Sam Altman Fires Back: ‘Training Humans Uses More Energy Than AI’—Here’s Why It Matters

OpenAI chief executive Sam Altman is once again in the spotlight after mounting a spirited defense of the massive energy and water footprint required to run generative-AI models, insisting that “it also takes a lot of energy to train a human.” Speaking on 23 February at an Indian Express-hosted tech summit in New Delhi, Altman dismissed criticism that data-center cooling systems are depleting local water supplies, calling some of the concerns “fake” and arguing that modern facilities rely far less on evaporative cooling than critics claim. He framed the debate as a broader question of cost–benefit, suggesting that advanced AI “will pay society back many times over” in productivity and scientific breakthroughs. Altman’s remarks come as environmental groups publish alarming estimates about the hidden water cost of training large language models; one recent study calculated that a single cutting-edge AI run can consume millions of liters of fresh water for server cooling. While the executive acknowledged that OpenAI’s systems are “not free,” he claimed that focusing solely on resource draw “misses the big picture of what AI returns to humanity.” The comments follow a week of escalating scrutiny across Silicon Valley. Earlier in February, Altman publicly ridiculed Elon Musk’s suggestion that placing data centers in low-Earth orbit could solve ground-based energy constraints, calling the plan “ridiculous” and “nowhere near commercial reality.” Industry analysts note that while satellite-powered data hubs might sidestep local zoning fights, they would introduce staggering launch costs and latency issues. Environmental activists counter that Altman’s human-versus-machine analogy is a false equivalence. “Humans don’t scale linearly the way model sizes do,” said Priya Verma, a researcher with the Center for Sustainable Computing, adding that the sector’s average data-center electricity demand has spiked more than 15 % year-over-year since ChatGPT’s debut. Regulators in the European Union and several U.S. states are already drafting disclosure rules that could force AI developers to publish annual water and carbon-emissions audits. Despite the controversy, investors appear unfazed. OpenAI is reportedly courting a new funding round that would value the firm at roughly $150 billion, banking on enterprise demand for its GPT-5 series and a forthcoming multimodal assistant tailored to corporate knowledge bases. Analysts at Bernstein predict that if usage growth continues, AI data-center power needs could equal those of a mid-size European country by 2028—an outlook that makes the efficiency debate more than academic. For now, Altman is betting that rapid improvements in chip design, immersion cooling, and clean-energy procurement will quiet the backlash. “We’ve cut cost per compute cycle by an order of magnitude in just a few years,” he told the Delhi audience, hinting that OpenAI plans to publish fresh sustainability metrics later this quarter. Whether those numbers will satisfy critics—or simply add fuel to a swelling conversation about AI’s planetary footprint—remains to be seen.