China impulsa megacentros de IA con energía barata

China pours massive resources into building massive AI hubs, but the speed of construction clashes with the difficulty of obtaining sufficient advanced semiconductors due to US restrictions. China is constructing data centers rapidly in its energy-rich interior in a bid for Beijing to turn abundant electricity and cheap land into a competitive advantage in the global AI race.

This initiative is creating huge centers, like the one in Ulanqab Inner Mongolia, where 89 data centers have been built or planned. The deployment of infrastructure for artificial intelligence is one of the fastest and largest in the world. China has defined the advancement of AI as a national strategic priority to boost economic productivity, consolidate domestic political influence, and ensure global technological dominance.

While China struggles to secure enough advanced chips to meet the growing AI demand, it has mobilized land, electricity, and construction capacity to raise the data centers that host them. China already has 24 gigawatts (GW) of operational computing capacity in data centers - a figure higher than the rest of Asia combined, but less than half of the 56 GW in the US, according to Semi-Analysis.

There are another 50 GW under construction or announced. Ulanqab, known for its potatoes and yogurt, has become a data center hub, hosting about 15 GW of capacity, according to a sector operator. It has abundant energy, cheap land, and a cold climate, advantages hard to find in the US, where electrical connections are scarce. Engineers are even converting an abandoned high school into a data center.

The operator ZData took over the facilities after their closure due to declining student enrollments caused by population decline in the region. Soon, rows of hidden computing equipment will fill the schoolyard where students used to gather between classes. The scale of construction is visible in Ulanqab's outskirts, with at least a dozen companies spread across four extensive data center complexes.

Among them are Alibaba and Huawei, along with operators like ZData and VNET, constructing facilities and leasing computational capacity to tech groups. ByteDance has become the largest source of demand for leased infrastructure in China, while DeepSeek and Z.ai labs are also building installations in Ulanqab. Eric Xu, Huawei's rotating president, who has three data centers in Ulanqab, said the tech giant evaluated numerous locations before settling on the city due to its cheap electricity and proximity to Beijing, just two hours by high-speed train.

Inner Mongolia had 117 GW of installed electricity capacity by June, the highest in China and nearly four times the total installed electricity capacity of the United Kingdom. It also has about 130 GW of fossil fuel-generated electricity generation capacity, mainly coal. The region's cold climate also helps reduce the massive cooling needs of AI clusters.

Ulanqab's average annual temperature is around 4.3°C, and its proximity to Beijing allows engineering teams to quickly reach the facilities. Authorities have authorized some data center projects to build exclusive energy generation systems connected directly to their computational clusters, bypassing some of the electrical network limitations faced by developers in other areas.

However, the region is not without resource limitations. Inner Mongolia's interior lacks abundant water reserves, a problem that could become more significant as the industry grows. But China's advantage goes beyond energy access. Developers in Ulanqab are also building data centers at lower costs and faster than in the US. Contractors from across the country have been awarded contracts for these projects; construction costs in the region are about 20% lower than in major cities, according to Goldman Sachs.

China has mobilized thousands of workers to build data centers in 12 to 18 months, compared to the 18 to 24 months typical in the US. The speed of construction is attracting manufacturers and suppliers to set up nearby to reduce costs and timelines. In Ulanqab, Envision is developing renewable energy projects, and Vertiv supplies cooling equipment.

Beijing is also pushing for the use of domestic technology: data centers using Chinese processors instead of Nvidia's receive tax benefits and discounts on electricity and water. Analysts believe it is unlikely China will accumulate data centers without using them. The problem is that while China builds these facilities at a rapid pace, it needs to secure enough advanced processors to take advantage of the massive computing power it is deploying.

The restrictions imposed by Beijing and Washington on Chinese companies' purchases of Nvidia chips, combined with the limited local production capacity, have made this a fundamental obstacle. The question is whether data centers can secure enough chips to equip the installations.