Sam Altman dismisses AI water claims and compares model training to raising humans

Sam Altman’s latest contribution to the AI-energy debate is a rhetorical move that would make any utilities lobbyist proud: stop counting data-center electricity and start counting human calories.

Speaking at an event hosted by The Indian Express, the OpenAI CEO argued that claims about ChatGPT’s water footprint are “totally fake,” and insisted that common comparisons — such as a query consuming the equivalent of multiple iPhone charges — are wildly off. TechCrunch reports that Altman acknowledged overall energy consumption is a legitimate concern “in total,” but said the fair comparison is not model training versus a single query; it’s AI’s energy use versus what it takes to “train a human,” i.e., “20 years of life and all of the food you eat” plus the evolutionary history of humanity.

When your industry is being accused of straining grids, you reframe your load as a philosophical inevitability. The catch is that power systems do not bill in metaphors. They bill in peak demand, transformer capacity, and interconnection queues.

Communities are increasingly responding in the only language that matters to land-use politics: “not here.” Business Insider reports that New Brunswick, New Jersey, just removed data centers from the list of permitted uses in a redevelopment plan after residents and environmental activists showed up at a city council meeting to oppose even the possibility of a future facility. The city planner said the data-center option was meant to “diversify” commercial development but was “not critical.” Council voted to nix it; the crowd celebrated.

The New Brunswick fight is part of a broader pattern. Business Insider cites its earlier reporting that more than 1,200 data centers had been approved for construction across the U.S. by the end of 2024, with potential electricity demand comparable to entire states and substantial water use in drought-prone regions. Residents in other towns — from Oklahoma to Texas — have packed hearings, protested, and in at least one case seen police arrest a speaker for overrunning a time limit.

Altman’s “humans use energy too” line is, in effect, a bid for social license: if AI is framed as merely a more efficient substitute for human cognition, opposition becomes anti-progress. But local governments are not deciding whether AI is metaphysically efficient. They are deciding whether to allocate scarce grid capacity, tolerate noise and diesel backup generators, and accept industrial-scale buildings in residential-adjacent areas.

There is also a transparency problem: as TechCrunch notes, there is no legal requirement for tech companies to disclose energy and water use, leaving researchers to estimate impacts from the outside. That opacity is convenient when you’re selling the future — and maddening when you’re asking a town to absorb the infrastructure.

The emerging conflict is less about “misinformation” than about bargaining power. AI companies want to move fast and plug in. Communities want enforceable numbers, not evolutionary poetry.