A new study from the University of Utah quantifies what flexible data center operation is worth to North America's western power grid, and cooling sits at the center of the math. According to TechXplore's coverage of the peer-reviewed paper, researchers led by professor Masood Parvania and power systems engineer Mohammad Amin Mirzaei modeled three forms of flexibility across the Western Interconnection, the 136,000-mile network that serves more than 80 million people across 11 U.S. states, two Canadian provinces, and part of Mexico. Stacking all three approaches produces an estimated $590 million in annual operating savings, or about $468,000 per day, plus roughly 3.5 million fewer tons of CO2 each year.
The model separates flexibility into temporal shifting, spatial coordination, and on-site energy resources. Temporal flexibility alone is worth $62 million a year. Adding spatial coordination, moving compute between regions to follow cheaper or cleaner generation, lifts the figure to $171 million. The on-site layer, which includes solar, small modular reactors, and battery storage, carries the total to $590 million.
The temporal lever is where cooling earns its place in the study. The researchers model cooling systems augmented with thermal energy storage, letting a facility pre-cool or bank chilled capacity when power is abundant and coast through peak grid hours. That turns the cooling plant from a fixed draw into a dispatchable buffer. The same logic underpins the trade-offs we have covered between water and power in data center cooling, where the choice of heat rejection method dictates how much a site can flex without breaching thermal limits.
Treating cooling as grid-responsive load is a different engineering target than treating it as a constant. A site built to shift load needs chilled-water volume, tank capacity, or phase-change mass sized for hours of ride-through, and a control layer that ties setpoints to grid price signals. Air-cooled halls have limited room to bank thermal mass, which pushes the economics toward liquid loops where coolant inventory and tank storage are easier to scale. That reinforces the broader move toward liquid cooling as the standard architecture at high rack densities.
The savings also depend on coordination that most operators do not yet have. Spatial flexibility assumes workloads can migrate between regions on the timescale the grid needs, and that cooling capacity at the receiving site can absorb the inbound load. The study was presented at the 2026 Hawaii International Conference on System Sciences and frames flexibility as a grid-planning input, which matters as developers face mounting grid constraints on new data center projects.
For operators, the Utah numbers reframe thermal storage from an efficiency nicety into a revenue and siting argument. A cooling system that can shift hundreds of megawatts of load off peak becomes a grid asset, and the $590 million figure is the size of the prize the whole interconnection leaves on the table when cooling runs flat. The question for cooling vendors is whether the next generation of plants ships with the tank capacity, controls, and liquid inventory to make that flexibility real, or whether it stays a modeling exercise.