Deep Dive delves into hot issues in Hong Kong and mainland China. Our easy-to-read articles provide context to grasp what’s happening, while our questions help you craft informed responses. Check sample answers at the end of the page.
News: Fuelled by cheap wind power and a cool climate
-
Ulanqab is leveraging its cool climate and abundant wind and solar energy to build massive green-powered AI data centres
-
Rapid AI expansion is straining energy grids worldwide
While an artificial intelligence (AI) query might take mere seconds for users to receive an answer, the process behind that response is complex. It involves a journey back to places like Ulanqab.
In this region, which plays a crucial role in driving the AI boom, servers consume electricity, cooling systems manage heat and wind turbines harness the movement of air across the grasslands to generate power.
For generations, Ulanqab – a city several hundred kilometres northwest of Beijing – was known for its windswept pastureland. Strong breezes sweep through the Yinshan mountain range, powering giant wind turbines.
As AI’s rapid growth increases global demand for computing power and electricity, Ulanqab’s cool climate is transforming this pastoral city in Inner Mongolia into one of China’s most valuable industrial assets.
Envision, a green technology company, is developing its Galaxy Campus in Ulanqab. It is set to become one of the world’s largest AI computing hubs.
Envision selected this city to power its data centres with local renewable energy sources, which currently provide 67 per cent of Ulanqab’s electricity. The region’s chilly temperature allows the facility to use outside air for cooling instead of energy-heavy air conditioning.
Leading data-centre clusters in Ulanqab’s region have a power usage effectiveness (PUE) below 1.15, against a national average of nearly 1.5.
This means in Ulanqab, a greater proportion of each watt of electricity is effectively reaching the server chips.
According to the National Energy Administration (NEA), China’s data centres consumed approximately 170 billion kilowatt-hours of electricity in 2025. This accounted for around 1.6 per cent of the country’s total electricity usage.
By 2030, the annual consumption by data centres is expected to increase to about 800 billion kilowatt-hours, nearly five times the 2025 levels. It will represent roughly 6 per cent of national electricity use in 2030, as stated by Wang Hongzhi, the director of the NEA.
Greenpeace’s recent Clean Cloud report tracked the renewable energy use of 25 Chinese technology companies. By August, it found that eight reported achieving annual renewable energy ratios that exceeded 30 per cent. In their 2024 report, there was only one company achieving that.
Technology giants are betting that Ulanqab’s wind and solar power, combined with a cooler climate, can provide a more sustainable foundation for AI infrastructure.
Staff writers
Question prompts
1. How is an AI query related to Ulanqab in China? Choose two statements that apply.
(1) Ulanqab’s wind turbines run the servers that might power an AI chatbot’s answers.
(2) Ulanqab’s cold weather helps to cool down the data centres behind many AI companies.
(3) Ulanqab’s economy relies heavily on importing electricity from Beijing to support its AI companies.
(4) Ulanqab provides the human effort needed to manually calculate AI queries.
A. (1), (2) only
B. (1), (3) only
C. (2), (4) only
D. (3), (4) only
2. Explain two factors that make Ulanqab a good location for hosting AI data centres.
3. Do you think Chinese tech companies are getting most of their power from sources like those found in Ulanqab? Explain your answer using information from News.
Chart
Question prompts
1. What is the difference between the electricity consumption of China’s AI data centres in 2025 and what is projected for 2030?
2. How does the power usage effectiveness (PUE) of data-centre clusters in the Ulanqab area compare with the national average? If more data centres had a PUE closer to what Ulanqab has, how might this affect the electricity consumption projection for 2030?
Issue: Surging demand for AI tokens
-
Ulanqab converts its vast wind potential and cold climate into major cost savings for AI companies
-
Soaring AI token consumption is driving Ulanqab’s push to reach over 200,000 petaflops of computing capacity
China’s appetite for AI tokens has exploded in recent years. Daily consumption rose from about 100 billion at the beginning of 2024 to 100 trillion by the end of 2025. It then exceeded 140 trillion in March 2026, according to government data.
This surge in token demand is driving a greater need for the electricity and computing infrastructure required to produce them, and every AI token comes with a cost.
“Every 1 million tokens generated requires two to three kilowatt-hours of electricity,” said Luo Xiyong of data centre operator Zhonglian Data.
In Ulanqab, data centre operators have access to some of China’s cheapest power.
Rates vary, but some operators have secured electricity for about 0.35 yuan (HK$0.41) per kilowatt-hour, less than half the roughly 0.70 yuan charged in eastern provinces. Electricity can be 60 to 70 per cent of a data centre’s operating costs, making even modest savings significant.
For China’s burgeoning AI industry, which is still searching for viable ways to monetise its models, cheap power is a central part of the business model.
For Ulanqab, the next challenge is not simply building more servers but making sure those servers are connected to real demand.
Regions such as Ulanqab have attracted data centres because of cheap electricity, abundant land and favourable climates.
Meanwhile, much of the demand for AI computing remains concentrated in China’s eastern cities and affluent regions, where technology companies and users are based.
Beijing is pushing to connect its AI infrastructure with the country’s power system. In Ulanqab, wind farms could provide renewable electricity. Data centres could then convert that power into computing capacity, and a national network could help deliver that capacity to users far beyond Inner Mongolia.
The city is also seeking to move up the AI value chain. Under a three-year action plan, Ulanqab plans to expand computing capacity to more than 200,000 petaflops – a measurement of computing capacity – by 2026 and develop a platform for trading and settling AI tokens.
The plan calls for Ulanqab to evolve from a data centre hub into a broader computing and AI services centre, with a focus on green computing for large AI model training and inference. It also aims to encourage investment in liquid cooling, computing hardware, AI software and talent.
The aim is to capture more value from its renewable power: not simply selling electricity or computing capacity, but turning that power into AI computing and ultimately into the tokens generated by AI models.
Staff writers
Question prompts
1. Why is low-cost electricity considered a central part of the business model for AI data centres?
2. What would be the benefit of connecting AI infrastructure with the wider power system in China?
3. What would it mean for Ulanqab to “move up the AI value chain”?
Graph
1. Using your own knowledge, what do you think has changed from 2024 to 2026 that has caused this increase in AI token usage?
2. What is the cost of generating these AI tokens? To what extent do you agree that companies and people should continue this trend of using increasingly more tokens?
Glossary
-
Ulanqab: accounts for about one-tenth of China’s total wind resources and one-third of the Inner Mongolia autonomous region’s wind resources, according to China’s Xinhua news agency. As of June, Ulanqab had signed 89 data centre deals, 86 of which were AI computing projects, based on local government data.
-
wind turbines: giant spinning towers with blades that convert wind movement into electrical power
-
power usage effectiveness: a measure of energy efficiency of a data centre. Perfect efficiency is represented by a PUE of 1.0, where every watt goes directly to computing, while higher numbers indicate energy wasted on supporting systems.
-
token: a measure for the units of text processed by AI models
-
petaflops: a unit measuring computing speed, equal to one quadrillion floating-point operations per second
Sample answers
News
1. A. (1), (2) only
2. Ulanqab’s cool climate allows for efficient cooling of data centres using outside air, reducing the need for energy-heavy air conditioning. Additionally, the region’s renewable energy sources ensure a sustainable power supply.
3. Only eight out of 25 Chinese tech companies mentioned in the Clean Cloud report achieved annual renewable energy ratios above 30 per cent. This indicates that the majority of these companies rely on non-renewable energy sources, which are harmful to the environment.
Chart 1
1. In 2025, China’s AI data centres consumed approximately 170 billion kilowatt-hours of electricity. By 2030, this consumption is expected to rise to about 800 billion kilowatt-hours, indicating an increase of nearly five times the 2025 levels.
2. The PUE of data-centre clusters in Ulanqab is below 1.15, significantly better than the national average of nearly 1.5. If more data centres could achieve a similar PUE, they would be more efficient in electricity usage, potentially lowering the overall electricity consumption projections for 2030. This could help reduce the anticipated increase in demand from 800 billion kilowatt-hours. More efficient data centres would allow for greater sustainability in powering AI infrastructure.
Issue
1. Low-cost electricity is crucial for AI data centres as it constitutes 60 to 70 per cent of their operating costs. With the demand for AI tokens soaring, the electricity needed for generation significantly impacts overall expenses. Affordable power enables operators to reduce costs, making their business models more sustainable and competitive.
2. Connecting AI infrastructure with the wider power system in China would enable efficient delivery of computing capacity to users across the country, particularly in demand-rich eastern cities. It would also leverage renewable energy sources like wind farms, promoting sustainability. This integration can optimise the production of AI tokens and enhance the overall value captured from renewable power.
3. For Ulanqab to “move up the AI value chain,” it means transitioning from just being a data centre hub to a comprehensive AI services centre. This includes expanding computing capabilities, developing platforms for trading AI tokens and focusing on green computing. By enhancing these services, Ulanqab can create greater value from its renewable energy resources.
Graph
1. The increase in AI token usage from 2024 to 2026 likely stems from advancements in AI technology, growing demand from businesses for AI applications and improved infrastructure supporting larger-scale AI model training. Additionally, the availability of cheap electricity in regions like Ulanqab has made token generation more economically viable, fuelling further growth.
2. Generating 1 million AI tokens requires two to three kilowatt-hours of electricity, contributing to substantial operational costs for data centres. While the demand for tokens indicates innovation and growth, it is crucial to balance this with sustainability and the environmental impact of increased electricity consumption. Companies and individuals should pursue tokens thoughtfully, ensuring that the benefits align with ecological considerations and responsible energy use.
