Why AI is moving to the middle of nowhere

Major Chinese technology firms are building vast arrays of servers in the arid plains of Inner Mongolia. While tech giants used to rent space in shared facilities, they are now racing to build their own dedicated infrastructure. Ulanqab has become the preferred location for these projects. It is a region traditionally known for sheep farming and coal mining, but it now boasts nearly 100 data centers, with projects planned that could eventually exceed the total energy capacity of even the most ambitious global infrastructure projects currently on the drawing board.

AI requires two main things on a massive scale: electricity and cooling. To train an AI, companies run thousands of processors at peak capacity for months at a time. This generates a massive amount of heat. If these processors get too hot, they crash or melt. Data center operators usually have two ways to solve this: use energy-intensive air conditioning or find a naturally cold environment. By building in Ulanqab, companies save money because the climate does much of the cooling work for them. Furthermore, the region offers abundant, low-cost electricity from a mix of coal, wind, and solar, which keeps the astronomical energy bills of these computer arrays manageable.

The shift to remote locations highlights the massive, often invisible, environmental footprint of AI. While these companies are moving toward renewable energy, the sheer scale of the power required is putting a strain on local resources, particularly water. Cooling systems often rely on water to regulate temperatures, yet Ulanqab is a dry region already struggling to supply its residents. As companies push to build faster, we are learning that AI is not just a digital phenomenon; it is a physical industry that demands vast amounts of power, water, and land. We are effectively choosing where to sacrifice local resources to feed the global hunger for faster and smarter AI models.