Researchers modeled a way to store cooling underground and use it when data centers need electricity most. Commercial-scale performance remains to be demonstrated.
Data centers must remove heat from computing equipment around the clock. Cooling demand can become especially costly during hot hours, when mechanical chillers draw more electricity and the surrounding grid may already face heavy demand. Developers usually address that need with on-site cooling equipment and access to sufficient electric capacity.
A Department of Energy project spotlight describes another possibility: store cold energy underground when conditions or electricity prices are favorable, then use it to reduce chiller operation during peak hours. The approach could change when a facility needs cooling power. Whether it can do so affordably at a real site remains the central question.
Cold underground thermal energy storage, or cold UTES, connects a data center’s cooling system to a subsurface storage system. During cooler periods or lower-demand hours, the facility uses available cooling capacity to chill the ground or a circulating fluid. Later, stored cooling helps absorb heat from the facility when outside temperatures and electrical demand rise.
This is not an underground refrigerator that replaces every other component. Pumps, heat exchangers, controls, and conventional cooling equipment still have roles. Storage lets operators shift some electrically intensive cooling away from peak hours; its value depends on how effectively the entire system works together.
The schematic shows how stored cooling could reduce chiller use during high-demand hours, via National Laboratory of the Rockies.
Storage can use arrays of drilled boreholes or deeper underground reservoirs. Boreholes exchange heat with surrounding rock and soil. Reservoir designs circulate fluid through suitable subsurface formations. Neither option fits every parcel: developers need to assess underground conditions, available space, drilling feasibility, and how the storage system would connect to the facility. Site testing matters because a promising model cannot establish the performance of unexamined geology.
An October 2026 revision of the National Laboratory of the Rockies’ Phase 1 report analyzed hypothetical 1-gigawatt data centers in Loudoun County, Virginia, and Maricopa County, Arizona. The locations differ in climate, electricity systems, and cooling needs. Researchers compared designs with and without cold UTES rather than measuring results at commercial installations.
For the Virginia reference case, the report modeled a reduction of about 200 megawatts in peak electrical demand for cooling. Its modeled annual cooling-electricity expense fell from roughly $30 million to $10 million. Arizona’s modeled expense fell from about $22 million to $9 million. Those figures describe particular assumptions about facility design, weather, electricity prices, and storage performance; they are not savings promises for other projects.
The laboratory’s model shows lower annual cooling-electricity costs in both reference cases; the bars represent projections, not measured commercial savings, via National Laboratory of the Rockies’ revised Phase 1 report.
Lower peak demand might also let a data center use more of an existing grid connection for computing instead of cooling. That potential benefit matters to operators, but it creates a separate public question: who receives any savings if a large customer needs fewer grid upgrades?
For policymakers and utilities, the useful test is whether storage reliably reduces a facility’s peak demand enough to change an actual interconnection or infrastructure decision. Planning should compare installation and operating costs with other cooling designs, demand-management options, and grid upgrades. Any claim of avoided public expense should identify which project or cost was actually avoided.
Households should not interpret a developer’s modeled energy savings as an automatic reduction in residential rates. Utility tariffs and cost-allocation rules determine who pays for shared infrastructure. Cold storage could help limit pressure on a local grid if it performs as expected, but a data-center owner might retain its direct operating savings. Transparent contracts and measured peak-load data would make the consumer implications clearer.
Researchers still need commercial-scale demonstrations that measure charging and discharge performance across seasons. Such projects should document drilling costs, groundwater and other site constraints, cooling reliability, maintenance, and performance during prolonged heat. Comparisons should use a realistic conventional-cooling baseline rather than assuming every site would otherwise build the same equipment.
The laboratory says a follow-on collaboration aims to develop demonstration sites. That is a step toward testing the concept, not evidence that its modeled savings have already been delivered across the data-center industry.
Underground cold storage offers a plausible way to shift some data-center cooling demand away from costly peak hours. Modeling in Virginia and Arizona shows why researchers and operators consider it worth testing.
Geology, construction costs, cooling reliability, and local electricity rules will determine whether an individual project works. Demonstrations must establish those results before projected savings become claims about proven performance.
For households, the final test extends beyond a data center’s power bill. Policymakers and utilities will need to show whether reduced peak demand actually avoids shared grid costs and how any savings are allocated.