The Thirst Machine
AI data centers consumed 17 billion gallons of water in 2023. By 2028, they’ll need 68 billion. Sam Altman says the concern is “fake.”
On February 23, 2026, OpenAI CEO Sam Altman told a summit audience that water concerns about AI data centers were “fake.” One week earlier, Bloomberg published an investigation showing that AI data center water use was still “too much of an afterthought.” Three days later, reporting from Central Texas revealed that at least 14 counties experiencing moderate to extreme drought were staring down five new data center projects competing for the same disappearing water.
The numbers tell a different story than the CEO does.
The scale
AI data centers consumed approximately 17 billion gallons of water in 2023. By 2028, that number is projected to hit 68 billion gallons — a 300% increase in five years. Klaus Reichardt, CEO of Waterless Co., put the context clearly: “Never in the history of this country has demand for water increased so dramatically in such a short time.”
A typical data center uses 300,000 gallons of water per day — equivalent to the daily needs of about 1,000 households. Large data centers use 5 million gallons per day, equivalent to a town of 50,000 people. A single Meta data center in Newton County, Georgia, consumes 500,000 gallons daily — 10% of the entire county’s water supply.
The International Energy Agency estimates global data center water use at roughly 560 billion liters per year, potentially rising to 1.2 trillion liters by 2030. Water consumption for cooling is projected to increase by 870% as more facilities come online.
Each 100-word AI prompt costs approximately one bottle of water — 519 milliliters — according to University of California, Riverside researchers. ChatGPT alone processes over one billion queries per day.
Where they’re building
Many AI data centers are located in the driest regions of the country. The logic is economic: solar power and renewable energy are cheapest in deserts and arid plains. The consequence is ecological: the facilities consuming the most water are built where water is scarcest.
In Hays County, Texas, aquifers have reached historic lows. At least five new data center projects are planned in the region. Throughout Central Texas, 14 counties are experiencing moderate to extreme drought — including Travis, Williamson, Guadalupe, and Caldwell. The data center projects will compete directly with agricultural, residential, and municipal water supplies.
This isn’t a Texas-specific problem. Arizona, where a $14 billion data center project in Goodyear recently withdrew under community pressure, faces chronic water stress. Northern Virginia, which hosts roughly 13% of the world’s data center capacity, is already straining its water infrastructure. The planned Bessemer, Alabama facility would consume 2 million gallons of water daily.
What cooling does
The reason data centers consume so much water is straightforward: heat. Every computation generates heat. AI computations generate more heat than traditional data processing because the processors run harder and longer. Cooling systems — primarily evaporative cooling towers — use water to absorb and dissipate that heat.
The average new data center being planned requires 450 megawatts of power — ten times the 45-megawatt average of existing facilities. More power means more heat, which means more cooling, which means more water. The relationship is linear and relentless.
Closed-loop cooling systems can reduce freshwater use by up to 70%. But they’re more expensive, and most facilities are still being built with evaporative cooling. The industry’s standard response — that they’re investing in water efficiency — doesn’t change the trajectory. Even a 70% reduction in per-facility water use can’t offset a 300% increase in total facilities.
Who pays
The economic impact flows downhill. When data centers consume water at industrial scale, water costs rise for everyone else in the region. Industries that depend on water — manufacturing, hospitality, healthcare, commercial real estate, agriculture — face cost increases they didn’t create and can’t control.
In Indiana, household utility bills climbed 17.5% in 2025, driven partly by data center infrastructure costs. Virginia’s utility regulator estimates that data center expansion will add $37 per month to every household’s energy bill by 2040. The water costs haven’t been similarly quantified yet, but the dynamic is the same: industrial consumption at scale makes the resource more expensive for everyone.
Farmers are the most directly affected. Agriculture accounts for roughly 80% of consumptive water use in the western United States. When a data center that employs 100 people competes for water with farms that employ thousands and feed millions, the conflict isn’t abstract. It’s a zero-sum competition for a finite resource in a region that’s running out.
The efficiency paradox
The AI industry’s standard defense is efficiency: newer facilities use less water per unit of computation. This is true. It also doesn’t matter at the scale being planned.
If each new data center is 50% more water-efficient but the industry builds ten times more data centers, total water consumption still increases fivefold. Efficiency gains are real. They are also irrelevant at the growth rates the AI industry is pursuing. The math only works if the industry stops growing, which is the opposite of every business plan in the sector.
This is the same pattern that played out with energy efficiency in automobiles, home appliances, and electronics: each unit gets more efficient, but total consumption increases because there are more units. Economists call it the Jevons Paradox. The AI water crisis is the Jevons Paradox applied to a resource that’s already scarce in the places where the industry wants to build.
What Altman actually said
When Altman called water concerns “fake” at a February summit, he framed it in context: “humans use energy too.” The argument is that data centers’ total water consumption is a small fraction of national water use, and that the benefits of AI — measured in economic productivity and innovation — justify the resource cost.
This is true as a fraction of total use. It’s misleading as a description of local impact. A data center that consumes 10% of a county’s water supply changes life in that county, regardless of what fraction it represents nationally. The communities fighting data center projects aren’t worried about national averages. They’re worried about their wells, their farms, and their water bills.
The AI industry’s water problem isn’t that it uses too much water globally. It’s that it uses too much water in specific places where there isn’t enough — and it plans to use dramatically more in those same places. That’s not a fake concern. That’s an infrastructure crisis being built on schedule.
Originally published at https://noahaust2.github.io/strategist-dashboard/blog/the-thirst-machine.html
Write a comment