Last updated: May 2026
Quick answers
Which data center is causing the Georgia water problems?
Meta’s Stanton Springs data center, located on the border of Newton and Morgan counties in Georgia, is the facility at the center of the controversy. The development consumes roughly 10% of Newton County’s total daily water supply and has been linked to brown tap water and a 33% rate hike for local residents.
Why is the EPA investigating Meta’s Georgia data center?
EPA Assistant Administrator for Water Jessica Kramer pledged an investigation on May 20, 2026 after Rep. Alexandria Ocasio-Cortez presented jars of brown drinking water from Morgan County during a congressional hearing. Kramer stated she would review whether data center construction violated EPA water quality standards.
How much water does a data center use per day?
Enterprise data centers consume 300,000 to 500,000 gallons per day. Large hyperscale facilities, like those operated by Meta, Google, and Microsoft, use between 1 and 5 million gallons daily for cooling. That’s equivalent to the daily water use of 30,000 to 50,000 households.
On May 20, 2026, Rep. Alexandria Ocasio-Cortez held up two jars of water during a hearing of the House Energy and Commerce Subcommittee on Oversight and Investigations. The jars were murky, brown, and visibly contaminated. She told EPA Assistant Administrator Jessica Kramer that the water came from Morgan County, Georgia, where Meta had built a data center. Families in the area had started relying on bottled water to drink and cook.
Meta’s Stanton Springs data center in Morgan County, Georgia is the focus of a federal water quality investigation after residents reported brown tap water and a 33% rate hike tied to the facility’s water consumption. The EPA pledged to review the situation immediately.
The image went viral. Over 26,000 upvotes on Reddit. AOC’s clip circulated on every major platform. But here’s what most of the coverage missed: Meta’s Georgia mess isn’t a one-off. It’s a preview of what happens when $785 billion in planned AI infrastructure collides with small-town water systems that were never built for it.
For founders building anything that touches AI compute, this story isn’t just policy news. It’s a cost input. And the costs are moving in one direction.
What happened in Morgan County, Georgia?
Meta’s Stanton Springs facility sits on the border of Newton and Morgan counties, about 50 miles east of Atlanta. According to Mike Hopkins, executive director of the Newton County Water and Sewerage Authority, the data center consumes approximately 10% of the county’s total daily water supply.
That figure alone is striking. One facility, one company, one-tenth of a county’s water.
But the consumption numbers are only part of the problem. Residents in the surrounding area began reporting degraded water quality after construction ramped up. Tap water turned brown. Appliances stopped working. Water pressure dropped. A Fortune investigation published May 13, 2026 reported that Newton County is on track for a total water deficit by 2030 if current consumption patterns hold, with rates expected to surge 33%.
Meta’s response has been to point to a third-party groundwater study the company commissioned, which found no connection between data center operations and the well water issues. Spokesperson Ryan Daniels said Meta works closely with local utilities to ensure no negative impacts. The company maintains all its water comes from the local utility, not groundwater sources.
Residents aren’t buying it. And neither, apparently, is Congress.
During the May 20 hearing, Kramer acknowledged she hadn’t received complaints about the correlation between data center construction and drinking water contamination prior to Ocasio-Cortez’s presentation. She pledged to investigate immediately upon returning to her office, stating that ensuring EPA water quality standards are met remains a priority.
How much water does a data center actually use?
A large hyperscale data center consumes between 1 and 5 million gallons of water per day for cooling. That’s equivalent to the daily water needs of a town of 30,000 to 50,000 people, according to data from the Environmental and Energy Study Institute.
The numbers scale fast. U.S. data centers collectively consumed 17.4 billion gallons of water in 2023. Annual consumption is projected to climb to between 38 and 73 billion gallons by 2028, according to research from the Lincoln Institute of Land Policy. A January 2026 analysis by Global Water Intelligence found that 30 minutes of AI usage requires roughly 0.16 gallons of water.
Up to 85% of the water data centers consume evaporates through cooling systems and never returns to the local water supply. That’s what makes this different from, say, a manufacturing plant that discharges treated wastewater back into a river. The water is gone.
And Georgia isn’t an isolated case. A separate data center project in Fayette County, Georgia consumed 29 million gallons of water over 15 months through two connections that the county didn’t even know existed, Tom’s Hardware reported. The operator, QTS, owed $147,474 in retroactive charges. The county refused to fine them.
Google reported consuming over 6 billion gallons of water across all its data centers in 2023 alone. The company uses reclaimed or non-potable water at over 25% of its campuses. Amazon Web Services said 20 of its facilities cool with purified wastewater instead of potable water. But these are the exceptions, not the standard.
Why are communities blocking data centers across the U.S.?
Water is becoming the primary reason communities say no. And they’re saying no a lot.
According to Fortune’s May 18 investigation, at least 48 data center projects representing $156 billion in investment were blocked or stalled by local opposition in 2025. Project cancellations jumped from 6 in 2024 to 25 in 2025. In Q1 2026 alone, more than 20 additional projects were killed, a record quarterly pace.
There are now 188 local opposition groups operating across 40 states. A Gallup survey found 71% of Americans would oppose a data center in their community, a higher disapproval rate than for nuclear plants or gas facilities. Data centers are now America’s least-wanted neighbor.
The opposition pattern is consistent. A hyperscaler identifies a rural community with cheap land, low regulatory friction, and available power. The project arrives with NDAs for local officials. Researchers at the University of Mary Washington found that 25 of 31 Virginia communities with data center projects had nondisclosure agreements that prevented the public from learning a facility was even being proposed. The public finds out after construction starts, not before.
By the time residents realize what’s happening, construction is underway. Water pressure drops. Rates climb. And the tax incentives that lured the data center in the first place mean the community often doesn’t see proportional revenue. In Georgia, the QTS facility in Fayette County racked up $147,474 in unpaid water charges through unauthorized connections. The county didn’t fine them.
The financial asymmetry is worth noting. Moody’s pegs hyperscaler capex at $785 billion for 2026. The communities absorbing these facilities operate on municipal budgets measured in tens of millions. When venture-backed AI companies need more compute, they contract with cloud providers who contract with hyperscalers who build in these communities. The cost of brown water in Morgan County is externalized down the chain.
The backlash is reshaping where AI gets built. In Utah, 53% of voters oppose the Box Elder proposal, a major data center project. Multiple cities are adding local guardrails, from conservation-plan requirements for large water users to outright prohibitions on potable water for cooling.
What did AOC say about data centers and water?
Ocasio-Cortez’s May 20 hearing appearance was specific, not rhetorical. She brought physical evidence: two jars of water from Morgan County collected during a visit earlier that month. She told Kramer that the only difference between clean water and what she was holding was the data center.
The exchange landed because it was concrete. Not a policy brief. Not a position paper. Brown water in a jar.
She pointed out that families near the Stanton Springs facility can’t use their tap water to drink or prepare meals. Appliances are failing because sediment and contamination are destroying internal components. And the data center consumes a tenth of the community’s daily water while residents absorb rate increases.
The timing amplified the impact. Just 10 days earlier, on May 11, EPA Administrator Lee Zeldin had proposed a rule change that would let data center developers begin pre-construction activities, including clearing vegetation, grading, excavating, and installing utility infrastructure, before receiving their final environmental permits. The rule redefines what counts as “beginning actual construction” under the New Source Review provisions of the Clean Air Act.
The proposed change is open for a 45-day public comment period. Environmental groups, including the Sierra Club, have criticized it as weakening public health protections. Industry groups have praised it as reducing permitting delays for AI infrastructure.
For founders watching this space, Zeldin’s rule creates a paradox: it accelerates construction timelines while the EPA simultaneously commits to investigating the impacts of facilities already built. The regulatory environment is moving in two directions at once.
What this means if you’re building with AI
Every AI product runs on compute. Every compute facility runs on water and power. And the siting playbook for those facilities just broke.
Moody’s Ratings projects that the top six U.S. hyperscalers, Microsoft, Amazon Web Services, Meta, Alphabet, Oracle, and CoreWeave, will spend $785 billion on capital expenditure in 2026 and nearly $1 trillion in 2027. That money is funding the construction of hundreds of new data centers. Each one needs a water source.
The Morgan County story reveals three risks that founders should track:
Compute costs will absorb community resistance. Blocked projects don’t disappear. They move to locations with weaker opposition, which often means higher land costs, longer permitting timelines, or less favorable utility agreements. Those costs flow through to the price of compute. If you’re budgeting AI infrastructure costs for 2027 and beyond, build in 15-20% uncertainty for siting friction. The Q1 2026 cancellation pace, more than 20 projects killed in 90 days, represents a structural shift, not an anomaly.
Water is the new constraint on AI scaling. Energy has dominated the AI infrastructure conversation for years. Water is catching up. Phoenix-area data center cooling water demand could rise from 385 million gallons to 3.7 billion gallons annually, according to research from the Brookings Institution. In Arizona, Nevada, and Utah alone, new data center proposals could demand roughly 7 billion gallons of water per year. Startups building applications that require dedicated compute should evaluate their cloud provider’s water sourcing strategy. AWS and Google have public water stewardship commitments. Others don’t. That distinction will matter more as municipalities start conditioning permits on water conservation plans.
Policy risk is now a real cost input. The EPA investigation, Zeldin’s permit reform, the growing network of 188 opposition groups, and the record pace of project cancellations all point to the same conclusion: the regulatory environment for data center construction is getting harder to predict. Founders who lock in long-term compute contracts should understand where their provider’s facilities are located and what local policy risks those sites carry. This is especially relevant for AI automation agencies and startups in the AI acquihire pipeline, where compute access is table stakes.
The data center buildout won’t stop. The economics are too strong. But the assumption that rural communities will keep absorbing the costs of AI infrastructure, their water, their rates, their quality of life, is no longer valid. Morgan County proved that.
The bigger picture for AI infrastructure in 2026
Meta’s Georgia situation exists inside a larger pattern. The AI boom requires physical infrastructure at a scale the tech industry hasn’t attempted since the fiber optic buildout of the late 1990s. Except this time, the resource constraints are water and power, not just capital.
Consider the numbers. Hyperscaler capex projections have been revised upward three times in 2026 alone, from $700 billion to $785 billion. Moody’s separately flagged $662 billion in off-balance-sheet data center commitments across just five companies: Meta, Amazon, Microsoft, Oracle, and Alphabet. These commitments represent binding obligations for facilities that haven’t been built yet, in communities that haven’t been asked yet.
The transparency problem compounds the resource problem. University of Mary Washington researchers found that 25 of 31 Virginia communities with existing, approved, or proposed data centers had nondisclosure agreements with local officials. These NDAs kept the public from learning a facility was even being proposed, let alone its size or water demands. When residents eventually find out, the response isn’t measured. It’s 188 opposition groups across 40 states.
Some companies are adapting. Google now uses reclaimed water at more than a quarter of its data center campuses. Microsoft has committed to becoming water-positive by 2030. AWS says 20 of its data centers cool with purified wastewater instead of potable water. But the industry average Water Usage Effectiveness ratio sits around 1.8 liters per kilowatt-hour, and most facilities still rely on municipal potable water systems that serve residential populations.
The alternatives exist but aren’t scaling fast enough. Liquid cooling, which circulates coolant directly through server components, uses far less water than evaporative cooling. Immersion cooling eliminates water from the equation entirely by submerging servers in dielectric fluid. Both approaches are gaining adoption in new builds, but the hundreds of hyperscale facilities already operating or under construction were designed around evaporative cooling. Retrofitting those facilities isn’t simple or cheap. And the timeline pressure from AI demand means most new builds are still choosing the proven, water-intensive approach over the experimental one.
The tension between AI’s growth trajectory and local resource limits isn’t going away. Newton County’s projected 2030 water deficit isn’t speculation. It’s what happens when you add hyperscale demand to infrastructure built for a county of 120,000 people. And the current administration’s regulatory approach is to accelerate construction, not slow it down.
For founders in the AI compute space, the lesson from Morgan County is operational, not political. Water access will shape where compute gets built. Where compute gets built will shape its cost. And cost determines which AI products are economically viable. AI search engines and the tools founders build on top of them all run on this same physical layer.
That’s the supply chain nobody’s modeling yet. Maybe they should start.
