There is a big problem with building underground data centers

Data centers have become the necessary evil that we all must contend with for the sake of rapid advancements in AI. However, they are problematic and most ordinary people are strongly opposed to them for obvious reasons. Data centers are so noisy that their noise can extend for a mile, and they can also drive up local utility costs, take up valuable land, and waste water supplies. So why not put them underground? This would theoretically solve most of these problems.

If only it were that simple. Although the UAE has considered fortifying its own data centers with weapons and burying some underground, this is essentially a solution due to the dangers of Iranian attacks. Under normal circumstances, the project would end up costing more, as it would require creating a strong shell capable of supporting the weight of all the ground above it. This doesn’t even take into account drilling, insulation and waterproofing. Then, these facilities wouldn’t be easy to scale because they would require additional blasting or drilling, meaning companies are unlikely to invest in something that will make people happy, at extra cost and loss of scalability.

But the biggest challenge? Maintain good air quality. When you’re underground, air-related complications can add up quickly. As such, the data center would require expensive systems to achieve the same environment that is practically given on the surface, so add the expensive humidity control and ventilation system to an already hefty bill. Ultimately, it’s far more feasible than creating orbital data centers, but far too tricky for companies to seriously consider.

The advantage of underground data centers, at least in theory, is that they would save operators money on cooling because there is no sunlight. They could also use geothermal energy to reduce electricity waste. It really does seem like a magic bullet when you lay it all out – if only the construction costs weren’t so high. Well, new AI data centers can be built inside existing underground structures, which automatically solves most of the problems, and we have real-world examples that prove it works well.

For example, there are a few data centers operating in abandoned mines in the United States, making them extremely secure by design. The best part is that they don’t cost much to build and maintain. In addition, the temperature in the mines, and especially the humidity, sometimes borders on perfection. Unfortunately, few mines are both close to a power source and already have the necessary connections. So it’s not like corporations can move into the mines en masse, Fallout style.

In a perfect world, profits would be secondary to preserving our natural resources. Ultimately, the reason there isn’t a lot of effort put into finding better solutions is that it simply costs more than just taking land and building a data center. Additionally, the alternatives we have so far are all difficult to implement. Data centers aren’t being built in Antarctica because it’s too remote and doesn’t have a power grid, despite the naturally fresh air seeming like a perfect solution for the AI ​​boom. Until we figure something out, unfortunately we will all probably have to deal with the effects of new facilities popping up.