OpenAI CEO Sam Altman has pushed back against widely circulated claims about the water consumption associated with ChatGPT, arguing that viral estimates overstate the amount of water required to operate modern AI infrastructure.

Speaking on the Sources Podcast, Altman addressed claims that running one ChatGPT query consumes as much water as leaving a household shower running for six hours. He described that comparison as inaccurate and offered a different estimate, while acknowledging that he was recalling the figure from memory.

According to Altman, roughly 38,000 ChatGPT queries could correspond to the amount of water used to produce a single almond in California.

He stressed that he did not have the precise calculation available during the discussion and said the figure could be incorrect.

Altman questions the six-hour shower comparison

The water consumption of AI data centres has become a subject of public debate as demand for computing infrastructure grows.

Altman said the comparison between an individual ChatGPT query and several hours of shower use has gained traction online despite, in his view, lacking a sound basis.

He also pointed to the water footprint of everyday products, using almonds as an example. His argument was that consumers may overlook water consumption associated with familiar products while focusing heavily on AI infrastructure.

Modern data centres have changed cooling systems

Altman also distinguished current data centres from older facilities that relied heavily on evaporative cooling.

According to him, earlier data centre designs could consume significant quantities of water through evaporation, but modern facilities have adopted different cooling approaches.

He compared the water consumption of a large modern data centre to that of a conventional office building, particularly when accounting for water used by people for sinks, toilets and other standard facilities.

Altman said the idea that modern data centres routinely consume extraordinary amounts of fresh water has become difficult to challenge because the claim is widely repeated online.

AI infrastructure faces growing resource scrutiny

The debate comes as technology companies continue expanding data centre capacity to support AI services.

Water use is one part of the broader resource discussion surrounding these facilities, alongside electricity demand and the infrastructure required to operate increasingly powerful computing systems.

Altman's comments challenge a specific viral comparison, but he also acknowledged that the exact figures depend on the technology and infrastructure involved. His estimate of 38,000 queries per California almond was presented as a recollection rather than a definitive calculation.

The discussion highlights the difficulty of assigning a single water-use figure to an AI query when data centre cooling systems, location and operating conditions can vary significantly.