Every time someone sends a message to an AI assistant, a data center somewhere uses electricity, generates heat, and cools that heat away, usually with water. A large data center can use up to 5 million gallons of water in a single day, about the same as a town of 50,000 people. Across all US data centers, the daily total is around 550 million gallons.
Here is the part that matters right now: the majority of new AI data centers are being built in places that are already short on water. Since 2022, nearly two-thirds of new US data centers have been built in high water-stress areas like California, Arizona, and Texas. Over 60% of the US is currently experiencing drought conditions, with reservoirs in many regions sitting below historical averages.
The collision between those two facts is already producing visible friction. In Georgia, residents in one county noticed drops in water pressure. Regulators investigated and found a data center had consumed over 29 million gallons through meters that were not properly tracked. In Iowa, state officials found 40 unpermitted wells at a data center site. In Arizona, the Tucson City Council unanimously rejected a proposed large data center complex over water concerns. These are not isolated incidents.
The reason data centers cluster in dry states comes down to simple economics. Texas has cheap electricity and no state income tax. Arizona has extended data center tax breaks through 2033. Virginia, the single largest hub, offers cheap power and fast permitting. Water has historically not been a factor in these location decisions. That is changing, but slowly.
The water problem is not straightforward either. Cooling systems at most data centers work by evaporating water, which means roughly 80% of the water they draw never comes back. It disappears into the air. And heat makes the problem worse: in Arizona, a data center's summer water usage can be nearly double its average monthly level, exactly when local residents and farms need water most.
Texas illustrates the scale of what is coming. One study estimates data centers in that state alone will use 49 billion gallons of water in 2025. By 2030, that figure could reach 399 billion gallons, equivalent to drawing down the largest reservoir in the US by over 16 feet in a single year. Texas is already dealing with drying reservoirs and at least one city preparing to declare a water emergency.
The tech companies are not ignoring this. Google says it only uses water cooling where local sources are healthy. Microsoft is deploying closed-loop systems that eliminate water evaporation entirely, cutting usage by over 125 million liters per facility per year. Several major companies have pledged to become what they call "water positive" by 2030, meaning they plan to put more water back into local systems than they take out. Whether those commitments hold as construction accelerates is an open question.
On the regulatory side, states are starting to act. In 2025, over 200 bills were introduced across all 50 states targeting data centers, covering water reporting, usage limits, and siting restrictions. Utah, Michigan, Virginia, and South Carolina have passed or introduced laws requiring detailed water usage reporting. Some local governments have imposed temporary building pauses while they figure out the rules. The federal government, by contrast, has moved in the opposite direction, pushing to streamline environmental permitting for large data center projects.
For business operators, the practical implication is this: water supply is becoming a genuine operational risk in the same regions that house the most AI infrastructure. If your business depends on cloud services or AI tools, your supplier's physical infrastructure sits in places that may face resource constraints within the next few years. That does not mean anything breaks tomorrow, but it does mean the cost and reliability picture for AI-dependent services may shift as local governments, regulators, and communities push back harder on what is being built in their backyard.