'AI data center water consumption is inflated'

"There's information circulating about ChatGPT's water consumption, that each query consumes the equivalent of running a six-hour shower and that water never comes back, just lost," Sam Altman said on the Sources podcast with Alex Heath this week.

OpenAI CEO believes that the above information is incorrect. Although he couldn't calculate exactly at the program, he estimated: "About 38,000 ChatGPT queries consume the same amount of water as producing one almond in California. This is the actual total consumption, not just the amount of water used in a data center." People who "gobble 12 almonds at a time" don't usually feel like they're "doing something terrible in terms of water use," he added.

According to Altman, data centers used to use evaporative cooling systems, but have long since converted. A large, modern center now consumes as much water as an office building when employees use the sink, toilet or do other tasks.

"The view that data centers consume a lot of water is very popular and difficult to eliminate," the head of OpenAI said, saying that the AI ​​industry needs to change public thinking. "The right way to get people to like something is to give them value. The industry needs to work harder to make products easy to use and provide great value to users."

According to Business Insider, water consumption is the main concern in the wave of opposition to data centers. Gallup's survey of Americans in May showed that 71% disagreed about building data centers where they live, of which 70% were concerned about environmental impact, 50% mentioned the impact on resources, and 18% specifically mentioned water.

According to Tom's Hardware, almonds are one of the most water-consuming crops in the US. Researchers say it takes an average of 4.16 liters of water to grow one almond. Meanwhile, according to Altman's old statement, each response from ChatGPT requires 1-50 ml of water, apparently then using evaporative cooling technology. Meaning, producing one almond takes as much water as 85-5,500 ChatGPT queries.

Training AI models in legacy facilities requires large amounts of water, while new technology offers significant savings. For example, Nvidia introduced a liquid cooling system that operates in "hotter than a hot tub" conditions to help reduce the amount of water, or Microsoft introduced a closed cooling solution that uses water equivalent to a restaurant. Amazon even claims its data centers consume only 0.075% of the water Americans use for their lawns and gardens.