Sam Altman Defends AI’s Environmental Impact, Dismisses Water Usage Claims
OpenAI CEO Sam Altman has addressed concerns about the environmental impact of artificial intelligence (AI), specifically pushing back against claims that AI is driving a surge in water consumption. While acknowledging the growing energy footprint of AI as its use scales globally, Altman asserts that widely circulated claims about water usage per AI query are inaccurate.
Water Usage Claims “Completely Untrue”
Speaking with The Indian Express during his visit to India for the AI Impact Summit, Altman stated that claims regarding water usage are “completely untrue” and based on outdated assumptions about data center cooling methods. “Water is totally fake. It used to be true. We used to do evaporative cooling in data centres, but now we don’t do that,” he explained. He dismissed online claims suggesting that a single ChatGPT query consumes as much as 17 gallons of water as “completely untrue, totally insane, [and having] no connection to reality.”
Energy Consumption: A Valid Concern
Altman’s comments come amidst a broader debate about the environmental impact of large-scale AI systems, particularly as data centers expand and electricity demand increases to support the training and operation of increasingly complex models. While dismissing the water-related concerns, Altman conceded that energy consumption is a “fair” concern, especially when considering the aggregate impact of widespread AI use.
“It’s fair to be concerned about the energy consumption — not per query, but in total, because the world is now using so much AI,” Altman said, emphasizing the need to accelerate the transition to nuclear, wind, and solar power.
He also refuted comparisons equating a single AI query to significant energy consumption, such as claims that a ChatGPT prompt uses the equivalent of multiple smartphone battery charges, stating, “There’s no way it’s anything close to that much.”
AI Training vs. Human Training
Altman argued that discussions surrounding the energy cost of training AI models are often presented unfairly, particularly when not compared to the energy and resources required to “train” a human being. “People talk about how much energy it takes to train an AI model. But it also takes a lot of energy to train a human,” Altman said. “It takes like 20 years of life and all of the food you eat during that time before you get smart.”
He suggested that a more relevant comparison is the energy required for an AI system to answer a question versus a human performing the same task, implying that AI may already be comparable in energy efficiency.
India’s AI Ambitions and Energy Context
Altman’s remarks were made as Indian policymakers increasingly link AI growth to questions of power and infrastructure. At the AI Impact Summit, Union IT minister Ashwini Vaishnaw highlighted that AI is both energy- and resource-intensive, and meeting future compute demands will require a substantial expansion of clean power and supporting infrastructure. CNBC reported on this discussion.
Vaishnaw noted that India is in a relatively strong position, with over 51% of its installed power generation capacity coming from clean energy sources. He also indicated that nuclear power is being considered alongside continued investments in renewable energy to support the growing compute requirements of AI systems. The government is actively working to expand clean energy capacity and collaborate with startups to address infrastructure constraints and support AI-led growth.