and the ability to standardise integrated 800V DC designs across projects. If the model scales, it could influence how AI campuses are planned: renewable-rich locations, direct DC power delivery, embedded storage, and millisecond-level control. The Ulanqab deployment is a commercial first step, but its larger importance lies in proving that power-compute coordination can operate in production.