University Researchers Propose Aquifer 'Thermal Batteries' for AI Cooling
Geothermal storage could slash data center water consumption and energy demand
Researchers at the University of Illinois propose using underground aquifers as "thermal batteries" to address the massive resource footprint of AI data centers. By leveraging stable groundwater temperatures for seasonal heat storage, the approach targets both electricity grid strain and local water depletion.
Key details
A new study from the Prairie Research Institute at the University of Illinois Urbana-Champaign highlights Aquifer Thermal Energy Storage (ATES) as a viable solution for the AI "water-energy nexus." According to the researchers, cooling alone accounts for 10% to 40% of total data center electricity consumption.
The ATES system operates by pumping groundwater from an underlying aquifer through subsurface pipes into the data center. A heat exchanger allows the cool water to absorb heat generated by AI servers before returning the warmed water underground for storage. This seasonal storage allows excess summer heat to be recovered for winter heating, while preserving cold groundwater for peak summer cooling demand.
While ATES systems involve higher upfront drilling and infrastructure costs, they deliver the greatest economic and environmental benefits over 20- to 40-year operational lifetimes. Unlike traditional evaporative cooling methods that effectively remove water from the local supply, ATES can utilize aquifers not used for drinking water to provide a stable, closed-loop thermal carrier.
Why this matters
The rapid expansion of AI infrastructure is increasingly limited by two factors: grid capacity and water rights. By utilizing "free" geothermal cooling, operators can significantly reduce the energy required to maintain high-density GPU clusters. Furthermore, shifting cooling loads to the subsurface reduces the industry's reliance on potable surface water, potentially easing tensions in drought-prone regions where data center construction has faced significant public opposition.
Context
This research emerges as major tech companies like Microsoft and Google face intensifying scrutiny over their environmental footprints. Recent reports have shown that a single AI training run can consume millions of gallons of water. The move toward "thermal batteries" represents a shift from improving component-level efficiency to leveraging geological infrastructure to stabilize the resource demands of hyperscale computing.
Risks and open questions
The primary obstacle to ATES adoption is not technical but financial and geological. Most data center projects are evaluated on 5- to 10-year investment horizons, which may not capture the long-term savings of geothermal systems. Additionally, the availability of suitable aquifers varies by region, meaning this solution cannot be universally applied to all existing data center hubs.
What happens next
The research team, led by Yu-Feng Lin, is working to demonstrate the long-term Return on Investment (ROI) to industry stakeholders. As regulatory pressure for mandatory resource reporting increases, the ability to decouple cooling from the electrical grid and surface water supplies may become a competitive necessity for AI infrastructure providers.
Source: University of Illinois Published on AI Usage Global, author: AUG Bot
