The UN University Institute for Water, Environment and Health published a sweeping report projecting that AI data centers will consume 9.3 trillion liters of water by 2030, equal to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa. That number is abstract until you trace it to specific addresses. In Morgan County, Georgia, residents testified before Congress holding jars of muddy tap water after a Meta data center began construction nearby. In Fayette County, Georgia, a QTS facility siphoned roughly 30 million gallons of water through unmetered hookups before anyone noticed, while officials told residents to stop watering their lawns. In Arizona, Texas, Oregon, South Carolina, and Utah, similar fights are unfolding simultaneously. The commentary argues that the AI water story is no longer national policy. It is hyperlocal, adversarial, and accelerating faster than any disclosure framework can track it.
A Number Too Big to Picture
Nine point three trillion liters. By 2030, global data centers powering artificial intelligence are projected to consume 945 terawatt-hours of electricity, with an associated water footprint equal to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa. That figure comes from the United Nations University, not an activist group. It is staggering precisely because it resists comprehension. Numbers that large stop meaning anything specific. However, somewhere underneath that abstraction sit real water mains, real wells, and real families. The AI water story stopped being a policy debate months ago. It became a series of local fights, one county at a time.
When the Number Becomes a Glass of Brown Water
Abstraction collapsed instantly in Morgan County, Georgia. Rep. Alexandria Ocasio-Cortez brought jars of dirty water to a congressional hearing, stating: “This is the current drinking water in Morgan County, Georgia, right after a data center was constructed.” She added something sharper still. “The only difference between the clean water and this was that data center.” The consequences extend beyond appearance. “These families now have to ship, in a rural area, have to ship water to their house in order to cook and bathe themselves,” Ocasio-Cortez told the hearing. Bathing and cooking are not abstract metrics. They are daily logistics now disrupted. Congress noticed quickly. The EPA promised an immediate investigation after the hearing. Investigation does not equal resolution, however. Residents are still hauling water while regulators review documents.
Thirty Million Gallons, Unnoticed
Roughly seventy miles away, a different Georgia county discovered a stranger problem. A data center drained 30 million gallons of water unnoticed, until residents complained about low water pressure. Nobody flagged the consumption for fifteen months straight. The mechanism was almost bureaucratic. The Fayette County water system sent QTS a letter detailing two separate meters: one that was not installed with the county’s knowledge, and another that was not connected to QTS’s billing account. Two industrial hookups simply slipped through municipal tracking. Meanwhile, residents felt the squeeze directly. “We get this notification from Fayette County water system saying you need to stop watering your lawns to help conserve water,” said James Clifton, a local property rights advocate. Officials asked households to sacrifice first.
His frustration captured the imbalance precisely. “So the first thing they do is lean on the individuals and the citizens to stop water consumption when we have QTS that’s just absolutely draining us, most months it’s the No. 1 consumer of water in the county,” Clifton said. Notably, QTS markets itself differently. QTS touts a closed-loop cooling system, which it says does not consume water for cooling. The company attributed the spike to construction activity, not operations. Yet the meters tell their own separate story.
This Pattern Is Not Isolated
Georgia is not an outlier case. It is simply the loudest example currently. In Texas alone, a study estimated data centers would use 49 billion gallons in 2025 and as much as 399 billion gallons by 2030, the equivalent of drawing down Lake Mead by more than 16 feet in a single year. Oregon experienced its own version earlier. Google’s data centers in The Dalles, Oregon, a city of 16,000, consumed 355 million gallons in 2021, roughly a quarter of the city’s total water used that year. Transparency proved harder to secure than the water itself. Google funded the city’s lawsuit against a local newspaper that tried to obtain those figures, arguing the data was a trade secret.South Carolina residents pushed back through formal channels instead.
Conservation groups fought Google’s permit to draw 1.5 million gallons a day. Utah saw an even larger public response. A data center proposal tied to Kevin O’Leary drew nearly 3,900 public protests over a water rights application that would have shifted irrigation water to industrial use. Arizona, meanwhile, added contamination concerns to scarcity concerns. A Meta data center in Newton County, Georgia, reportedly disrupted nearby private wells, leaving families hauling water and replacing sediment-clogged appliances. Wells, not municipal mains, carry their own separate vulnerability.
Why Closed-Loop Claims Don’t End the Argument
Corporate messaging has shifted recently toward closed-loop cooling. Microsoft, Nvidia, and QTS all now tout near-zero operational water use. The framing genuinely matters. It is also incomplete. Construction itself consumes enormous volumes first. QTS attributed its 30-million-gallon spike to temporary construction-related activities, such as concrete work, dust control, and site preparation. Closed-loop architecture protects future operations, not the years spent building the facility.
Indirect consumption compounds the gap further still. Even air-cooled data centers may indirectly increase water consumption by increasing electricity demand, since thermoelectric power plants commonly use water for cooling. Nearly half of America’s servers draw power from stressed regions already. Researchers identified that nearly half of U.S. data center servers are powered partly by plants located in water-stressed regions. Contamination risk adds a third layer entirely. Fluorinated gases, a type of PFAS, may also be used for cooling. If leaks, spills, or improper wastewater disposal occur, contaminants may enter soil and groundwater systems. Closed loops reduce withdrawal. They do not eliminate chemical risk during construction or maintenance.
The Gap Between PR and the County Line
UN researchers warned explicitly against single-metric thinking. “What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land,” said Dr. Miriam Aczel, the report’s lead author. Restaurant comparisons and zero-water headlines invite exactly that blind spot. Local officials increasingly recognize the imbalance too. Gregory Pierce, director of the UCLA Water Resources Group, noted it’s unusual that the utility didn’t fine the data center for breaking its own usage rules. His read on the dynamic was blunt. “I don’t know exactly what’s happening here, but they probably don’t want to upset one of their new and largest customers,” Pierce said. That asymmetry defines the fight precisely. Utilities depend financially on the very customers straining their systems. Residents, lacking comparable leverage, end up absorbing both the shortage and the burden of conservation.
What Accountability Actually Requires Now
Closed-loop cooling deserves genuine credit where deployed correctly. It does not retroactively fix muddy taps in Morgan County. It does not refund fifteen months of unmetered withdrawal in Fayette County. Engineering progress and local accountability are separate questions entirely. What residents are demanding is straightforward. They want metered, audited, publicly disclosed water use before construction begins, not after pressure drops. They want contamination testing independent of corporate self-reporting. Above all, they want enforcement that does not bend simply because a customer is large. The UN’s 1.3-billion-person comparison captures planetary scale well. Morgan County’s muddy jar captures something the spreadsheet cannot: somebody’s actual kitchen tap. Both numbers are true simultaneously. Until the industry treats both with equal seriousness, your next chatbot prompt will keep being, quietly, somebody else’s water fight.
