Asger Risom
Sam Altman has dismissed claims about ChatGPT’s water usage and defended AI’s electricity demands, arguing that humans themselves require vast amounts of energy to “train” over a lifetime.
OpenAI CEO Sam Altman has pushed back against mounting criticism over artificial intelligence’s growing electricity and water consumption, arguing that the way people frame the debate is often misleading.
Speaking at the India AI Impact Summit, Altman was asked directly about reports highlighting ChatGPT’s environmental footprint — particularly claims about how much water individual AI queries consume.
Water claims called ‘totally insane’
Altman dismissed suggestions that ChatGPT uses gallons of water per query as “completely untrue” and “totally insane,” according to footage shared by The Indian Express.
He said many data centers powering OpenAI’s models have shifted away from traditional evaporative cooling systems, which rely heavily on water to prevent overheating. In their place, newer facilities increasingly use alternative cooling methods designed to reduce water intensity.
However, broader industry data paints a more complex picture. A January report from water technology company Xylem and Global Water Intelligence found that 56% of data centers globally still rely on evaporative cooling in some form.
Electricity is a ‘fair’ concern
On the question of electricity, Altman struck a more conciliatory tone.
“We need to move toward nuclear, or wind, or solar very quickly,” he said, acknowledging that AI’s power demands are significant and will continue to grow as models become more advanced.
But he also argued that comparisons between AI and human intelligence are often incomplete.
“It also takes a lot of energy to train a human,” Altman said, prompting laughter from the audience. “It takes, like, 20 years of life, and all of the food you eat during that time before you get smart.”
He expanded on the analogy, suggesting that human knowledge itself is the product of thousands of years of accumulated learning across generations.
Measuring efficiency differently
Altman’s broader point was that judging AI purely on raw energy consumption per query ignores context. A more relevant comparison, he argued, would examine how much energy a trained human consumes to answer a question versus how much energy a trained AI model requires.
In a June 2025 blog post, Altman estimated that a single ChatGPT query uses about 0.34 watt-hours of electricity — roughly equivalent to running an oven for about one second. Energy use varies depending on the complexity of the request, and newer models may consume different amounts.
The environmental debate isn’t going away
Despite Altman’s defense, projections suggest AI’s environmental footprint will expand sharply over the coming decades.
The Xylem and Global Water Intelligence report estimates AI-related water consumption could grow by around 130% by 2050. Rising electricity demand from data centers is also expected to increase water use for power generation, while more advanced chip manufacturing could significantly raise water requirements during production.
OpenAI’s 800-acre data center complex in Abilene, Texas, will reportedly use a closed-loop cooling system that recirculates water, though it will initially require millions of gallons to fill the system.
As AI becomes more embedded in daily life, scrutiny over its environmental cost is likely to intensify. Altman’s stance suggests the solution lies in accelerating clean energy development rather than slowing AI progress — a position that will continue to fuel debate as the technology scales.
Sources: The Indian Express, Xylem and Global Water Intelligence report, Texas Tribune, OpenAI blog
