Sam Altman claims ChatGPT’s water footprint is as small as ‘one fifteenth of a teaspoon’

OpenAI CEO Sam Altman has claimed that a single ChatGPT query consumes about 0.000085 gallons of water, or “roughly one fifteenth of a teaspoon” in a new blog post outlining his broader views on the future of artificial intelligence.

The figure, which Altman presented without citing a specific source, was shared as part of an attempt to contextualise the energy and resource demands of large language models like ChatGPT. “People are often curious about how much energy a ChatGPT query uses; the average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes,” Altman wrote. He added that “the cost of intelligence should eventually converge to near the cost of electricity.”

Altman’s remarks come at a time of heightened scrutiny over the environmental footprint of AI technologies. Studies have increasingly raised concerns about the carbon emissions, power usage, and water demands of AI systems, particularly as models grow in size and scale.

Earlier this year, a research paper forecast that AI could outpace Bitcoin mining in energy consumption by the end of 2025. A 2023 investigation by The Washington Post and academic researchers found that generating a 100-word email using GPT-4 required “a little more than 1 bottle” of water. The actual water usage, they noted, often varies significantly depending on datacentre cooling systems and geographic location.

Water consumption is a particularly complex issue tied to the need to cool massive data infrastructure, often relying on evaporation-based systems in water-stressed regions. AI training runs and inference queries contribute to both direct and indirect water use. 

Altman’s attempt to quantify the cost per query could be viewed as a move to offer transparency or to minimise perception of AI’s real-world environmental impact.

As pressure mounts on the industry to account for its ecological footprint, the lack of transparent, verifiable data from major AI companies remains a growing concern among environmental researchers and policy experts.