OpenAI chief Sam Altman recently brushed aside worries about water use in artificial intelligence data centers, pointing to newer facilities that skip water altogether. But while water concerns might be fading, the real problem of massive energy consumption keeps growing.
Speaking at an event in India last week, Altman called online claims about water usage “completely untrue” and said newer facilities don’t depend on water for cooling anymore.
Traditional data centers have used millions of gallons to cool equipment, but technology is changing. A recent study still projects water demand for cooling could more than triple in the next 25 years as computing needs expand.
Altman said energy use is a fair concern. The world needs to shift quickly to nuclear, wind, and solar power as AI use grows, he said. When asked about comparisons between human and AI energy efficiency, he argued the comparison should look at energy per query after training, not the training process itself.
He compared training AI models to raising humans, noting it takes 20 years and lots of food before a person becomes productive, in an interview with the Indian Express.
That comparison got criticized. Indian tech billionaire Sridhar Vembu, who attended the same summit, said he doesn’t want to see technology equated with people. His comment shows broader worries about AI potentially replacing human workers.
I do not want to see a world where we equate a piece of technology to a human being.
I work hard as a technologist to see a world where we don't allow technology to dominate our lives, instead it should quietly recede into the background. https://t.co/PrbjbgCYde
— Sridhar Vembu (@svembu) February 22, 2026
Data centers used electricity comparable to entire countries like Germany or France in 2023, based on International Monetary Fund numbers. This happened shortly after ChatGPT launched and sparked an AI boom.
Companies race to build massive off-grid facilities
The GW Ranch project in West Texas will cover 8,000 acres and use more electricity than Chicago. It will make power through natural gas and solar panels, avoiding delays from utility companies struggling with capacity issues.
Similar projects are planned or underway in Wyoming, New Mexico, Pennsylvania, Utah, Ohio, and Tennessee. Meta, OpenAI, Oracle, and Chevron are backing these developments. Some states passed laws making approvals easier.
In West Virginia near Davis, a planned data center will include a gas plant big enough to power every home in the state. Local resident Amy Margolies told reporters that officials removed local control for what she called a “speculative gold rush” shrouded in secrecy.
The United States runs 5,246 data centers using at least 17 gigawatts of power. One large nuclear plant makes one gigawatt, enough for 300,000 to 750,000 homes.
These facilities will include some solar power, but most will run mostly on natural gas because other sustainable options don’t provide steady output. Without grid backup, gas becomes necessary.
Energy researcher Michael Thomas warned this approach is “catastrophic for climate goals.” His firm found 47 off-grid data center projects nationwide.
Elon Musk built an off-grid facility in Memphis last year using portable gas generators to avoid grid limits. His xAI data center went online in months instead of years. But the Environmental Protection Agency ruled in January that his setup broke emissions regulations and ordered the company to get proper permits.
Meta is moving ahead with multiple off-grid projects. One in New Albany, Ohio will use two gas power plants and launch later this year. Another in El Paso links 813 smaller generators. Local officials protested, saying Meta promised clean energy but delivered gas power instead.
The company said it would meet clean energy commitments by purchasing renewable energy credits and adding clean power to the grid elsewhere.
Electricity prices jump as demand strains nation’s largest grid
Communities have fought back against data center proposals in populated areas. Citizens in Tucson beat a project called Blue over water concerns in the desert and potential electricity bill increases. A large data center can use 5 million gallons of water daily, matching a town of 10,000 to 50,000 people, based on Environment and Energy Institute numbers.
San Marcos, Texas rejected a proposed 1.5 billion dollar data center last week after months of public opposition.
Electricity prices have jumped on PJM Interconnection, the largest grid in America, serving 65 million people across 13 states and Washington. The Trump administration and several governors signed an agreement in January requiring tech companies to fund new power plants. Companies committed 15 billion dollars for new generation capacity.
