The Department of Energy's Office of Geothermal published its position on data centers, and the framing is unusually direct. The page puts the demand picture on the table: data centers consumed 4.4 percent of US electricity in 2023, up from 1.9 percent in 2018, and projected to land between 6.7 and 12 percent by 2028. Then DOE answers with two products. Geothermal generation at roughly 90 percent capacity factor. And Cold Underground Thermal Energy Storage, a cooling tool rather than a power one.
Cold UTES is the underrated piece of the page. The idea is to inject cold water into a subsurface formation during low-demand periods and pull it back during peak periods to cool the facility. It functions as a battery for chilled water rather than electricity. For a data center, that translates into smaller chiller plant sizing because the peak does not have to be met by mechanical cooling capacity, and it translates into a lower marginal cost of peak cooling because the cold was banked when energy was cheap. DOE is positioning Cold UTES as a peak-shaving mechanism that reduces the power required for cooling. That is the same efficiency lever the Penn State reinforcement learning controller captures on the software side, here delivered through subsurface geology.
The DOE-funded Southwest Virginia effort is exploring abandoned mine water as a data center cooling resource. The framing is an underground cold energy reserve sitting in billions of gallons of naturally cool subsurface water in mined-out coal seams. The engineering challenge is straightforward in concept and non-trivial in execution: pump cool mine water through a heat exchanger that serves the facility cooling loop, return the now-warm water to the subsurface, and avoid the legacy contamination issues that come with abandoned mines.
If it works, it converts an environmental liability into a cooling asset. Southwest Virginia has the geology, the population stress around the AI buildout, and the historical workforce that knows the mines. The same pattern exists across Appalachia, the UK Midlands, and German and Polish coal regions. A successful reference deployment becomes a template for every operator who wants a low-water cooling story without having to permit evaporative draw.
Federal endorsement of geothermal power and Cold UTES for data centers does two practical things. It sends a procurement signal to operators and developers that this architecture has technical support and likely future funding. And it puts geothermal and Cold UTES into the conversation utility regulators have with operators about interconnection and load shape. The 90 percent capacity factor figure DOE highlights is a direct counter to the intermittency framing that has dominated renewable cooling discussions. A data center that pairs geothermal generation with Cold UTES has a credible argument that it does not need the same transmission upgrades a wind-heavy facility would, which is increasingly the gating policy item.
Two markers. First, named DOE-supported deployments at meaningful megawattage, not pilots. A 50 MW geothermal-cooled data center is the proof point. Second, whether the cooling vendor base picks up Cold UTES as a standard option in design packages, the way they picked up rear-door heat exchangers. If Vertiv, Schneider, and the other major thermal vendors start quoting Cold UTES on cooling-load proposals, the architecture is in the mainstream. If they do not, it stays a DOE pilot category.