As resistance mounts against the ongoing construction of data centers across the United States, a common topic of discussion is the impact that data centers have on the prices and availability of local utilities, particularly water supplies. AI data centers in particular tend to be extremely thirsty, with most requiring enormous amounts of water in order to keep their various components cool and functional. This is how Amazon’s data centers managed to consume more than 2.5 billion gallons in a single year.
While all data centers require cooling, simpler data centers that serve only telecommunications purposes can get by with air cooling from large fans (although large fans are not particularly energy efficient). It’s specifically AI data centers that get hot enough that air won’t shut them down, and so water-based cooling methods are effectively the only viable options. Of course, using water to cool the many servers and components in these facilities isn’t as simple as simply dumping buckets of water down the halls, and some cooling methods use more water than others. Still, almost all require regular injections of fresh water, with a single email written by ChatGPT consuming more water than expected.
Water is almost always needed, but it can be used in different ways
Make more antennas/Shutterstock
Cooling data center equipment is no different than cooling the components of your own PC. Water is used to absorb ambient heat generated by servers and GPUs, dissipating it safely and preventing equipment from overheating.
One type of liquid cooling is evaporative cooling, a simple system used by a large number of data centers. It uses hot water to siphon heat from the components, then sends the water to a cooling tower to evaporate into the air. Of course, once water evaporates, it is gone and more water must be brought in to replace it, leading to increased costs and water consumption.
Another type of liquid cooling is closed-loop cooling, in which water circulates through closed pipes, capturing and removing heat from components as it passes. This type is more sustainable, as it uses much less water than evaporative cooling, although it is also much harder and more expensive to install en masse and can use more electricity overall.
It is technically possible for even an AI data center to avoid the need for water-based cooling, but only if the center itself is built in an ideal location. For example, data centers in exceptionally cold locations, such as the atNorth ICE03 data center in Akureyri, Iceland, can bring in cold air from outside to provide temperature regulation comparable to or better than liquid cooling. However, this approach still requires a lot of power to spin the fans. Additionally, research from the University of Hawaii suggests that increasing levels of ambient heat and humidity around the world could one day threaten this particular cooling pathway.
