ChatGPT water use is less than you think, says Sam Altman
&imwidth=600&imheight=450&format=webp&quality=medium "ChatGPT water use is less than you think, says Sam Altman")
ChatGPT water use is less than you think, says Sam Altman
Sam Altman has challenged concerns about ChatGPT’s water consumption by comparing AI queries with California almonds. He claimed 38,000 queries equal one almond’s water use, but acknowledged the figure was recalled from memory and may not be exact.
OpenAI CEO Sam Altman has pushed back against growing concerns over the amount of water used by artificial intelligence, making a surprising comparison between ChatGPT queries and almonds.
Speaking on the first episode of the Sources podcast, Altman claimed that around 38,000 ChatGPT queries use as much water as producing one almond in California. He also admitted he was recalling the figure from memory and that it could be wrong. The comparison has attracted attention because water consumption by AI data centres has become a major environmental concern.
How much water does one ChatGPT query use?
Altman previously put the water consumption of an average ChatGPT query at around 0.32 millilitres.
Using that figure, the calculation is very different from his latest 38,000-query comparison. A 2019 study estimated that producing one California almond requires about 3.56 litres of water. At 0.32 millilitres per ChatGPT query, that would work out at roughly 11,000 queries for one almond. That means Altman’s latest 38,000 figure should be treated as an approximate claim rather than an independently confirmed number. Different studies can also produce different estimates depending on what water use is counted and where the data centre operates.
Sam Altman says AI water fears are overstated
Trending Stories
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
Altman argued that the public perception of AI data centres as huge water consumers has become a “robust meme” that is difficult to correct.
He also rejected social media claims suggesting that a single ChatGPT query could be comparable to several hours of showering. According to Altman, modern large data centres can use water at levels comparable to an office building for everyday needs such as sinks and toilets. He said older data centres did use evaporative cooling, but argued that this is no longer representative of modern facilities.
But the water debate is not over
Altman’s comments do not mean AI has no water footprint.
Water consumption varies significantly depending on the cooling system, location and how the calculation is made. Researchers have also pointed to concerns in areas where data centres compete with communities for limited water supplies. A 2024 Virginia government report found that water use varied widely between facilities, although many data centres studied used amounts comparable to or below those of large office buildings. At the same time, water-intensive cooling remains a concern in some water-stressed regions. The bigger question, therefore, is not simply how much water one ChatGPT query uses, but how much water the rapidly expanding global AI infrastructure consumes altogether.
