Sam Altman Defends AI Energy Use by Comparing It to Human Training Costs

OpenAI CEO Sam Altman has ignited a fierce debate over the environmental impact of artificial intelligence, asserting that concerns about AI’s energy and water consumption are overstated—and even "fake"—when compared to the resources required to train human beings. Speaking at an event hosted by The Indian Express in early 2026, Altman argued that the energy expended in raising and educating a single human from infancy to professional competence vastly exceeds the computational power needed to train even the largest AI models. "It also takes a lot of energy to train a human," Altman said, according to TechCrunch. "We’re talking about decades of food, heating, transportation, medical care, and education—all powered by fossil fuels and water-intensive systems. If we’re going to critique resource use, let’s be consistent."

Altman’s comments come amid growing public and regulatory scrutiny over the environmental costs of generative AI. Data centers powering large language models consume vast amounts of electricity, with some estimates suggesting that training a single model can emit as much carbon as five cars over their lifetimes. Water usage for cooling these facilities has also drawn criticism, particularly in drought-prone regions like California and India, where tech giants are expanding their infrastructure. Yet Altman dismissed these concerns as disproportionate. "The water used to cool a server farm is a fraction of what’s needed to grow the food that feeds a single child for a year," he reportedly stated, as reported by MSNBC.

While his analogy resonates with some in the tech industry, it has drawn sharp rebuttals from environmental scientists and ethicists. Dr. Elena Rodriguez, a climate policy researcher at Stanford, told TechSpot: "Comparing the energy intensity of raising a human to training an AI is a false equivalence. Humans are not machines. We don’t train billions of humans to perform the same task at scale, every week. AI models are replicated and deployed globally in minutes. The marginal cost of inference is near-zero, but the cumulative impact is not."

Moreover, critics argue that Altman’s framing ignores systemic inefficiencies. While human development is a one-time, biologically necessary process, AI training is iterative and often redundant. Multiple versions of the same model are trained for A/B testing, fine-tuning, and regional customization—each consuming significant energy. A 2025 study from the University of Cambridge found that over 70% of AI training runs produce models that are never deployed, representing a massive waste of computational resources.

Altman’s defense also sidesteps the growing pressure on tech companies to adopt renewable energy and water-recycling technologies. While OpenAI has pledged to power its data centers with 100% renewable energy by 2030, critics note that such commitments remain aspirational and lack independent verification. Meanwhile, utilities in regions hosting AI infrastructure report rising demand and strain on local water supplies, with some communities facing rationing.

Despite the backlash, Altman’s remarks reflect a broader trend in Silicon Valley: reframing environmental criticism as moral hypocrisy. By invoking the human cost of energy use, he attempts to shift the burden of proof onto critics. "If you’re worried about resource consumption," he said, "then advocate for sustainable agriculture, public transit, and universal education before you single out AI."

But as AI’s footprint expands—with estimates suggesting it could consume 8% of global electricity by 2030—many argue that technological innovation must be paired with responsible stewardship. The question is not whether humans use more energy than AI, but whether AI’s exponential growth demands new ethical and environmental guardrails. As the debate intensifies, Altman’s analogy may be rhetorically clever, but it risks obscuring the urgent need for transparency, accountability, and sustainable design in the age of artificial intelligence.