China’s data centers rise on the Inner Mongolian plains, infrastructure shaping the AI race

In the vast plains of Ulanqab, where potatoes once formed the sole economic identity, massive concrete blocks are now rising, reshaping the global map of computational power. Potato farms now sit alongside rows of data centers that will feed Chinese AI platforms, from Huawei and Baidu to DeepSec, in a race where computational power has become the most important currency.

From the potato capital to the computing front

Ulanqab, with its volcanic architecture and vast plains, has become the latest front in the AI infrastructure battle. Electrical engineers speak of a “second phase” of construction, followed by “many subsequent phases.” A worker smiles as he describes his country as “infrastructure crazy,” in an off-the-cuff summary of a unrelenting pace. These centers are not merely technical buildings; the Chinese Communist Party deems them “critical to economic and national security,” with the declared goal of integrating AI into 90 % of industries and society by 2030.

Talent flow: a reverse current

The facilities attract a shifting human tide. Jack Chang, a researcher at the Chinese University of Science and Technology, studied at Yale and Cambridge before returning to lead a high-level national program. “I wanted to apply my research in health and medicine, and that requires resources, people and government support that are hard to obtain abroad,” he says. The pattern repeats: fewer students and researchers head to the United States, and more return to China. Tensions with Washington and immigration restrictions under Trump accelerated this shift.

Open model, different sovereignty

China’s response to chip denial was not to copy the closed model but to go the opposite way: open-source models at affordable prices that let companies and countries modify the technology as they wish. Xi Jinping openly encourages this path. The result: faster innovation, partial bypass of the advanced-chip bottleneck, and Beijing’s desire to “set the rules governing AI” on the global stage. While the United States clings to bans, China builds an exportable, adoptable alternative.

Environmental and human cost

The green ambition is declared: the majority of data centers will run on renewable energy by 2030, with wind and solar abundant in these remote provinces and the cold climate reducing cooling needs. Yet the network still relies partially on coal, and water is scarce in arid Inner Mongolia. Whole villages have been emptied, and the figures and compensations are shrouded in uncertainty. “A topic people seldom discuss because voicing concerns can be dangerous,” says a resident who asked to stay anonymous. Another citizen sums up the gap: “These massive data sets have nothing to do with ordinary people… what benefit have they brought us?” Few locals have found jobs, while others fear losing their agricultural land.

Five million graduates annually: the silent engine

In contrast to rural concerns, China produces about five million science and technology graduates each year, versus half a million in the United States. More of them stay domestically, finding a growing innovation ecosystem: researchers integrate AI with nuclear fusion, astrophysics, new materials, and drug development. Dong Ronghao, 27, sees that “Chinese campuses have become more open,” collaborating with researchers worldwide from his base in Shanghai. But Professor Liu Jun, who studied at Harvard and Stanford, warns: “China still rewards compliance over challenge… part of the culture, part of the system.”

What this means for the Gulf, Egypt and the Levant

China’s construction of hundreds of data centers reshapes the global computing market: Chinese demand for advanced chips keeps prices high, and its open-source models (such as DeepSec) give regional developers a runnable alternative on modest hardware. The Chinese data centers’ green-energy strategy parallels Gulf investments in hydrogen and solar, both aiming to decouple computing from carbon. Most importantly, Beijing’s “computational sovereignty” model, owned infrastructure, open models, local talent, offers a template that regional capitals are considering as they build their own sovereign capacities. The race is no longer just about models, but about who controls the land they run on.