Last summer, residents in parts of the American Southwest were told to limit lawn watering as reservoirs dropped under extreme heat. At nearly the same time, construction crews nearby were preparing sites for new AI data centers expected to consume millions of gallons of water each year for cooling.
For many local officials, the contradiction is becoming harder to ignore.
Artificial intelligence is often described as a digital revolution. But the infrastructure behind it is deeply physical — massive warehouses of servers, cooling towers, pipes, pumps, and electrical equipment running day and night. And as AI systems become more powerful, the amount of water needed to keep them from overheating is rising fast.
What worries regulators is not just the scale of consumption. It is where the demand is growing: many of the world’s newest AI-focused data centers are being built in regions already struggling with drought, groundwater depletion, and long-term water stress.
The Hidden Cost of AI Cooling
Inside a hyperscale data center, temperatures climb quickly.
Training large AI models requires thousands of specialized GPUs operating simultaneously for weeks or months at a time. The hardware produces enormous heat, and without continuous cooling, servers can fail within minutes.
Many facilities still rely on evaporative cooling systems, where water absorbs heat before evaporating into the atmosphere. The process is efficient. It is also water intensive.
Researchers from the University of California, Riverside and the University of Texas estimated that training GPT-3 in Microsoft-operated facilities may have consumed approximately 700,000 liters of freshwater during cooling operations. Their research warned that global AI-related water withdrawals could reach between 4.2 and 6.6 billion cubic meters annually by 2027.
Dr. Shaolei Ren, one of the researchers behind the study, said public awareness remains surprisingly low.
“People are generally aware about electricity usage, but water consumption is not something people think about.”
That disconnect matters because water shortages are local. Communities do not experience them as abstract environmental statistics. They experience them through restrictions, shrinking reservoirs, dry wells, and rising utility costs.
And unlike electricity demand, water stress becomes visible very quickly.
Why Data Centers Keep Expanding Into Water-Stressed Regions
The locations attracting AI infrastructure are not random.
Data center developers typically look for:
- cheap land,
- stable energy access,
- tax incentives,
- fiber connectivity,
- and fast permitting.
Many drought-prone regions happen to offer all five.
A Bloomberg investigation published in 2024 found that roughly two-thirds of newly developed AI-oriented data centers since 2022 were located in areas already facing high levels of water stress.
In states like Arizona and Texas, local governments have aggressively competed for data center investment because of the jobs and tax revenue these facilities can generate. But water agencies are increasingly warning that existing infrastructure may not be designed for continuous industrial cooling demand at this scale.
The International Energy Agency estimates that a single 100-megawatt data center can consume up to 2 million liters of water per day depending on cooling systems and climate conditions.
That is roughly equivalent to the daily water use of thousands of households.
For communities already under drought restrictions, those numbers land differently.
- Related post The US Is Making Clean Energy More Expensive Right as AI Data Centers Need Massive Amounts of Power
“We Just Want Straight Answers”
In St. Joseph County, Indiana, concerns over water use became a flashpoint during discussions tied to a proposed Microsoft data center expansion.
Residents crowded into public meetings asking officials whether groundwater studies had been completed and how much water the project would ultimately require. According to reporting from Windows Central, some community members said basic details about water consumption remained unclear even after multiple hearings.
One resident told officials:
“Nobody explained how much water this thing would actually need.”
Another added:
“We’re not against technology. We just want straight answers before these projects move forward.”
Those comments reflect a broader frustration emerging in communities across the United States. Residents often hear announcements about economic growth and innovation, but far fewer specifics about water sourcing, drought contingency plans, or long-term environmental impact.
That gap is beginning to create political tension.
The Policy Problem Regulators Are Now Facing
For policymakers, the dilemma is real.
AI infrastructure brings major economic incentives. Cities want the investment. Governors want the jobs. Technology companies argue that advanced computing capacity is becoming strategically important for economic competitiveness and national security.
But environmental regulators are increasingly being forced to ask a different question:
What happens if water demand outpaces supply?
The issue is no longer theoretical in drought-prone regions where aquifers are already declining and climate pressures are intensifying.
Corporate sustainability reports show water demand moving upward alongside AI expansion.
Google reported consuming approximately 6.1 billion gallons of water in 2023, according to its 2024 Environmental Report. The company acknowledged that AI growth may place additional pressure on cooling infrastructure.
Microsoft also reported a 34% increase in water consumption compared to 2021 levels, partly tied to the expansion of AI-related infrastructure.
The numbers themselves are significant. But for local governments, the harder issue is uncertainty.
Many municipalities still lack standardized reporting systems that would allow them to compare:
- projected water withdrawals,
- seasonal demand spikes,
- groundwater impact,
- and recycling efficiency across facilities.
That makes long-term planning difficult — especially in regions where drought conditions can worsen rapidly.
AI’s Resource Footprint Is No Longer Invisible
For years, the technology industry marketed cloud computing as something almost immaterial. Data lived “in the cloud.” AI appeared frictionless to users typing prompts into chatbots.
But the physical footprint behind those systems is becoming harder to hide.
Researchers at the World Resources Institute have repeatedly warned that water stress is expected to intensify globally due to climate pressures and rising industrial demand.
Some companies are experimenting with alternatives, including recycled-water systems, liquid immersion cooling, and air-cooled facilities that reduce freshwater dependence. In China, developers recently launched an underwater AI data center cooled partly by surrounding seawater.:
Still, those projects remain exceptions.
Most AI infrastructure today continues to depend heavily on water-intensive cooling systems.
And that is forcing policymakers toward difficult decisions:
- Should AI facilities be required to publicly disclose water usage?
- Should drought-prone cities impose stricter cooling regulations?
- How much industrial water demand is acceptable during periods of scarcity?
Those debates are likely to intensify as AI expansion accelerates.
Because for communities living through water restrictions and declining reservoirs, the cloud no longer feels weightless.
It feels very heavy.