Sam Altman Claims 38,000 ChatGPT Queries Use Same Water as Producing One Almond

OpenAI CEO Dismisses AI Water Consumption as Overblown

Sam Altman has sparked debate by claiming that AI water consumption concerns are vastly exaggerated, stating that 38,000 ChatGPT queries consume the same amount of water needed to produce a single almond in California1

1

. Speaking on the Sources podcast with Alex Heath, the OpenAI CEO addressed what he called a "robust meme" about ChatGPT's environmental impact, arguing it doesn't hold up to scrutiny3

3

. Altman acknowledged he was citing the figure from memory but maintained it was accurate, contrasting it with viral claims that a single ChatGPT query uses as much water as running a shower for six hours2

2

.

The Math Behind the Almond Comparison

Researchers estimate that producing a single almond requires approximately 1.1 gallons of water, making almonds one of the most water-intensive crops grown in the United States1

1

. Using Altman's 38,000-query estimate, each ChatGPT query would consume roughly 0.11 milliliters of water, equivalent to about two laboratory drops2

2

. However, this figure contradicts Altman's own previous statements. In his June 2025 blog post "The Gentle Singularity," he wrote that a typical ChatGPT query uses about 0.32 milliliters of water, which would translate to roughly 11,000 queries per almond, not 38,0002

2

. The average American consumes 2.3 pounds of almonds annually, making almond production a significant water consumer in California, a state prone to droughts5

5

.

Modern Data Centers Versus Evaporative Cooling Methods

Altman emphasized that modern data centers have abandoned evaporative cooling methods, which consumed massive amounts of fresh water by releasing cool water vapor into the atmosphere1

1

. Today's facilities use closed-loop systems where water or specialized coolant circulates through sealed pipes, losing virtually no liquid to evaporation2

2

. According to Altman, a modern very large data center uses the equivalent amount of water as an office building for sinks and toilets1

1

. Nvidia has developed liquid cooling systems that run hotter than a hot tub, cutting water use by 100%, while Microsoft employs closed-loop cooling systems using the same amount of water as a restaurant1

1

. Amazon claimed its data centers consume only 0.075% of the water Americans use for lawns and gardens1

1

.

Research Shows Higher Water Footprint Numbers

A widely cited study led by researchers at the University of California, Riverside and the University of Texas at Arlington painted a different picture of AI's environmental footprint. Published in Communications of the ACM in 2025, the study titled "Making AI Less Thirsty" estimated that a brief conversation of 10 to 50 responses from a large language model like GPT-3 consumed roughly 500 milliliters of water, about one standard plastic water bottle2

2

. This translates to 10 to 50 milliliters per response, potentially hundreds of times higher than Altman's estimate2

2

. The discrepancy stems partly from what each measurement counts. Altman's figure likely reflects only on-site cooling water consumed directly at the data center, while the University of California, Riverside study includes water consumed at power plants to generate electricity for those servers, known as scope-1 and scope-2 water use2

2

. Independent analyses suggest the electricity-generation layer can account for up to 75% of the total water footprint behind a single AI query2

2

.

Real-World AI Data Center Water Concerns Continue

Despite Altman's reassurances, multiple data centers across the United States have faced criticism over resource usage. In Fayette County, Georgia, a facility consumed 29 million gallons of water over 15 months1

1

. This spring, residents of Fayetteville, Georgia, discovered a new 6.6 million-square-foot data center had guzzled up 30 million gallons of water, leaving residents with weak water pressure during a drought5

5

. In Morgan County, Georgia, residents complained that their water turned muddy brown, allegedly due to a nearby Meta data center development1

1

. Training AI models in older facilities can use far more water, with estimates suggesting GPT-3 consumed almost 185,000 gallons during training with older systems1

1

. Neighborhoods around the country have been revolting against the rapid buildout of massive data centers, complaining of noise, light and water pollution5

5

.

Growing Demand Raises Future Questions

ChatGPT's popularity continues to surge, with OpenAI reporting 2.5 billion user prompts daily as of last summer5

5

. The increasing popularity of AI agents, which solve more complex software problems and use significantly more tokens than chatbots, means data centers could consume more water compared to current levels1

1

. AI agents are being deployed across industries including tech, finance and retail5

5

. Social media users have responded to Altman's almond comparison by creating memes, with "Sam Almond" becoming a viral topic4

4

. Questions around AI-related water usage have become an increasingly hot topic on social media, though tracking exactly how much water AI data centers use remains extremely difficult4

4

. Research reveals that despite growing demand for water related to data centers, other industries like agriculture still top AI in terms of water consumption4

4

. Watch how tech companies balance AI expansion with environmental responsibility as the industry faces mounting pressure to demonstrate sustainable practices.