3 Minutes
A viral stat — 'each ChatGPT query uses 64 liters of water' — has been shared so often it sounds like fact. It isn’t. Sam Altman, OpenAI’s CEO, pushed back hard on that figure while speaking at an event hosted by The Indian Express in India, calling those water-consumption claims 'baseless' and misleading.
Short, sharp point: the old story about massive water use came from a time when some data centers relied on evaporative cooling. That technique does use water. But it’s largely been phased out in many modern facilities. Altman says the numbers floating online ignore how infrastructure and cooling strategies have changed — and they paint a distorted picture of what systems like ChatGPT actually cost in water.
Does that mean there’s no environmental concern? Not at all. Altman conceded that energy use is a fair and growing worry. The scale matters. One query might use a fraction of a kilowatt-hour; global, 24/7 demand for AI is rising fast. The cumulative energy load — not isolated per-question estimates — is the metric that should worry policy makers and operators.
Then he threw in a provocative comparison. How do you judge the energy cost of intelligence? Humans take decades to learn. Altman pointed out that a person consumes roughly 20 years’ worth of food and metabolic energy to reach mature cognitive ability. He also framed human cognition as the product of evolutionary effort spanning roughly 100 billion people who have lived on Earth — a slow, diffuse, and costly process.
Altman’s conclusion: if you want a fair comparison, measure the energy required for a trained AI model to answer a question versus the energy a human uses to answer the same question. By that yardstick, he suggested, AI could be more efficient — but only if the power behind it is low-carbon.
So what’s the fix? Faster deployment of clean energy. Altman urged a rapid shift to nuclear, wind, and solar so the world can accommodate rising electricity demand without deepening the climate crisis. That’s not an optional add-on, he argued; it’s the backbone of responsible AI scale-up.
One more wrinkle: there’s currently no global legal requirement for tech firms to publish precise, audited water and energy footprints. Independent researchers try to model impacts using indirect estimates, and in some regions data centers have contributed to higher electricity prices and stress on distribution grids.
Debate and transparency will keep pace with deployment. But if we’re serious about sustainable AI, the conversation should move from sensational water-scoop headlines to real questions about clean power, infrastructure investment, and public reporting.
Leave a Comment