Sam Altman: humans take a lot of energy to train, too

Whenever critics sound the alarm on AI's staggering appetite for electricity and water, Silicon Valley always finds a novel way to frame the debate. This week in India, OpenAI CEO Sam Altman brought a rather cosmic defense to the table: humans aren't exactly cheap to run either.

Speaking at an event hosted by The Indian Express, Altman argued that the current public discourse around ChatGPT's environmental footprint is fundamentally unfair. In his view, critics keep comparing the massive energy required to train a frontier model against the tiny calories a human burns to answer a single question.

Raising a human takes twenty years of groceries

Altman pointed out that producing a capable human mind requires decades of heavy biological investment.

It takes around twenty years of life and all the food consumed during that time before someone actually gets smart, Altman explained. On top of that, you have to account for the evolutionary history of roughly 100 billion people who learned how to dodge predators and invent science just to produce a modern human.

Because of that, Altman argued the only fair benchmark is inference versus human thinking. Once a model is fully trained, how much juice does ChatGPT burn to answer a query compared to a person? Measured that way, he claimed AI has probably already caught up on an energy efficiency basis.

Busting the water myths and iPhone metrics

Altman also pushed back aggressively against popular viral statistics about AI consumption.

He called internet claims that a single ChatGPT query consumes 17 gallons of water totally fake and disconnected from reality. While he acknowledged that evaporative cooling in older data centers used to make water consumption a genuine problem, he insisted modern facilities no longer operate that way.

When the interviewer brought up a metric previously discussed with Bill Gates—that a single prompt burns the equivalent of 1.5 iPhone battery charges—Altman dismissed it outright, saying the real consumption is nowhere near that high.

Still, Altman conceded that worrying about aggregate power use is entirely fair. Because the whole world is adopting AI at once, total demand is skyrocketing, which he believes requires an urgent global pivot toward nuclear, wind, and solar energy.

Why it matters

Comparing server clusters to a child's dinner table is classic Sam Altman rhetorical flair, but in our view, it conveniently sidesteps the actual friction on the ground.

Human metabolism does not strain regional power grids or drive up residential utility bills. High-density data centers do. Without mandatory reporting requirements forcing tech giants to disclose verifiable water and power metrics, researchers are left making educated guesses while communities near server farms deal with rising power costs.

For users and developers, this rhetoric confirms that the cost per query might be falling, but the aggregate footprint will keep expanding as agentic workflows take over. The real bottleneck for OpenAI and its rivals is no longer just securing GPUs—it is securing dedicated clean megawatts, which is why nuclear power has suddenly become the tech industry's hottest new side hustle.