The AI Water Usage Threat
Image courtesy of Learntek under CC0 1.0 Universal Deed, resized to 700 x 391 pixels.
Artificial intelligence (AI) is obviously expected to consume massive amounts of electricity, but unbeknownst to many, there is also an AI water usage threat looming.
The reason is that AI technology runs extremely hot, and water-based cooling systems are often used to keep servers from overheating. Additionally, the indirect water usage associated with powering these facilities contributes greatly as well. The question is – how much of a risk is this to water utilities?
Quantifying the Impact of AI Water Usage
According to a study by the Lawrence Berkely National Laboratory, data centers in the U.S. consumed approximately 17.5 billion gallons of water in 2023. The report goes on to say that it expects this water consumption could potentially quadruple in just 5 years.
When viewed nationally, the impact is modest, but there are pockets of activity across the U.S. where the impact is more than modest. In fact, the strain on local water utilities in areas where data centers are concentrated can be extreme. Unfortunately, about 66% of data centers built in the last 3 years are in high-water-stress areas.
Luckily, there are some ways to reduce AI water usage. One method is to build water storage facilities on-site at data centers to manage peak demand. Another method is to enhance water infrastructure by expanding distribution or fixing leaks to improve the ability to manage demand spikes.
Additionally, there may be better options for the evaporative cooling systems that are commonly used in the industry. These include air-based and liquid-immersion cooling systems, using recycled or gray water to reduce the amount of potable water that needs to be consumed, and even so-called closed-loop systems that in theory don’t consume any water because they simply use the same water over and over again (that said, these systems use more electricity which can increase indirect water use).
Of course, these solutions are easier said than done, which means that AI water usage will continue to be a threat to water utilities operating in data-center-heavy areas.
