AI infrastructure expansion could increase global water demand by 30 trillion liters

The rapid expansion of artificial intelligence infrastructure is projected to significantly increase global water demand. An industry analysis suggests that by 2050, AI-related infrastructure could increase annual water demand by approximately 30 trillion liters, a 129% increase from baseline models.

This water usage is distributed across three primary layers: 54% is attributed to electricity production, 42% to semiconductor manufacturing, and 4% directly to the expansion of data centers. In semiconductor production, ultra-pure water is essential to remove impurities from modern chips.

While data centers require massive amounts of water for cooling, technological shifts may mitigate this impact. For instance, Nvidia has transitioned to 100% liquid cooling for its entire computing systems in certain generations, a method that can potentially reduce water consumption compared to traditional air cooling. However, researchers at UC Berkeley note that high-quality, comparable data regarding specific data center water consumption remains scarce.