Envision's 2GW AI Data Centre in China Runs Entirely on Renewables

Envision Energy has commissioned its Galaxy Campus in Ulanqab, Inner Mongolia, a 120,000-square-metre AI supercomputing facility designed to scale beyond 2 gigawatts of capacity, supporting up to one million AI accelerators once fully built out. The facility is powered entirely by dedicated renewable generation, transmission infrastructure and large-scale energy storage rather than drawing from a conventional grid connection. Envision says the design delivers up to ten times more compute output per square metre than conventional data centres, though this figure and the facility's "world's only" gigawatt-scale green data centre claim come from the company itself rather than independent verification.

 

Why Building Power Infrastructure Into the Facility Itself Matters

 

The most consequential design choice at Galaxy Campus is treating the power system as an integral part of the data centre's architecture rather than an external dependency the facility simply connects to. Ricky Zheng, general manager of Envision's AI Data Center division, described this as requiring a purpose-built AI Power System integrating renewable generation, dedicated transmission infrastructure and large-scale energy storage specifically to deliver the stable, low-cost green power a cluster of this size requires.

That distinction addresses a genuine and increasingly acute problem facing AI infrastructure development globally: conventional data centre expansion is frequently constrained by existing grid capacity, since utility grids were generally not built to accommodate the concentrated, rapidly scaling power demand a single AI cluster containing hundreds of thousands of accelerators requires. By designing dedicated generation and storage capacity directly alongside the compute infrastructure from the outset, Envision is attempting to avoid competing with existing grid users and interconnection queues entirely, a similar logic to the onsite gas generation projects covered elsewhere in recent reporting, except built specifically around renewable rather than fossil generation.

 

UPCOMING WEBINAR

How Two Sustainability Teams Used AI to Get Audit-Ready

Online, 18 August 2026 · Cleaning spend data, building audit trails and meeting what auditors expect

 

Why Ulanqab's Combination of Grid Connectivity and Renewable Resources Matters

 

Ulanqab has developed into a significant digital infrastructure hub within China, giving the project access to the network performance an AI cluster of this scale requires for accelerators to communicate with each other at the speed large AI training workloads demand. That existing connectivity infrastructure matters as much as the power supply itself, since a data centre with abundant renewable power but poor network performance would still be unsuitable for the kind of large-scale AI training and inference workloads this facility is designed to support.

The region's wind and solar resources independently make it well suited to Envision's renewable-first model, providing the underlying generation capacity dedicated transmission and storage infrastructure can be built around. That combination of existing digital infrastructure and strong renewable resource availability in the same location is relatively uncommon, which likely explains why Envision selected this specific site as the flagship demonstration for its broader strategy.

 

RELATED EVENT

Reuters Sustainability Europe 2026

London, 20–21 October 2026 · 400+ sustainability, finance and procurement leaders on regulatory uncertainty and Scope 3 data

 

What the 10x Compute Density Claim Would Require to Hold Up

 

Envision's claim of delivering up to ten times more compute output per square metre than conventional data centres is a substantial performance claim that would require verification against a clearly defined baseline and methodology to assess independently, since compute density comparisons depend heavily on assumptions about cooling technology, accelerator generation, power delivery efficiency and facility design that vary considerably across different data centre operators. Without independent third-party benchmarking data accompanying this claim, it should be read as the company's own performance target rather than an externally validated result.

Ricky Zheng framed the underlying strategic thesis directly, arguing that as AI models grow larger and more compute-intensive, the binding constraints on further scaling increasingly become power availability, network performance and energy efficiency rather than compute hardware itself, a framing that positions Envision's integrated energy-and-compute approach as addressing what the company views as the actual bottleneck facing the AI industry's continued expansion.

 

UPCOMING WEBINAR

Moving Sustainable Procurement Beyond Cost to Strategic Value Creation

Online, 09 September 2026 · Balancing cost, managing supplier risk and building long-term value

 

What the Broader Mission Gobi Initiative Signals

 

Galaxy Campus is the first flagship project under Envision's Mission Gobi initiative, unveiled at VivaTech in Paris in June 2026, which targets 5 gigawatts of green AI computing capacity across desert and arid regions globally by 2030. That broader ambition suggests Ulanqab is intended as a proof-of-concept demonstrating the model's viability before Envision attempts to replicate it across other arid, renewable-resource-rich locations worldwide, a strategy that depends on finding additional sites combining the same rare intersection of strong renewable resources and adequate digital connectivity infrastructure that made Ulanqab suitable.

Whether Envision's claimed compute density and cost advantages hold up under independent scrutiny once the facility reaches full operational scale, and whether the broader Mission Gobi initiative successfully identifies and develops comparable sites to reach its 5 gigawatt 2030 target, will determine how significant a template this integrated renewable-and-compute architecture becomes for the wider AI infrastructure industry's approach to its mounting power constraints.