After a quick 40 min high speed rail trip from Beijing North Station, I step off the platform in the sleepy station of Donghuayuan Bei in Hebei’s Zhangjiakou. From the elevated platform I get a glimpse of the wind turbines dotting Guanting Reservoir 官厅水库. On the other side of the platform I look out across a semi-rural landscape, dotted with apartment blocks and in the distance the hulking warehouses housing new servers to power China’s computing needs. The reservoir was built in the 1950s to provide Beijing with water. In the lead up to the Winter Olympics in 2022, Zhangjiakou saw a boom in infrastructure investment, including in renewable power, much of it to supply Beijing. Now, the city is being mobilized to supply the capital again, this time with its growing cloud computing infrastructure.
An aerial shot of the emerging data center cluster, with Guanting Lake Wind Farm in background
Data centers stretch into the dusk horizon in Huailai (Author, 2025)
Wind turbines on Guanting Lake Reservoir in Huailai County (Author, 2025)
This is one of the eight data center clusters数据中心集群 of China’s Eastern Data Western Compute dongshu xisuan 东数西算 project. In this post, I’ll touch on some of the dynamics of China’s AI infrastructure and data center development through Zhangjiakou. Why this city of all the eight clusters? Zhangjiakou is adjacent to Beijing, and therefore offers lower latency for cloud//AI usage than more remote clusters, and also has solar and wind energy. This is why Zhangjiakou is attracting significant private data center investment in addition to those of state-operated telecoms, making it a relative success compared to some of the other more far-flung clusters such as in Gansu and Ningxia, or those in Xinjiang (which are not officially part of the EDWC plan).
The EDWC project was conceived to coordinate the growing energy needs of data centers with renewable energy buildout. However, the process of data center and energy permitting has turned into a new arena of political contestation. Some data center operators bribe their way to obtain permits for building from local governments, fueling a speculative boom that may outstrip the capacity of renewable energy in the areas where they are built. Even as data centers are built adjacent to renewable sources, they actually draw heavily on dirty energy like coal or gas-fired power plants, which provide a more reliable energy source required by data centers. According to a recent report from Xinhua , Zhangjiakou’s data centers used 1.859 billion kWh, up 43.47% YoY—-”green electricity” accounted for 574 million kWh, about 30%.
By some measures the effort has been a success—according to official statistics released by China’s National Data Bureau 45.5 bn Yuan ($6bn) of capital has been invested in data centers via the plan. However, other reports suggest that many data centers lie underutilized or even empty. When the effort was conceived, the idea was to “optimize the layout of China’s data centers.” The plan sought to avoid overbuilding of data centers in crowded urban areas where land and energy prices are higher and concentrate data centers where there was ample cheap and renewable power supply. But in fact, the EDWC policy has encouraged many cities beyond the eight official clusters to pursue data center development. Real estate companies and other conglomerates have tried to grab a piece of the data center building frenzy, leading to creation of companies that have little experience in data center buildout. The result is a familiar story of overbuilding and overcapacity—time will tell if data center demand will catch up with the overinvestment, or if the frenzy will end similarly to China’s real estate boom, with empty and/or underutilized facilities.
In 2022, four agencies jointly issued a plan called Eastern Data Western Compute 东数西算, which aimed to boost China’s computing power by constructing new data centers in eight designated clusters across the country. China’s western provinces are rich in energy: both traditional oil and natural gas, as well as renewables like wind and solar. Meanwhile, most data centers had been built in the prosperous eastern metropolises, to be nearer to most customers. Latency, a measure of the time it takes for a packet of data to travel between a user and the server and back, is highly correlated with distance: closer computing facilities enable lower latency, and faster computing. But the western hubs would be suitable for certain data operations: cold storage and backup. With the advent of energy-intensive AI model training in the wake of ChatGPT’s release, suddenly inland data hubs had a newfound purpose as ideal sites for intensive model training operations. As many of you know, I’ve been covering the EDWC policy for a while, but this year many English language outlets began to pay attention—writing about newly built data centers in Xinjiang, Qinghai, and Wuhu—which the FT called “China’s Stargate”, alluding to the massive OpenAI backed data center being built in Abilene Texas.
The EDWC plan built on the initial success of Guizhou, which began attracting data centers by major Chinese tech companies around 2013, and later Inner Mongolia, which sought to replicate Guizhou in the North. Both inland regions are blessed with suitable cooler climates, and energy: Guizhou with hydropower, and Inner Mongolia with wind and solar. In 2022, the EDWC plan was released, adding six additional clusters to a national framework. Four of these (Shaoguan, north of Guangzhou; Wuhu West of Shanghai, Zhangjiakou outside Beijing; and Chengdu-Chongqing) are located in or near major metropolitan areas, while four of them were in remote Western regions: Qinyang in Gansu, Hohhot (Inner Mongolia), Guizhou, and Ningxia.
Meanwhile, many cities outside these began building data centers. Datong is a city in northern Shanxi province, just southwest of Zhangjiakou. While no cities in Shanxi 山西 province are part of the official EDWC plan, the province’s climate, location, and cheaper land and energy costs make it an ideal location for data centers like its larger neighbor to the north, Inner Mongolia. Chindata, a leading Beijing-based data center operator, built a massive hyperscale facility in Datong. Other cities in Inner Mongolia have cultivated their own data clusters. Ulanqaab 乌兰察布is a prefecture-level city in Inner Mongolia, located about halfway between the provincial capital Hohhot and Hebei’s Zhangjiakou. It has developed a big data industrial park with large cloud computing facilities, including Apple’s second data center in China. Even within Inner Mongolia, the rise of two cloud computing hubs is indicative of the decentralized rollout of data center development, seen by local governments as a new engine of growth. Specialized data centers in Xinjiang and Tibet have also attracted media attention recently. While these regions have high solar and wind potential, their distance from urban areas makes them relatively unattractive for most types of data centers, but they could be used to train AI models, which does not require lower latency needed for real time computing. Desert data centers in remote Yiwu, Xinjiang got attention from Bloomberg earlier this year, but in reality China’s data center boom is mostly still happening in areas closer to major cities—in places like Zhangjiakou 张家口which offers both proximity to Beijing, and benefits of cooler climate and abundant renewable energy. Getting trained engineers and reliable infrastructure to operate data centers in Inner Mongolia and Gansu is hard enough, let alone in remote Xinjiang.
... continue reading