Sam Altman Addresses ChatGPT Water Usage: What 38,000 Queries and One Almond Really Mean

 September 4, 2026 — Concerns about the environmental impact of artificial intelligence are gaining renewed attention in the United States after OpenAI CEO Sam Altman discussed how much water modern AI systems use.

During a recent interview on the Sources podcast with Alex Heath, Altman pushed back against claims that individual ChatGPT questions consume huge amounts of water.

He offered a striking comparison: Altman said that roughly 38,000 ChatGPT queries use about as much water as producing a single almond in California. He also acknowledged that the figure was being recalled from memory and might not be exact. 

What Did Sam Altman Actually Say?

Altman's comments came amid growing discussion about the amount of water used by data centers that operate AI services.

He argued that some popular descriptions of AI water consumption are based on older information and do not necessarily represent how modern data centers operate.

Altman said modern large data centers can use water at levels comparable to an office building for everyday activities such as sinks and toilets, while newer cooling technologies can reduce reliance on water-intensive cooling methods. 

Does One ChatGPT Question Really Use Very Little Water?

The answer is more complicated than a single number.

AI models run on computer systems inside data centers. Those machines generate heat and require cooling. Depending on the facility, climate, cooling technology, electricity source and other factors, water consumption can vary considerably.

That is why researchers caution against treating one universal number as applying to every ChatGPT query.

CalMatters reported that researchers say the publicly available information about individual data centers is too limited to independently verify Altman's 38,000-query comparison. 

Where Did the Older 500-Milliliter Claim Come From?

Another figure frequently mentioned online comes from earlier research into AI's water footprint.

That research estimated that approximately 500 milliliters of water could be associated with a series of roughly 20 to 50 questions, depending on conditions. It was not saying that every individual ChatGPT question consumes a 500-milliliter bottle of water.

That distinction is important because the viral version of the claim can make AI's water footprint appear much larger than the original research suggested.

Why Is AI Water Usage Being Discussed Now?

The issue has become more important as the number of AI data centers grows.

Data centers can require water for cooling, and their impact depends heavily on where they are located and which cooling systems they use.

Researchers have also warned that peak water demand can matter for local communities, especially during hot periods when cooling requirements increase. 

At the same time, newer cooling technologies are being developed to reduce water consumption.

Experts Say More Transparency Is Needed

The debate isn't simply about whether Altman's number is right or wrong.

A major issue is the lack of detailed, publicly available data from individual data centers.

Researchers cited by CalMatters said factors such as location, outdoor temperature, cooling design, prompt length and computing requirements can all affect water consumption.

That means a single statistic may not accurately describe every AI query.

What Does This Mean for ChatGPT Users?

For everyday users, the discussion does not mean that sending a normal ChatGPT question is equivalent to using a bottle of drinking water.

Instead, it shows that AI has an environmental footprint that includes electricity, cooling and infrastructure.

The important question for the future may be how efficiently AI companies can operate increasingly powerful systems while reducing pressure on local resources.

The Almond Comparison Needs Context

Altman's almond comparison has attracted significant attention because almonds are known to require substantial amounts of water to grow.

Some reports estimate that producing one California almond requires roughly 1.1 gallons of water, although agricultural water footprints vary by location and methodology. 

Yahoo Tech

But experts caution that comparing agricultural water use with data-center water use isn't necessarily a simple apples-to-apples calculation.

CalMatters reported that researchers could not independently verify Altman's exact 38,000-query figure because detailed data needed for the calculation is not publicly available. 

The Bigger Question: Can AI Become More Efficient?

As AI becomes a bigger part of everyday life, efficiency is likely to become increasingly important.

Companies operating data centers are exploring different cooling approaches and infrastructure designs. Some systems can recirculate water, while other approaches can reduce direct water consumption but may require additional electricity.

This creates a broader technology challenge: how can AI companies provide more computing power while using fewer natural resources?

Bottom Line

Sam Altman's latest comments challenge the popular idea that every ChatGPT question c

onsumes a large amount of water.

His claim that 38,000 ChatGPT queries are comparable to the water used to produce one California almond is attracting attention, but it should be treated as Altman's estimate rather than an independently verified universal measurement. 

The broader environmental impact of AI is real, but the exact water footprint of an individual query depends on several factors.

As AI infrastructure expands, greater transparency about data-center water and energy use could help consumers and communities better understand the technology's environmental cost.

Editor's Note: This article was prepared using the latest available reporting and expert commentary. Water-use estimates can vary by data center, cooling technology and methodology. Figures attributed to Sam Altman are presented as his claims and are not treated as independently verified measurements.