OpenAI CEO Sam Altman on Friday defended the resource requirements of artificial intelligence, calling concerns about data center water use "bogus" and comparing the energy consumed by AI systems to humans. In an interview with The Indian Express on the sidelines of the India AI Impact Summit, Altman was asked about common criticisms of AI — such as energy and water consumption.
The CEO responded that the statement circulating on the Internet that "ChatGPT consumes several gallons of water for each question" is "completely untrue, extremely ridiculous, and has no relation to the facts."
Data centers have traditionally used large amounts of water to cool electronic components and prevent overheating. Although data center cooling technology has promised to reduce water consumption, some new data centers are even eliminating water altogether.
But even as efficiencies continue to improve, a report released last month by water treatment technology company Xylem and Global Water Intelligence predicted that water use for cooling will more than triple over the next 25 years as computing demands rise, putting pressure on water systems.
While dismissing water concerns, Altman said energy consumption is indeed a legitimate concern for AI.
“Not per question, but in aggregate — because the world is using AI so much... We need to move very quickly to nuclear power, wind power, and solar power.”
Asked about Microsoft founder Bill Gates' previous remarks - Gates once said that the energy efficiency of the human brain proves that AI can also become more energy-efficient in the future - Altman disagreed.
"There's always something unfair in this comparison: People always talk about how much energy it takes to train an AI model... but it also takes a lot of energy to train a human being," he said. "It takes almost 20 years of life, and all the food you eat before that, for you to become smart."
He added: "A fair comparison should be: after the model training is completed, you ask ChatGPT a question, and the energy consumed by it to answer is compared with the energy consumed by a human to answer. If measured by this standard, AI may have caught up with humans in terms of energy efficiency."
The process Altman is referring to is called inference, which is using an already trained AI model to generate new output. AI inference typically consumes far less power than the training phase.
Altman's comments, particularly the contrast between AI and humans, sparked some controversy online amid growing anxiety about AI taking over human jobs.
Sridhar Venbhu, co-founder and chief scientist of Indian software company Zoho, who attended the summit, criticized this approach of equating AI with humans. “I don’t want to see a world where a piece of technology is equated with a person,” the billionaire said on the X platform.
The debate comes as governments and companies invest tens of billions of dollars in building new data centers to meet the computing needs of AI systems.
A May report from the International Monetary Fund (IMF) showed that global data center power consumption has reached levels comparable to Germany or France in 2023, not long after OpenAI launched its landmark ChatGPT model.
In response, some governments are speeding up approval processes to connect new cheap energy sources to the grid, with some environmentalists warning such moves could conflict with global net-zero emissions targets.
Some local communities in the United States and other countries have also expressed opposition to data center development projects, fearing that they would put pressure on the power grid and push up overall electricity prices.
Just last week, the San Marcos, Texas, city council rejected a $1.5 billion data center project after months of public opposition.
In the face of these objections, many technology industry leaders, including Altman, believe that data centers need a more diversified energy supply, including renewables and nuclear power.