Somewhere between the hand-wringing about AI’s environmental footprint and the very real strain that new data centers can put on local utilities, the discourse has gotten… weird. And OpenAI CEO Sam Altman seems done playing defense against the most viral versions of it.
Speaking at the India AI Impact summit in an interview with The Indian Express, Altman dismissed the internet’s splashiest water-use claim—specifically the idea that ChatGPT somehow guzzles “17 gallons of water per query”—calling it “completely untrue” and “totally insane.” He also argued that water-related concerns were more applicable back when OpenAI relied on evaporative cooling, while stopping short of offering a full, detailed accounting of current energy and water usage across its infrastructure.
That doesn’t mean water use is a non-issue for the industry at large. Many data centers still consume enormous amounts of water for cooling, and the expansion of AI infrastructure is colliding with local resource constraints in ways that have already sparked public backlash in multiple regions. What is changing is the engineering: Microsoft, for example, has promoted a newer data center design that uses zero water evaporation for cooling (though sites may still use water for other facility needs).
Altman’s bigger concession was on electricity. He acknowledged that concern about the energy consumed by data centers is “fair” in a world increasingly built around AI services—and suggested the grid needs to move faster toward low-carbon generation, including nuclear as well as renewables like wind and solar.
The “iPhone battery” claim, and Altman’s pushback
In the interview, Altman was also asked about a claim attributed to Bill Gates that a single ChatGPT query uses the equivalent of 1.5 iPhone battery charges. Altman flatly rejected it: “There’s no way it’s anything close to that much.”
This is the recurring problem with AI footprint arguments: they often get boiled down into sticky, shareable numbers that are hard to trace, easy to misinterpret, and frequently detached from how data center accounting actually works. The result is less “informed debate” and more “meme math.”
The rhetorical pivot: training AI vs training humans
Then Altman went for a comparison that predictably lit up the internet. People often focus on how much energy it takes to train AI models, he said—but “it also takes a lot of energy to train a human.”
His point wasn’t a scientific equivalence so much as a reframing: rather than fixating on training runs, he argued the more relevant comparison is the marginal cost of answering a question once the model is trained—AI vs a human doing the same work. In that framing, Altman suggested AI is probably already competitive on energy efficiency per answer.
That’s a slippery argument—because it depends on what you’re comparing, what you’re counting, and what “efficiency” even means (time-to-answer, energy-to-answer, accuracy-per-watt, or something else). But it’s also a reveal: OpenAI’s leadership wants the public conversation to move away from headline-grabbing training costs and toward what AI looks like as day-to-day infrastructure.
The bigger picture: data centers are already a major power consumer
Even without getting into the weeds of OpenAI’s specific footprint, the macro trend is clear: data centers are becoming a larger and more visible piece of global electricity demand.
The International Energy Agency estimates global data center electricity use is around 415 TWh, roughly 1.5% of worldwide consumption, and projects it could roughly double by 2030 to around 945 TWh—just under 3%—with AI a major driver of that growth.
In the U.S., a Department of Energy analysis (with Lawrence Berkeley National Laboratory) projects data centers could reach 325–580 TWh by 2028, which would represent roughly 6.7% to 12% of total U.S. electricity demand—depending on how the buildout accelerates.
So yes: the energy questions are real. The water questions are real. But the loudest numbers circulating online aren’t always.
Altman’s frustration seems aimed at that gap—between legitimate scrutiny and viral claims that don’t hold up. Whether that’s a productive reset of the conversation, or simply a more aggressive PR posture from one of AI’s most influential CEOs, depends on what comes next: clearer reporting, better measurement, and far more transparency from the companies building this new layer of infrastructure.