AI data centers and cooling technologies

There's no doubt that data centers have become pivotal for society for decades, storing data for people to watch shows, use tap-to-pay, or for businesses to run databases. However, as AI development and use have grown, demand for new data centers has surged. Nowadays, AI data centers are sliding into existence, from rural Utah locations to just outside of our backyards in Northern Virginia.

Why Should We Care?

Data centers are reliant on significant electricity and water consumption for cooling purposes. This energy usage is predicted to consume up to 17% of all U.S. electricity by 2030, and water usage can go up to 5 million gallons per day, equivalent to the water usage of 50,000 people. Furthermore, the water is not reused. Roughly 80% of consumed water, mostly fresh water, evaporates in the cooling process. 

The evaporation of water is a primary concern, since cooling through evaporation leaves behind salts and other sediments, lowering the surrounding water’s quality. It also poses a concern for the environment and nearby community, since used water isn’t entering back into the community water stream.

Not only that, data center companies withdraw local water. Google, for example, reported that 31% of its freshwater usage came from medium or high water scarcity. By taking freshwater from local communities, including those that already have water scarcity, data centers are only harming them further by contributing to water usage.

Where Does Engineering Come Into Play?

Several engineering companies and researchers are in the process of optimizing cooling technologies for data centers, which would result in less water and electricity usage. 

Ferveret, a start-up started by Massachusetts Institute of Technology (MIT) researchers, is developing a specialized liquid that’s adapted from “subcooled boiling,” or using a liquid that has a low boiling point. This liquid, dubbed “Ferveret’s Adaptive Phase Cooling,” produces smaller bubbles in comparison to other liquid cooling methods. These bubbles then more frequently travel to recondense, accelerating the cooling process and preventing complications that would come up in traditional liquid cooling.

At the University of California San Diego, researchers have developed a “cold plate,” which consists of a new fiber membrane with pores that draw cooling liquid across it through capillary action. The pores are key to effective evaporation since pores that are too small clog, and pores that are too large end up boiling in the process of evaporating. Housed directly on a CPU or GPU chip, this cold plate would mitigate heat from electronics through evaporation as the liquid travels. Although this uses liquid cooling, the fiber membrane development allows for increased evaporation efficiency, harnessing the ability to cut energy use.

Although the exact methods of cooling improvement in data centers are still unknown, innovations will undoubtedly occur as research progresses. New ideas, such as closed loop liquid cooling, are already being implemented at a Microsoft AI data center, already significantly reducing water waste. With projects such as the above continuing, the future seems bright for engineering solutions for the future.