The opportunity is real, but the operating model matters

A data centre can sell computing capacity, maintain cooling stability and still discard thermal energy that another facility could use. That apparent contradiction becomes more significant as AI workloads increase electricity demand and concentrate heat inside increasingly dense computing environments. Yet recovering that energy is not simply a matter of attaching a pipe to a cooling loop and sending warm water to a nearby building. The temperature, volume and timing of the available heat must match what a receiving facility can actually consume, while the data center continues to meet its primary operational requirements. Heat pumps, heat exchangers, thermal storage and distribution networks can make recovery practical, but each introduces equipment, control requirements and maintenance obligations.

Technical guidance from the International Telecommunication Union recognises low-grade heat, integration challenges and economic constraints as important considerations in data centre heat reuse. For operators, the relevant question is therefore not whether the heat has value in principle, but whether that value survives the cost and complexity of delivering it. For customers buying AI capacity, the distinction matters because a sustainability initiative should not introduce avoidable uncertainty into the service on which their applications depend.

Reuse cannot become another critical dependency

AI infrastructure already depends on coordinated power delivery, cooling, networking and equipment management, leaving little room for an additional system that complicates routine operations without delivering measurable benefits. A heat recovery installation can add pumps, valves, heat exchangers, sensors and control logic that engineers must commission, monitor and maintain. If those components interact poorly with the primary cooling architecture, a project intended to improve energy utilization could make fault diagnosis and maintenance planning more complicated. That does not make heat recovery inherently risky; it means engineers must separate the requirements of heat delivery from the requirements of compute cooling.

The data centre must retain a dependable way to reject heat even when the receiving building needs less warmth, a distribution network becomes unavailable or a heat pump requires maintenance. Engineers should design the recovery system so that it can be isolated without interrupting the facility’s ability to protect IT equipment. Operators should also define which team owns each control boundary, alarm, maintenance task and recovery procedure before the system enters service. The strongest project is not the one that captures the greatest theoretical quantity of heat, but the one that can do so without compromising the reliability of the computing service.

The economics begin outside the data hall

The commercial case depends on more than the quantity of energy leaving the cooling system, because useful heat requires a buyer with a suitable demand profile and a practical connection to the source. A neighbouring residential development may need substantial heating during winter but little space heating during warmer months, when the data centre continues to reject heat. Industrial processes, swimming pools, hospitals and hot-water systems may offer different demand patterns, although each application requires an assessment of temperature requirements, operating schedules and connection costs. District heating can expand the potential market, but a project may still require substantial investment in pipework, pumping equipment, heat pumps and network integration before any revenue becomes available.

European Commission assessments identify limited heat reuse, the availability of suitable customers and the infrastructure needed to transfer heat as important considerations in project feasibility. Distance also matters because transporting heat requires infrastructure, and the economics can deteriorate when the customer sits too far from the source. A prospective buyer should therefore be assessed as an energy customer with defined consumption, temperature and availability requirements, rather than treated as an automatic outlet for surplus heat. Without that commercial discipline, a facility can end up financing a heat delivery system whose utilization remains too low to justify its capital and operating costs.

Heat pumps can improve usefulness while increasing electricity demand

Much of the heat rejected by data centres sits below the temperature required for conventional building heating or industrial applications. Heat pumps address that mismatch by using electricity to raise the temperature of the recovered heat, making it more useful to the receiving system. Their performance depends on the temperature lift, equipment design, operating conditions and the temperatures required by the customer, so the additional electricity cannot be treated as a negligible input. The project must account for heat pump consumption alongside pumping, heat exchange, storage and distribution losses when calculating the net energy and emissions benefits.

A system may deliver substantial useful heat yet produce a less attractive financial or environmental result if it requires expensive electricity or displaces a relatively efficient existing heating source. Conversely, a low-temperature network or a nearby customer with consistent demand may allow the project to use recovered heat with less upgrading. Thermal storage can help balance differences between the time heat becomes available and the time a customer needs it, although storage adds equipment, space requirements and further capital costs. The practical objective is to maximise useful heat delivered per unit of additional energy and expenditure, rather than maximise the volume of heat captured at any cost.

AI infrastructure needs a measurable definition of useful heat

The amount of heat a data centre produces is not the same as the amount it can sell, reuse or reliably deliver to another facility. Operators need measurements that distinguish heat generated, heat captured, heat upgraded and heat actually consumed by an end user. That distinction becomes especially important when equipment operates at variable loads or when the receiving network cannot accept the full available output. A project that reports only the quantity of heat captured may overlook electricity consumed by heat pumps, thermal losses, maintenance downtime and periods of weak demand. The Energy Reuse Factor (ERF) measures the proportion of total data centre energy consumption that is reused outside the facility boundary, but it does not independently establish the financial or environmental value of that reuse.

A European Commission assessment published in 2026 reported that approximately 1.8% of the heat generated by data centres in the European Union was reused during the reporting period covered by the assessment. That figure provides important context, but it does not establish that every facility can achieve the same recovery potential because local infrastructure, climate and heat demand differ considerably. Buyers and operators should evaluate projects against actual delivered heat, additional energy consumption, availability, cost and avoided emissions rather than rely on a single headline efficiency metric.

Contract design must protect both sides

Heat reuse also introduces a commercial relationship that differs from the usual arrangement between a data centre and its electricity or cooling suppliers. A district heating operator may expect a predictable supply, while the data centre must retain the flexibility to adjust equipment, conduct maintenance and respond to changing computing demand. Neither party should assume that the other can absorb every operational change without financial or service consequences. Contracts should specify expected temperatures, delivery volumes, seasonal availability, planned maintenance windows, metering arrangements and responsibility for equipment failures. They should also establish what happens when the heat customer cannot accept supply or when the recovery system becomes unavailable, including whether the data centre can continue rejecting heat through its normal cooling infrastructure.

Commercial terms should distinguish guaranteed delivery obligations from indicative forecasts, particularly where the computing facility cannot reasonably guarantee a fixed thermal output under every operating condition. Clear responsibility for electricity costs, network maintenance and capital expenditure can prevent disputes over whether the project creates value for the data centre, the heat customer or both. For AI capacity buyers, this contractual clarity is relevant because the recovery project should not quietly transfer new reliability or operating risks into the infrastructure service they purchase.

Site selection should determine the heat strategy

External heat networks are not the only option available to operators seeking to recover thermal energy. Depending on the design and climate, recovered heat may support domestic hot water, nearby buildings, selected industrial processes or other applications that can use the available temperature. Internal uses may reduce the need for additional distribution infrastructure, although their energy savings depend on the existing heating system and the amount of useful demand. A site with suitable neighbouring facilities and an established heating network may support a different commercial model from a remote AI campus built primarily around access to power and land. This makes heat reuse a planning decision that belongs alongside site selection, cooling architecture and utility negotiations, rather than a sustainability feature added after construction.

Under Article 26(6) of the European Union’s Energy Efficiency Directive, Member States must ensure that data centres with a total rated energy input exceeding 1 MW use waste heat or another heat-recovery application unless technical or economic infeasibility is established under the applicable assessment provisions. That approach recognises both the potential value of recovered energy and the need to assess local demand, technical feasibility and economic viability. Developers should evaluate the heat market before committing to a site, because a nearby customer with a compatible demand profile can matter more than a larger theoretical heat resource with no practical route to market.

Sustainability claims need a full-system calculation

Heat recovery can contribute to lower emissions when it displaces higher-emission heating, but the result depends on what the receiving customer would otherwise use and how much additional electricity the recovery system consumes. The calculation should account for heat pump efficiency, electricity emissions, distribution losses, backup heating and the emissions associated with the displaced energy source. It should also distinguish between energy that would otherwise have been rejected and energy that the data centre must continue consuming to operate the recovery equipment. A project that delivers heat during periods of genuine demand may provide a meaningful benefit, while one that supplies heat when customers do not need it may deliver little practical value.

Operators should avoid presenting all recovered heat as an automatic emissions reduction, particularly when the receiving facility already uses a low-carbon heating source. Instead, performance reporting should disclose the baseline, the volume of useful heat delivered, additional electricity consumption and the method used to estimate avoided emissions. This approach gives customers, investors and local authorities a more defensible basis for evaluating the project’s environmental contribution. It also creates a commercial incentive to improve the system’s actual performance rather than optimise the appearance of its sustainability figures.

Heat reuse should earn its place in the operating model

AI data centres will continue to face pressure to demonstrate that their energy consumption produces value beyond computing output, but waste heat recovery should not become another infrastructure obligation measured solely by installed equipment. A viable project must connect a reliable heat source with a customer whose demand, temperature requirements and willingness to pay justify the cost of delivery. It must also preserve cooling redundancy, establish clear ownership of new equipment and remain manageable during maintenance, seasonal changes and fluctuations in AI workloads. For developers, the decision begins with location, thermal design and a credible assessment of potential customers, not with an assumption that every megawatt of rejected heat represents a saleable product.

While Operators, success depends on measured delivery, manageable maintenance requirements and a clear financial case that survives realistic operating conditions. For AI buyers, the priority is to understand whether heat recovery improves the infrastructure’s overall efficiency without introducing hidden costs or weakening service reliability. The industry should therefore treat useful heat as a potential secondary product whose value must be demonstrated, not as an automatic benefit of consuming electricity. The real test is whether a data centre can recover heat, deliver it when needed and earn a measurable benefit without making reliable computing harder or more expensive to operate.