With every new technology comes horror stories about dangers, energy or water consumption, and the negative impact it will have on children, young people, or society in general. In other words, every new technology leads to the downfall of civilization, and if we don’t do something about it here and now, all hope for humanity is lost.

ChatGPT and all other language models immediately fueled fears about the energy and water consumption generated by each query. However, these horror scenarios turn out to be greatly exaggerated. And this seems to be systematic, because they were simply put out there and peddled without any figures to back them up.

A few weeks ago, Google published a study presenting the figures for the average energy consumption of a Gemini query: 0.24 watt hours of electricity and 0.26 milliliters of water. By way of comparison, a teaspoon holds about 2.5 milliliters, meaning that the response to an AI query consumes just under a tenth of a teaspoon of water for cooling.

Extrapolated to total water consumption, current data centers in the US consume as much water as a city with 122,000 inhabitants. This means that data centers consume almost exactly the same amount of water as crude oil processing (see chart), but far less than many other industrial sectors.

Data centers also account for 3.3 percent of the water consumption of all golf courses or 8 percent of the steel industry in the US. Even if, as predicted, water consumption for AI data centers triples by 2030, it will still be almost negligible compared to other industries.

However, it becomes interesting when you compare the tax revenues generated by these industries. Data centers generate 50 times as much tax revenue as golf courses, even though they consume only one-thirtieth of the water.

So if you want to get upset, then please do so about golf courses or the industry of your own hatred.

The graphics are from Andy Masley’s blog titled The AI water issue is a fake, and the water consumption data is from Berkeley Labs’ report 2024 United States Data Center Energy Usage Report.