AI Surge Accelerates the Transition to Liquid Cooling: Rethinking Energy and Water Use in Data Center Cooling

The digital world runs on massive warehouses of computers, known as data centers, which process the information we use every day. As artificial intelligence grows more powerful, these facilities are becoming incredibly hot, generating heat at a rate that traditional air conditioning struggles to manage. To keep the machines from overheating, engineers are shifting from blowing cold air over them to pumping liquid directly through the hardware, a method that carries heat away much more efficiently. However, this shift brings a new set of challenges. While liquid cooling uses less electricity, some versions of it rely on evaporating water to release that heat, which can drain local water supplies in areas that are already dry. The question facing the industry is how to cool these giant computers without either wasting massive amounts of power or draining the local rivers and aquifers.

A team of researchers set out to map out the best path forward for this transition. They built a detailed computer model to simulate how different liquid cooling systems would perform across fifty-seven cities around the world, covering every major climate type from the freezing north to the scorching deserts. Their goal was to find the sweet spot where energy use and water consumption are balanced, rather than just looking at one factor in isolation. They compared two main types of liquid cooling: one that uses fans to blow air over pipes (dry cooling) and one that sprays water to help cool the pipes (evaporative cooling). They also tested how changing the temperature of the liquid entering the computer racks would affect the total energy bill, including the electricity the servers themselves use to run.

The study found that switching to liquid cooling is a massive leap forward for efficiency. Compared to the old air-cooled systems, liquid cooling can cut the energy needed just for cooling by between 25 and 56 percent. In the most efficient scenarios, where the system uses the outside air to cool the liquid whenever possible, the total energy used by the entire facility drops to a level where almost all of the electricity goes directly to the computers, with very little wasted on cooling equipment. The researchers identified specific temperature thresholds where a data center could theoretically run entirely on free cooling from the outside air for nearly the entire year. For systems that spray water, this threshold is around 33 degrees Celsius, while systems that rely only on air can push that limit up to 49 degrees Celsius. This means that in many parts of the world, these facilities could stop using energy-intensive chillers almost entirely.

However, the researchers discovered that saving energy is not as simple as just turning the temperature up as high as possible. While warmer liquid allows for more free cooling, it can also make the computer chips themselves work harder and use more electricity. In some climates, the extra power the servers draw to handle the heat outweighs the savings from the cooling system. The team found that for every location, there is a specific, unique temperature that minimizes the total energy use for the whole building. This "total energy minimum" is often different from the temperature that makes the cooling system look best on paper. In cooler places, the best temperature is lower, because the servers don't need much help to stay cool, but in hotter places, the system can afford to run warmer to maximize the use of free outside air.

The study also tackled the water issue, showing that the choice between dry and evaporative cooling depends heavily on where the data center is built. In dry, hot regions, evaporative systems use a lot of water because the air is thirsty and pulls moisture away quickly. In these same dry places, the researchers found that dry cooling systems, which use no water at all, can actually be slightly more energy-efficient than their wet counterparts. This happens because the fans in dry systems can run slowly and quietly when the air is cool, whereas evaporative systems have a baseline energy cost just to keep the water pumps running. The findings suggest that in water-stressed areas, the best solution is often to use dry cooling with higher temperature settings, which saves both water and energy.

Ultimately, the paper argues that there is no single global rule for cooling the next generation of artificial intelligence. The most sustainable approach requires a tailored strategy that looks at the local weather, the specific type of cooling technology, and the behavior of the computer chips. By choosing the right temperature and the right system for each specific location, data centers can support the booming AI industry without exhausting the world's energy or water resources. The researchers emphasize that as chip designs improve and can handle even hotter liquids without using extra power, the potential for saving both energy and water will only grow, offering a clear path toward a more sustainable digital future.