That's the question that haunts Beverly and Jeff Morris. The retired couple quietly bought their dream home in the countryside of Newton County, Georgia. But Tap has been trickle-tricked since Meta began construction of a massive $750 million data center just 1,000 feet from the home. One bathroom doesn't work. The washing machine ran out.
“I'm afraid of drinking my water,” Beverly Morris told NYT.
The couple spent $5,000 trying to solve the problem. But replacing their wells costs five times more and they don't have the money. Meta denied responsibility, but Morrise's three neighbors also report similar water issues. And across the county, a bigger crisis is unfolding quietly.
Thirsty Center
Every time a data center trains powerful AI models or offers social media posts, it generates heat. That heat must be removed to keep the computer running, and the cheapest and most common way to do it is to pump water to cool the machine. This applies to all types of data centers, but especially AI centers.
The core driver of AI water demand is a dramatic increase in the power density of treatment units. Unlike traditional data centers, AI facilities are packed with high-performance graphics processing units (GPUs) and custom accelerators that dissipate much more heat in far more physical space than traditional central processing units (CPUs). This energy concentration creates a “heat wall” that requires a large amount of water to cool. Traditional data centers often rely on air-based cooling, circulating chilled air through a server rack. However, this method is inefficient for high density thermal loads of AI servers that require water.
Older data centers like Meta in Newton County can use 500,000 gallons of water a day. This is enough to supply thousands of households. Even the newer, thirsty centers seek millions of gallons of permission every day. One permit in Georgia required nine million gallons a day. This is equivalent to 30,000 homes.
Select a location above the resource
Maintaining the optimum operating temperature of this sensitive electronic device is unnegotiable for businesses. However, these companies prioritize locations with cheaper energy rather than large water availability, as energy is more expensive than water. Water is considered almost an afterthought. But for struggling communities, it is becoming a real problem.
Worse, much of this water is consumed entirely. In the evaporative cooling system, most of the water is lost to the atmosphere and is not returned to the local basin. This makes data centers different from other industrial processes. This process deals with most of the spent water and drains it into the system.
Researchers estimate that a series of 20-50 queries to large-scale language models like CHATGPT could require consumption of about half a liter (about a bottle) of freshwater to cool the servers that handle the requests. The image above may require approximately 0.5 liters of water. Multiplying these billions of interactions that occur every day around the world, this “invisible” water cost is a major factor in the consumption of the entire industry.
People caught in the middle
But that's a direct cost. Indirect costs come from AI centres that compete for coal, natural gas and nuclear facilities with water.
Of course, there are people in the middle of this.
Newton County is not alone. From Phoenix to Colorado to the United Arab Emirates, communities are wrestling with the water shortages associated with building booming data centers. Around Phoenix, developers have suspended home construction due to droughts exacerbated by high-tech facilities. In Colorado, water-hungry data centers are now part of a renegotiation of the Colorado River.
And the data is ambiguous – literally and figuratively. Companies rarely disclose water usage, making it difficult for regulators to act. Georgia lawmakers tried to push back a bill that would remove tax incentives for new data centers, but Gov. Brian Kemp refused, citing the risks of “economic development.”
This also happens when many cities across the US and beyond are already tackling long-term droughts supported by climate change. In the southwest, record-breaking reservoir levels cause emergency water restrictions, and climate models predict that these dry spells will only become more frequent and more intense. Adding large, opaque demands for AI data centers to this vulnerable system risks turning some areas into a full-scale water crisis.
Stay high and dry
The Newton County Water Department is competing to upgrade recycling facilities and maintain the pace, but even without dramatic action and cooperation, that's not enough. So far, there are no signs that will help the meta or other tech giants step into infrastructure bills. A Meta spokesperson said the company conducted its own research into Morris' property and concluded that it was “impossible” for the data center to affect groundwater.
Mike Hopkins, executive director of the Newton County Water and Sewerage Department in Georgia, said he is seeing applications requesting up to 6 million gallons of water per day, exceeding current daily usage across the county. Already, Meta's data centers are increasing every day about 10% of the county's total water use.
In a hurry to power smarter machines, we may become dry, poor and forget the whole community. And, as Beverly Morris said, perhaps the most troubling: “It feels like we're fighting a pointless battle that we didn't sign up.”
The most important development appears to be in the US, but elsewhere around the world, data centers are causing similar distress. Uruguay was haunted by the worst drought in history in 2023. Google's plan to build a data center that consumes 7.6 million liters of drinking water per day has sparked widespread public outrage. The protesters adopted the slogan. – “It's not drought, it's looting” – framing the issue as a diversion of important public resources businesses at the time of the crisis.
In the Netherlands, where stricter water laws exist, a 2021 survey revealed that Microsoft data centers consumed 84 million liters of drinking water. The figure was far higher than the 100 million to 20 million litres publicly estimated by the company and local officials.
What can you do?
If we improve the situation, policymakers should start treating water as a critical, finite resource, rather than simply a disposable after-paying.
First, there is a need for transparency regarding water use. Without clear water reports, the government cannot make informed decisions, even if it wishes. Next, this water usage should be taken into consideration and ensure that the community's water preparation can support it in a sustainable way. If a data center brings jobs and investments to the region, a portion of that money should be sent to water investments.
The AI revolution is in full swing and water demand continues to increase. Ultimately, future data center site selection should prioritize water sustainability along with energy costs and ensure that communities do not remain dry in the race to power artificial intelligence.
Otherwise, many communities will be left in battles that cannot be won.
