AnswersPublished29 sources
Ulanqab’s Green-Power Bet Is Turning a Grassland City Into an AI Compute Hub
Ulanqab’s advantage is the combination of 4.2 millisecond fiber latency to Beijing and renewable electricity priced at roughly 0.33–0.36 yuan per kWh—not cheap power alone. The city’s operating compute also expanded rapidly: reported online capacity reached about 165 EFLOPS by June 2026, while first half data center...
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- Ulanqab’s advantage is the combination of 4.2 millisecond fiber latency to Beijing and renewable electricity priced at roughly 0.33–0.36 yuan per kWh—not cheap power alone.
- The city’s operating compute also expanded rapidly: reported online capacity reached about 165 EFLOPS by June 2026, while first half data center electricity use rose roughly 90% year over year.
- Envision’s Galaxy Campus illustrates the model: a 120,000 square meter AI facility planned for more than 2 GW and up to 1 million accelerators, supplied through a system that links renewable generation, power delivery...
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How did Ulanqab, an Inner Mongolian city about 240 kilometers from Beijing, become one of Asia-Pacific’s fastest-growing AI compute clusters. Article summary: Ulanqab has turned proximity to Beijing into an AI-infrastructure advantage: near-real-time connectivity makes it suitable for Beijing-linked workloads, while cheap local renewable power makes energy-intensive model trai. Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Ulanqab is emerging as an important AI infrastructure hub because it solves two problems at once: it is close enough to Beijing for low-latency workloads, yet has access to cheaper renewable electricity than many eastern locations. Fiber round-trip latency to the capital is reported at about 4.2 milliseconds, while the city has more than 20 GW of renewable-energy installations, over 3,200 annual sunshine hours and a green-power share of about 67%.15
37
That combination matters as AI companies move from experimental model training to permanent, power-hungry compute fleets. Ulanqab’s “token factory” reputation is therefore less about a single hyperscale site than about building an operating system for AI compute: renewable generation, dedicated power connections, storage, cooling, networking and data-center capacity in one region.
Why Ulanqab is attractive to AI companies
Low latency keeps the city connected to Beijing
Ulanqab is far enough from Beijing to offer more land and energy capacity, but close enough to remain connected to the capital’s cloud and enterprise markets. The reported 4.2-millisecond fiber latency gives the city a plausible role in latency-sensitive overflow and online AI workloads, rather than restricting it to batch processing alone.37
That is the strategic logic behind China’s broader “east-to-west computing” approach: place energy-intensive workloads where power and land are more available, while using high-capacity networks to serve demand concentrated in eastern cities. Ulanqab’s connectivity makes that trade-off more practical for Beijing-linked customers.32
42
Renewable power lowers the cost of running dense AI clusters
AI data centers are unusually sensitive to electricity costs because accelerators run at high density and often operate continuously. Ulanqab’s reported data-center electricity prices of roughly 0.33–0.36 yuan per kilowatt-hour, combined with its wind and solar resources, give operators a cost advantage over many eastern-coastal locations.36
37
The city’s energy profile also aligns with China’s push for greener AI infrastructure. Reporting on Ulanqab says new AI-compute projects are expected to source at least 80% of their power from green sources, while new projects in the cluster have been reported at above 86%.27 Direct renewable-power connections and “source-grid-load-storage” projects are intended to reduce the distance between generation and the servers consuming the electricity.
42
43
Cool conditions help as well. Ulanqab’s average annual temperature has been reported at about 4.3°C, with natural cooling available for much of the year, potentially reducing the energy needed to remove heat from high-density computing equipment.53
How quickly has the cluster grown?
The headline expansion is substantial, but the numbers need to be separated by what they measure.
- Committed capacity: Operators and internet companies had committed roughly 12.5 GW of data-center capacity by June 2026, up from 3.3 GW in July 2025. More than 70% of those commitments were announced during the preceding year.
27
33
- Operating capacity: Officially reported figures put Ulanqab’s operating compute scale at about 165,000 PFLOPS, or roughly 165 EFLOPS, by the first half of 2026. Intelligent compute accounted for more than 90% of the total in reports citing local figures.
42
- Utilized data-center capacity: Utilized capacity rose more than 70% year over year to about 800 MW in 2025, according to reporting on the cluster.
- Electricity demand: Data-center electricity consumption in the first half of 2026 was reported to have increased by roughly 90% year over year, reaching about 3.3 TWh in one account.
These figures are not interchangeable. Gigawatts describe electrical capacity, PFLOPS and EFLOPS describe computing performance, and megawatts of utilized capacity describe the load being used. Most importantly, the 12.5 GW figure includes planned and operating projects; it does not mean that 12.5 GW of AI infrastructure was already running.27
40
Which companies are building or using capacity there?
Ulanqab’s ecosystem includes data-center operators, infrastructure companies, internet platforms and AI-model developers. Reports identify GDS Holdings, VNET Group and Envision among the major infrastructure players, while DeepSeek and Kimi are cited among the model companies with a presence or planned involvement. Tencent and ByteDance are among the large platform customers associated with the cluster.33
51
DeepSeek is also reported to be considering roughly 1 GW of additional capacity in Ulanqab through a combination of owned facilities and leased capacity. Some of that capacity could come online from late 2027 into early 2028, according to reporting based on people familiar with the matter.17
57
That plan remains a reported intention, not confirmed operating capacity. The available reporting says the company had not publicly confirmed the project, and key details—including the exact chip configuration and the meaning of the 1 GW figure—remain unclear.45
Envision’s Galaxy Campus shows the infrastructure model
Envision’s Xinghe, or Galaxy, Campus is the clearest example of Ulanqab’s strategy. The campus contains a 120,000-square-meter AI data-center building—roughly the area of 20 standard football pitches—with planned power capacity above 2 GW and room for up to 1 million AI accelerators.3
21
The significance is not only the building’s size. Envision is presenting the campus as an integrated energy-and-compute system, combining renewable generation, direct green-power supply, storage, power electronics and high-density AI computing.3
51
The first phase is smaller than the ultimate design. One energy-industry account describes an initial 120 MW phase allocated to local technology companies, with the wider campus planned to scale toward 2 GW.14 That distinction is important: a facility can be commissioned while its full power and accelerator capacity remain future milestones.
Why storage and transmission will decide whether the boom lasts
Wind and solar generation do not automatically match the steady and sometimes sharply fluctuating demand of AI training. Large accelerator clusters can create synchronized power changes, placing unusual demands on transformers, power electronics and grid controls.48
51
Storage is therefore more than a sustainability feature. It can absorb surplus renewable generation, provide power when wind or sunlight falls, and respond quickly to sudden changes in computing load. Envision’s proposed approach uses storage across multiple timescales, from longer-duration balancing to rapid power-quality control for dense AI systems.51
Transmission and interconnection are just as important. Renewable generation has economic value only when dedicated lines, substations and the wider grid can deliver dependable power to the racks. Ulanqab’s bottleneck could eventually be grid connection and power-quality infrastructure rather than available land or renewable resources.22
42
The 2028 test: commitments versus dependable compute
The period from 2028 will provide a useful reality check. It is the period when some of DeepSeek’s reported Ulanqab capacity could begin coming online, testing whether the city can turn announced projects into reliable, commercially used compute.17
45
The broader test is even more consequential. Ulanqab must deliver GPUs or other accelerators, complete construction, attract sustained customer demand, expand transmission, and keep renewable-heavy power reliable around the clock. A large project pipeline demonstrates investor interest; it does not by itself demonstrate utilization or profitability.
Ulanqab’s competitive advantage is best understood as a systems proposition: Beijing-grade connectivity paired with low-cost renewable energy and infrastructure designed around AI’s power profile. If storage and grid integration keep pace with construction, the city could become a durable example of how renewable-energy regions support AI growth. If they do not, its 12.5 GW headline will remain a pipeline rather than a working “token factory.”
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Ulanqab’s Green-Power Bet Is Turning a Grassland City Into an AI Compute Hub"?
Ulanqab’s advantage is the combination of 4.2 millisecond fiber latency to Beijing and renewable electricity priced at roughly 0.33–0.36 yuan per kWh—not cheap power alone.
What are the key points to validate first?
Ulanqab’s advantage is the combination of 4.2 millisecond fiber latency to Beijing and renewable electricity priced at roughly 0.33–0.36 yuan per kWh—not cheap power alone. The city’s operating compute also expanded rapidly: reported online capacity reached about 165 EFLOPS by June 2026, while first half data center electricity use rose roughly 90% year over year.
What should I do next in practice?
Envision’s Galaxy Campus illustrates the model: a 120,000 square meter AI facility planned for more than 2 GW and up to 1 million accelerators, supplied through a system that links renewable generation, power delivery...