A Fayette County, Georgia data center operated by Quality Technology Services drained 30 million gallons of water โ unbilled for months โ while nearby drought-stricken residents were under conservation restrictions. The county eventually retroactively charged QTS about $150,000, declined to fine the company, and water system director Vanessa Tigert told Politico the decision was partly because "they're our largest customer, and we have to be partners. It's called customer service." For anyone watching AI infrastructure expand into ordinary American counties, this is one of the first specific case studies of how the water side of data center economics breaks under stress.
The mechanics: two industrial-scale water hookups at the QTS facility weren't being monitored โ one had been installed without the utility's knowledge, the other was unlinked to the company's billing account. The county is mid-transition to smart cloud-based meters that should make leak detection and usage spikes more visible, but the transition is incomplete and the only meter inspector available was "spread pretty thin," per Tigert. QTS paid the $150K invoice once flagged and said the article's framing is "false and inaccurate" โ all water usage "followed relevant and applicable regulations." The county dismissed it as a "procedural mix-up" and imposed no penalties for exceeding the peak-usage limits established during the data center planning process. Residents on private wells reported sudden water-pressure drops; QTS counters that it doesn't draw from wells or groundwater, and the county sided with QTS on that point. James Clifton, the attorney who exposed the story via public records request, made the broader point: "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."
The bigger picture from a Xylem report cited by The Information: 40% of data centers and 29% of global chip fabs are built in "water-stressed" areas, and AI-associated water use will more than double over the next 25 years. Most of the water cost isn't in the data centers themselves โ it's at the semiconductor fabs producing the chips and the power plants generating electricity to run them. Microsoft is paying FIDO Tech to install "high-tech water leak detection systems" that feed sensor data into AI to isolate leaks, with the pitch that recovering 30% of world water currently lost to leaks could offset some of the buildout demand. The tension is real and obvious: AI infrastructure is creating water demand at the same time AI is being pitched as the solution to the water-grid inefficiency it's worsening. The QTS case shows what happens when the demand outruns the monitoring โ even absent malice, undermonitored hookups in stressed water systems mean residents pay the visibility cost while operators pay nothing.
For anyone evaluating new data center developments in their region โ and that's most of the US given the AI buildout โ the QTS pattern is the template to watch: industrial hookups installed during construction that fall out of meter inspection coverage, county staffing too thin to verify, residents bearing the drought-restriction cost while the operator's water use stays invisible. The fix isn't AI leak-detection (helpful but downstream). It's pre-construction commitments with enforceable monitoring and penalty schedules, plus the political will to fine the largest customer when they exceed them. Tigert's "we have to be partners" framing is the part residents should worry about. If your county's largest customer can't be held accountable, neither can the next one โ and the next ones are being permitted at speed across the South, the Midwest, and the Pacific Northwest right now.