Sam Altman Defends AI's Energy Use with Human Intelligence Comparison, Stirring Controversy

Sam Altman, CEO of OpenAI, took a firm stance against recent environmental criticisms targeting artificial intelligence. He dismissed alarming claims regarding AI's water consumption as exaggerated and out of touch with reality during a February 20 interview at the AI Impact Summit in New Delhi.

Altman criticized reports suggesting that ChatGPT's queries guzzle multiple gallons of water, labeling those figures as "totally insane." He cited an earlier estimate where a single ChatGPT query uses about 0.000085 gallons of water, roughly one-fifteenth of a teaspoon, though he didn't disclose the method underlying this calculation.

The water usage in AI primarily comes from evaporative cooling in data centers to prevent hardware overheating. Altman pointed out that many AI firms, including OpenAI, are moving away from traditional water-intensive methods. Instead, they favor cooling systems recycling liquid, which significantly lowers freshwater consumption. He also noted tech giants like Microsoft, Meta, Google, and Amazon have committed to replenishing more water than they consume by 2030.

Despite these shifts, data center water use remains substantial. A January report from Xylem estimated AI-related data center water consumption at 23.7 cubic kilometers in 2025-a 38% increase since 2020-and predicted it will more than triple in the next 25 years. The report further revealed that 56% of data center capacity still relies on some evaporative cooling, indicating ample room for improvement.

On the electricity front, Altman acknowledged the challenges but underscored a transition toward renewable energy sources such as nuclear, wind, and solar. He referenced a 2024 International Energy Agency report showing data centers accounted for about 1.5% of global electricity use, with consumption growing over four times faster than overall demand and expected to double by 2030.

In line with these concerns, numerous large tech players are sealing deals to power data centers with alternative energy, including emerging nuclear technologies. Altman has personal investments in nuclear initiatives like Oklo and Helion, both focused on innovative nuclear power solutions.

Perhaps the most provocative part of Altman's argument was his attempt to frame AI's energy use in the context of human intelligence development. He highlighted that training a human brain encompasses roughly 20 years of life fueled by food consumption before any meaningful 'inference' can happen. By contrast, training an AI model compresses this intensive development phase into a far shorter timeframe.

This comparison sparked pushback from commentators such as Matt Stoller of the American Economic Liberties Project, who criticized the analogy as equating a mere spreadsheet to a baby. Sridhar Vembu, Zoho Corporation's founder, emphasized on social media that AI should blend into the background rather than compete with human uniqueness, warning against equating technology with humans.

The debate raises a thorny question: Can we fairly benchmark AI's environmental impact without considering the extensive biological and developmental costs baked into human intelligence? Altman's framing certainly reframes the conversation, but whether it deflates or fuels scrutiny is far from settled.