A Chinese energy company has built what it claims is the world's largest single AI data center by solving a problem that's paralyzing Western infrastructure: the power grid queue. Envision commissioned a 120,000-square-meter supercomputing facility in Ulanqab, Inner Mongolia, that generates its own renewable electricity on-site rather than waiting years for grid connection approval. The approach highlights a structural bottleneck in the U.S. and Europe that no amount of capital spending can fix.

Why Is Grid Connection Taking So Long in the U.S.?

The United States faces a severe power shortage for data centers. Goldman Sachs Research estimates the country has a structural data center power shortfall of approximately 9.3 gigawatts in 2026, widening to an estimated 45 gigawatts by 2028. The problem is not money or computing chips. It is electricity access.

Grid interconnection queues in the densest data center markets are now running four to seven years. Northern Virginia, Phoenix, and Dallas are particularly congested. According to Schneider Electric, grid connection queues in the U.S. and UK have a 500% backlog, and the standard interconnection timeline for utilities serving data centers has stretched to between seven and ten years. Up to 30-50% of planned 2026 U.S. data center capacity is expected to slip to 2028 because of these constraints.

The timing mismatch is brutal. A modern data center can be physically constructed in 12 to 24 months, but securing a grid connection now takes seven to ten years. Grid interconnection queues across the U.S. hold more than 2,600 gigawatts of pending projects.

How Does Envision's Galaxy Campus Sidestep the Queue?

Instead of connecting to an overloaded regional power grid, Envision wired renewable generation assets directly to the compute facility. The Galaxy Campus uses what Envision calls its proprietary AI Power System, a platform that coordinates renewable generation, battery energy storage, dedicated transmission infrastructure, and computing consumption within a single operating system.

This is architecturally different from how most Western data centers claim to be powered. Most hyperscaler data centers in Western markets claim "100% renewable energy" through virtual power purchase agreements, contracts in which clean electricity is generated somewhere else and matched to the facility's consumption on paper. Galaxy Campus is physically co-located: generation assets sit next to the servers and are wired directly rather than routed through a regional grid.

The technical advantage is measurable. DC-coupled architectures, which connect generators and storage directly to computing loads without AC conversion stages, reduce transmission losses and enable what researchers call "islanded operation." This allows the facility to maintain stable power delivery even when regional grid conditions fluctuate. Vertiv's chief product and technology officer has described DC power as inherently better suited to renewables than traditional AC grid connections.

Why Did Envision Choose Inner Mongolia?

Ulanqab sits approximately 350 kilometers northwest of Beijing. The location is strategic in ways that extend well beyond land cost. The city's average annual temperature is approximately 4 degrees Celsius (39 degrees Fahrenheit), which substantially reduces mechanical cooling requirements, one of the largest operational costs in any data center. Electricity is priced at around 0.32 yuan per kilowatt-hour, approximately $0.05 USD per kilowatt-hour, roughly one-third to one-fifth of industrial electricity prices in U.S. markets under grid-connection pressure.

The energy mix already draws more than 80% from renewables. Inner Mongolia installed 135 gigawatts of combined wind and solar capacity in 2024, surpassing the region's coal power capacity for the first time and leading China in total, new, and generated renewable capacity. Despite the distance from Beijing, network latency remains below five milliseconds, sufficient for demanding AI inference workloads.

Ulanqab is also one of eight national computing hubs under China's "Eastern Data, Western Computing" initiative, the central government's program to shift compute-intensive workloads from resource-constrained eastern cities to infrastructure-rich western regions. The Horinger Data Center Cluster, which includes major parks in Hohhot and Ulanqab, carried 291,000 petaflops of total computing capacity and 273,000 petaflops of intelligent computing capacity at last official count, with green electricity usage running at 84.57%.

What Does This Mean for the Global AI Infrastructure Race?

The commissioning arrives at a moment when scale, not willingness to spend, is the binding constraint on AI development. According to Epoch AI research, five data centers at gigawatt scale or above are expected to come online globally in 2026, each operated by a different hyperscaler. The largest currently operational is xAI's Colossus 2 in Memphis, drawing approximately 946 megawatts.

The spending behind this buildout is staggering. Amazon, Google, Meta, and Microsoft combined for $413 billion in capital expenditures in 2025, an 84% increase from $224 billion in 2024, and are projected to spend $600 billion to $700 billion in 2026. McKinsey has estimated that companies across the compute-power value chain will need to invest $5.2 trillion into data centers by 2030 to meet worldwide AI demand.

Yet capital has not solved the power access problem. Every major Western hyperscaler has responded by signing nuclear or small modular reactor power deals to secure supply outside the standard interconnection queue. Envision's approach is structurally different from both the nuclear bets and the conventional grid-dependent model. By co-locating renewable generation with compute in a region already rich in wind and solar resources, Galaxy Campus bypasses the queue problem at its root.

How Are Western Data Centers Responding to Power Constraints?

Because grid interconnection queues are so congested, hyperscalers and developers are looking to alternative energy sources to get facilities online. Global turbine orders reached 110 gigawatts by the end of 2025, but global manufacturing capacity tops out at just 60 to 70 gigawatts annually. Because of this severe supply-and-demand mismatch, leading turbine manufacturers like GE Vernova are sold out for several years. Large-class gas turbine orders placed today face average delivery wait times stretching up to seven years.

To bridge the gap to cleaner solutions, hyperscalers are shifting to on-site power, using natural gas, fuel cells, or battery storage to keep projects on track. However, even this transitionary path is bottlenecked by severe equipment supply chain delays. As of today, there is around a 100-gigawatt gas backlog, and the earliest a newly ordered gas turbine can ship is 2031, a five-year lead time.

"When this AI boom hit, a lot of developers realised that the fastest way to get to this dispatchable power is natural gas," explained Mandar Pandit, Chief Strategy and Growth Officer for Data Centres for Electrification Systems at GE Vernova.

Mandar Pandit, Chief Strategy and Growth Officer for Data Centres for Electrification Systems at GE Vernova

Steps to Align Data Center Power Strategy With Grid Realities

Industry experts and energy researchers are collaborating on new frameworks to address the timing mismatch between data center construction and grid infrastructure development. Here are the key approaches emerging across the sector:

• DC Flex Initiative: The Electric Power Research Institute (EPRI) is leading an industry-wide initiative known as DC Flex to help data centers act as flexible resources for the grid rather than passive consumers. The program boasts 70 participating members and explores how onsite generation can keep facilities running when grid capacity is tight.
• On-Site Generation Mix: Rather than relying solely on gas turbines, operators are exploring renewables and battery storage as non-emitting or low-emitting options. GE Vernova has completed 100% hydrogen capability testing and is working on the first commercial carbon capture project on the UK side.
• Co-Located Renewable Architecture: Following Envision's model, operators can site facilities in regions with abundant wind and solar resources, wire generation directly to compute loads using DC-coupled systems, and bypass grid interconnection queues entirely.

The fundamental problem lies in a severe timeline mismatch between how quickly a data center can be constructed and how long it takes to build grid infrastructure. When data centers scale from 100 or 200 megawatts to gigawatt-scale loads, they become the size of a small city in terms of power demand. There is no way the grid can accommodate that size getting interconnected in two years.

"When we think about onsite power, it's not just gas, there are renewables and battery storage. There are options that are non-emitting or low emitting, and it becomes the data center operator's choice on how they set it up," noted Anuja Ratnayake, Emerging Technologies Executive at EPRI.

Anuja Ratnayake, Emerging Technologies Executive at EPRI

Envision is now the most ambitious operator to plant its flag in China's computing hub framework. DeepSeek, the Chinese AI startup, separately announced plans this week for its own gigawatt-scale campus in Ulanqab, underscoring that the region has crossed from regional ambition to genuine hyperscaler destination.

Galaxy Campus is the first flagship project under Envision's Mission Gobi initiative, which the company unveiled at the VivaTech conference in Paris in June 2026. The initiative targets 5 gigawatts of green AI computing capacity in desert and arid regions worldwide by 2030. The ambition reflects Envision's position in the global energy industry. The company ranked second globally by wind turbine installations in 2025, shipping 20.9 gigawatts, nearly a quarter installed outside mainland China.

The industry is also shifting its focus geographically. The Data Center Frontier Trends Summit, a major industry conference, will move to Glendale, Arizona, in May 2027 and co-locate with Microgrid Knowledge for the first time. The move reflects where the two audiences have been converging: power. The Southwest has become one of the fastest-growing data center corridors in North America, with Glendale and the greater Phoenix market drawing record development driven by land availability, transmission access, and an expanding energy generation mix.