Every watt of electricity that flows into an AI data center eventually becomes heat. This is not a flaw in the design. It is thermodynamics. Compute operations generate heat as an unavoidable byproduct, and the infrastructure built to manage that heat, cooling towers, chillers, air handlers, liquid cooling loops, represents a significant fraction of a facility's total cost and energy consumption. In a 60-megawatt data center, the thermal management challenge is substantial. In a 100-megawatt hyperscale facility, it is enormous.
What happens to that heat is, in most cases, a policy failure dressed up as an engineering inevitability.
The Evaporative Cooling Problem
The dominant cooling model in American data centers is still evaporative cooling, systems that use water to absorb heat and then release it into the atmosphere through cooling towers. This approach is effective, relatively inexpensive to build, and deeply problematic in water-stressed regions. A large data center using evaporative cooling can consume millions of gallons of water per year. In Texas, where drought conditions are increasingly common, the addition of large AI loads with evaporative cooling requirements is a compounding problem that utility regulators and environmental agencies have been slow to address.
Closed-loop liquid cooling eliminates most of the water consumption problem. It also produces something that evaporative cooling does not: a concentrated, usable heat stream. And this is where the policy failure becomes most visible.
"The thermal output of a large AI data center, properly captured, is a resource. District heating systems in northern Europe have demonstrated this for decades. In Texas, the opportunity exists but the infrastructure does not yet."
Heat That Could Be Recovered Is Instead Wasted
District heating systems in northern Europe have demonstrated for decades that data center waste heat can warm residential buildings, commercial spaces, and industrial facilities at meaningful scale. In Denmark, Finland, and Sweden, data centers are integrated into urban heating networks as heat sources rather than heat emitters. In Texas, that infrastructure largely does not exist, but the opportunity does. PVAMU's campus in Waller County includes buildings that require heating and cooling. A data center co-located on or adjacent to the campus could supply recovered heat to campus facilities, reducing the university's energy costs and demonstrating a model of thermal integration that has not been widely deployed in the American South.
EDC-3: Thermal Exhaust Recovery
The LegacyGrid framework designates thermal exhaust recovery as EDC-3, the third layer of the infrastructure model. It is positioned after the data center base layer and the battery storage layer because it depends on both. Permitting processes for large AI data centers should require applicants to evaluate thermal recovery options and to explain, in specific terms, why recovery is not feasible if they choose not to pursue it. The default should not be atmospheric dumping. The default should be recovery, with disposal as the exception that requires justification.
"Waste heat is a policy failure. It is also, in the right hands, a community asset waiting to be claimed."
Related reading: Batteries Are Not Backup · Water-Conscious Data Centers · Data Centers Should Clean Their Environments