Underground Water Could Be the Fix for AI’s Biggest Cooling Problem
Published · 5 min read
Researchers just found a way to cool data centers using something that’s been sitting under our feet the whole time: groundwater.
Artificial intelligence has an appetite. It eats electricity. It drinks water. And every new data center built to keep up with demand adds more strain to both.
A team at the University of Illinois Urbana-Champaign thinks they’ve found a fix. Not a flashy one. Not a new chip or a smarter algorithm. Just water, stored underground, doing a job it was never asked to do before.
They call it Aquifer Thermal Energy Storage, or ATES for short.
Why Data Centers Are Struggling to Keep Cool
Here’s the problem in plain terms.
Data centers run hot. Servers process massive workloads around the clock, and all that computing generates heat. Someone has to pull that heat out, or the whole system fails.
That’s where the trouble starts.
- Electricity costs pile up. Cooling alone can eat up 10 to 40 percent of a data center’s total energy use, depending on how the facility is designed
- Water gets wasted. Many cooling systems rely on evaporation, which means water disappears from the local supply and doesn’t come back
- Both problems grow together. As AI expands, so does the demand for power and water at the same time
Upasana Pandey, a postdoctoral researcher on the project, put it simply. She said data centers burn through large amounts of electricity, with a meaningful chunk of that going straight to cooling. She also pointed out that the water used often evaporates and never returns to the community it came from.
That’s not a small side effect. That’s the core tension.
Fixing One Problem Usually Breaks Another
Most cooling solutions trade one issue for another.
Cut energy use, and you often need more water. Cut water use, and you often need more energy. Researcher Yu-Feng Lin, who worked on the study, described data centers as sitting right at the center of that water-energy tradeoff.
His team wanted a fix that didn’t force that choice.
How Aquifer Cooling Actually Works
The idea itself is straightforward once you break it down.
Underground aquifers hold groundwater that stays at a fairly steady temperature, no matter what’s happening on the surface. That steady temperature is the whole trick.
Here’s the basic process:
- Cool groundwater gets pulled up through underground pipes
- It runs through a heat exchanger, where it absorbs heat from the data center
- The now-warmed water goes back down into the aquifer for storage
- That stored heat can later be pulled back out and used for winter heating
- Meanwhile, cold winter water gets stored the same way, ready for summer cooling
In short, the system doesn’t waste the temperature difference. It banks it.
Why Illinois Is a Good Testing Ground
Location matters a lot here, and Illinois happens to check the right boxes.
Andrew Stumpf, another researcher on the team, explained the advantage using real numbers. Illinois swings from around 90°F in summer to minus 10°F in winter on the surface. But underground, the temperature barely moves. It sits around 55°F year-round.
That changes the math completely.
Instead of forcing a system to adjust from 90 degrees down to a usable 70, the system only has to shift from about 55 degrees up to 70. That’s a much smaller gap to close, and a much smaller amount of energy needed to close it.
Illinois also has favorable underground geology and strong seasonal swings, which makes it a near-ideal proving ground for this kind of system.
Not Just Clean Water Required
One detail makes this idea more practical than it might sound at first.
The system doesn’t need drinking water to function. It can run on:
- Deep saline aquifers, which are too salty for human use
- Contaminated groundwater that isn’t safe to drink anyway
- Abandoned mines, which often hold standing water no one is using
That matters. It means this approach doesn’t compete with communities for clean drinking water, which has been one of the loudest criticisms aimed at AI infrastructure lately.
The Real Obstacle Isn’t the Technology
So why isn’t this everywhere already?
According to the researchers, the holdup isn’t technical. The science works. The obstacle is money and timing.
Installing an ATES system costs a lot upfront. Data center operators tend to want fast returns, and a system like this takes time to pay for itself. That mismatch, more than any engineering limitation, is what’s slowing adoption.
Lin summed up why water works so well for this job in the first place. He noted that water holds heat efficiently and moves that heat well when it flows, a combination that’s genuinely rare in nature. Groundwater lets engineers tap into those same properties for long-term energy storage.
Why This Matters Beyond One University Study
AI isn’t slowing down anytime soon. More models mean more servers. More servers mean more heat. More heat means more resources spent trying to manage it.
Communities near data centers have already pushed back against the strain on local water and power grids. Solutions that reduce that strain, without requiring new technology to be invented from scratch, carry real weight.
Aquifer cooling won’t fix every data center overnight. The upfront cost is real, and not every region has the right underground geology to make it work.
But it points to something bigger: some of the best answers to AI’s environmental footprint might not come from smarter software. They might come from paying closer attention to what’s already sitting beneath the ground.
What Is Confirmed
- Researchers at the University of Illinois Urbana-Champaign studied Aquifer Thermal Energy Storage as a cooling method for AI data centers
- The findings were published in the peer-reviewed journal Groundwater
- Cooling can account for 10 to 40 percent of a data center’s total energy use
- The system stores summer heat underground for winter use, and winter cold for summer use
- Illinois is well suited to this method due to its seasonal temperature swings and subsurface geology
- The system can use non-potable water, including saline aquifers, contaminated groundwater, and abandoned mines
- High upfront installation costs, not technical feasibility, remain the main barrier to adoption
The Bottom Line
AI’s energy and water problem isn’t going away on its own.
This research offers a rare kind of solution: one that doesn’t ask companies to choose between saving power and saving water. It asks them to use what’s already underground instead.
Whether that idea scales beyond research papers depends on whether data center operators are willing to pay more upfront for savings that show up later.
The technology is ready. The question now is who’s willing to dig for it.
Jonas Muthoni is Editor-in-Chief of Microgrid Media, where he oversees reporting and analysis on microgrids, energy storage, distributed energy resources, data center power demand, grid modernization, resilience, renewable energy, and electricity policy and markets. His work focuses on the infrastructure, technologies, and regulatory developments reshaping the power system.
As Editor-in-Chief, Jonas leads Microgrid Media’s editorial strategy and standards, including story selection, source verification, technical accuracy, and the development of original reporting and analysis. His coverage draws on regulatory filings, government records, utility and company disclosures, technical documentation, and independent industry sources to explain significant developments across the evolving energy system.
