Sam Altman is comparing ChatGPT’s water use with almonds as public concern over the environmental cost of artificial intelligence grows alongside the rapid expansion of data centres. Speaking on the Sources podcast, the OpenAI CEO argued that claims about AI consuming enormous quantities of water are exaggerated. He said that, based on a figure he was recalling from memory, around 38,000 ChatGPT queries use the same amount of water involved in producing one California almond. Altman acknowledged that the exact number could be wrong, but said it was close to the real figure.

A 2024 Virginia government report found that most data centres use as much or less water than large office buildings, although some consume significantly more. However, water-intensive cooling systems such as evaporative cooling can worsen shortages in parts of the western US.

Why Altman Brought Almonds Into the AI Water Debate

Altman’s comparison was intended to challenge the idea that every AI query places an enormous burden on water supplies. He criticised social-media claims suggesting that using ChatGPT could consume water on the scale of hours of showering, calling the perception around data-centre water consumption a “robust meme” that has been difficult to correct.

There is some context behind the almond comparison, but the numbers do not line up perfectly. Altman previously put the water consumption of a ChatGPT query at about 0.32 millilitres. A 2019 study estimated that producing a California almond required an average of 3.56 litres of water. Using those figures produces roughly 11,000 queries per almond, considerably below Altman’s 38,000 estimate.

Also Read: Global Warming Nears 1.5°C Threshold As UN Raises Fresh Climate Warning

Data Centres Still Have a Real Water Footprint

The debate is not simply about the amount of water consumed by one query.

• Data-centre water demand depends heavily on location, cooling technology, electricity sources and operating conditions.
• Some facilities use water-intensive evaporative cooling, while others rely more heavily on air-based or closed-loop systems.
• This means a single average figure can hide significant differences between facilities.

Public concern is also rising because data-centre construction is accelerating.

• A March 2026 Gallup survey found that about seven in 10 Americans opposed building AI data centres in their local area, with environmental impact and resource consumption among the concerns.

Also Read: Fashion Industry Emissions Increase 14%: Apparel Impact Institute

Why AI Companies Are Under Pressure to Use Less Water

Technology companies are increasingly publishing water-efficiency figures as scrutiny grows. Amazon reported that its global data centres withdrew about 2.5 billion gallons of water in 2025, while water use at facilities it directly owned and operated fell 2% from 2024 despite its expanding data-centre footprint. The company says its facilities use air cooling for most of the year and that its water-use efficiency has improved substantially since 2021.

Sam Altman is comparing ChatGPT’s water use with almonds to put the per-query impact into perspective, but the comparison does not settle the wider debate. The more important question is what happens when billions of queries are processed through increasingly large data-centre networks, particularly in regions where water is already scarce. Even if individual AI interactions use relatively little water, the scale and location of infrastructure remain important environmental questions.

Also Read: Coral Reef Cover Falls 9.5% Worldwide In Just Four Years, Study Finds