Sam Altman Defends AI’s Energy Use, Says Training Humans Is Even Less Efficient

OpenAI’s chief executive used the AI Impact summit in New Delhi to pivot the debate from the carbon cost of training large language models to the far larger, albeit less visible, energy bill of human development. Citing the Guardian AI interview, Altman argued that “it takes about 20 years of life—and all the food you consume during that time—before you become smart,” and that the cumulative evolutionary effort of roughly 100 billion people who survived predators and built scientific knowledge should be counted when measuring efficiency (The Guardian AI). By framing AI’s training phase as a drop in the evolutionary ocean, he suggested that critics were “over‑doing” the concern, even as he acknowledged that the public’s perception of AI’s power use is “fair” (The Guardian AI).

Altman’s remarks arrived amid mounting data on the sector’s electricity appetite. The International Energy Agency (IEA) estimates datacenters consumed about 1.5 % of global electricity in 2024, and projects a 15 % annual growth through 2030—four times faster than the rise in all other sectors combined (The Guardian AI). MIT climate‑impact fellow Noman Bashir warned that the rapid construction of new facilities forces “the bulk of the electricity to power them” to come from fossil‑fuel plants, a trend that could outpace the rollout of clean‑energy capacity (The Guardian AI). In December, more than 230 environmental groups called for a moratorium on U.S. datacenter construction, citing the “largely unregulated rise” that is already disrupting communities (The Guardian AI).

Altman did not dismiss the need for greener power, however. In the same interview he stressed that “we need to move toward nuclear or wind and solar very quickly,” echoing a broader industry consensus that AI’s future hinges on a breakthrough in clean energy supply (The Guardian AI). He also downplayed specific energy‑use estimates for individual queries, rejecting Bill Gates’s claim that a single ChatGPT request consumes the equivalent of 1.5 iPhone battery charges as “not close to that” (The Register). While Altman’s dismissal of the “water is totally fake” comment—referring to closed‑loop liquid cooling systems that avoid traditional evaporative cooling—was more rhetorical than technical, it underscored his view that the sector’s infrastructure is already moving toward higher efficiency (The Register).

The debate over AI’s carbon footprint is not confined to public statements. Reuters reported that Altman, speaking at Davos earlier this year, framed the energy challenge as a “breakthrough” problem, suggesting that the industry’s trajectory will be determined by the pace of clean‑energy innovation (Reuters). TechCrunch echoed the sentiment, noting that Altman repeatedly reminds audiences that “humans use a lot of energy” and that AI should be judged against that baseline (TechCrunch). Yet analysts remain skeptical that the analogy will sway policymakers. The IEA’s growth forecast implies that, even with aggressive decarbonisation, AI‑related demand could add hundreds of terawatt‑hours to the grid by 2030, a scale that dwarfs the per‑inference energy figures Altman disputes (The Guardian AI).

In short, Altman’s defense reframes AI’s energy use as a modest cost in the grand narrative of human evolution, while simultaneously urging a rapid transition to low‑carbon power sources. Whether that rhetorical shift will translate into concrete policy or investment changes remains to be seen, but the juxtaposition of evolutionary time‑scales with annual electricity growth curves highlights the growing tension between AI’s commercial momentum and the planet’s climate limits.