One loop, Two problems solved: Co-locating a data centre with a geothermal network
Two infrastructure problems are usually treated as unrelated. Our feasibility work started from the premise that they are actually each other's solution.
A data centre's growth is constrained less by chips than by water and electricity, and every cooling choice trades one scarce resource for another. Reject the heat through cooling towers and you cool cheaply, but you evaporate water: up to 80% of what the site draws is lost to the air. Reject it through dry or air-cooled systems and the water problem shrinks, but the power needed to move that heat rises sharply, straining the sector's tightest limit. Neither escape is clean; one burns more electricity and drives up emissions, the other pushes heat and humidity into the local environment. Either way, a facility spends enormous resources to throw low-grade heat away.
Meanwhile, a quieter shift is underway in the buildings next door. Networked geothermal is spreading across the country as communities look for ways to cut the cost of heating and cooling. Twelve states have now passed legislation enabling these thermal energy networks, utilities in Colorado, Massachusetts, Minnesota, and New York are running pilots, and dozens of communities are commissioning feasibility studies to explore whether a shared loop can decarbonize their heat and lower bills at neighborhood scale. The appeal is efficiency through sharing: because buildings on the loop balance each other's heating and cooling, the network recycles thermal energy that would otherwise be wasted. Its one stubborn cost is the ground itself, the boreholes and the trenching, which dominates the capital budget.
That cost is exactly what a data centre could offset. Our assessment looked at co-locating one within such a network: instead of rejecting its heat to the air, the data centre feeds it continuously into the shared loop, carrying part of the load the ground would otherwise supply and in summer recharging the boreholes rather than wasting the heat. The data centre loses its cooling burden, the network builds less ground, and the community gets cheaper, steadier heat.
The mechanism is simpler than it sounds, and it's where the economics live. The network is one shared loop — a communal body of circulating water that every building and every source taps into. The data centre becomes another tap, putting heat in year-round instead of drawing it out. Its 40–50°C waste heat raises the temperature of the shared loop, and each building's heat pump draws from that warmer, easier source. For the data centre, this is the difference between fighting its heat and selling it. Feeding a loop requires the closed-loop cooling that cuts water consumption by roughly 80% against evaporative towers, therefore lowering operating costs by 8–18%.
The same heat reshapes the network's economics by shrinking the ground it has to build. The ground is not an infinite furnace: draw heat from it all winter and the rock around the boreholes cools faster than the earth replenishes it, so operators drill more boreholes, spread apart, purely to survive the coldest weeks and a large share of borefield cost exists only to meet that winter peak. A data centre feeding constant heat into the loop supplies part of that load directly, so the ground is drained less and fewer boreholes are needed. In summer, its surplus is pushed down into the rock, recharging the reservoir and cancelling the slow temperature drift that erodes efficiency. Because the ground is the network's dominant capital cost, this is the saving that matters most. Analyses of load-balanced borefields show on the order of 20–28% fewer boreholes and around 22% less drilled length for the same demand.
None of this works as an afterthought. The value only materialises if the data centre sits close enough to a thermal network for heat to move cheaply between them, and proximity, load profile, and offtake terms are set at siting, not retrofitted later. Waste heat can offset the ground a network would otherwise drill, and an adjacent network can turn a data centre's hardest constraint into a delivered product, but only while the site is still a choice on a map. As data-centre development accelerates, screening for nearby or planned thermal energy networks belongs in the earliest feasibility stage, alongside power and water. The cheapest time to capture this is before the site is chosen, which makes it a design question long before it is an engineering one.