US voters this week approved 40 of 41 local ballot measures placing limits on AI data centres, with water and power the main concerns, but measuring how much water each site uses remains difficult, CNBC reports. Operators usually publish figures for the whole company rather than for individual sites. University of California, Riverside researcher Shaolei Ren estimates that the largest one-gigawatt AI facilities, such as those planned by Meta, Google, Amazon and SpaceX, can use up to 10 million gallons a day, though the real figure depends heavily on hardware, location and cooling method. Modern AI chips run so hot that liquid cooling is needed, but most large sites use closed-loop systems that consume relatively little water day to day. Experts say the complexity has fuelled misleading claims on both sides of the debate.
Source: CNBC