From Server Halls to Steam Mains: Britain's Overlooked Thermal Export Opportunity

The UK's data centre estate has spent a decade being discussed as a source of electricity demand, grid constraint and local opposition. In 2026, a different question is moving up the agenda: what happens to the heat those buildings produce? As AI compute clusters raise rack densities and cooling loads, operators, planners and infrastructure investors are beginning to treat waste heat not as an unavoidable by-product but as a saleable output. If that shift holds, it could do more than improve the energy balance of the UK's digital economy. It could create a modest but durable export category in thermal engineering, controls and heat-network delivery.

The new geography of compute and heat

Most legacy data centres reject heat at low temperatures. Air-cooled halls may exhaust air at 30–40°C, which is useful for little more than preheating or low-temperature greenhouses. Newer liquid-cooled and direct-to-chip systems can produce return water at 45–60°C, and with careful design some high-density AI halls can run warmer. That is still below the 70–90°C flow temperatures that many UK district heating networks need, particularly in older buildings. The engineering gap is therefore not a reason to dismiss reuse; it is the commercial opportunity.

High-temperature heat pumps can upgrade low-grade data-centre heat to network temperatures, albeit with a coefficient of performance that falls as the temperature lift rises. Sorption chillers offer a complementary route, using waste heat to drive cooling and reduce electrical demand in the same facility. Thermal storage can decouple compute load from heat demand, allowing a data centre to supply heat at times when the network needs it rather than only when servers are busy. The result is a more complex thermal system, but one that can be designed, modularised and sold.

Engineering the handover

The technical work begins at the interface. A data centre is designed for uptime, not for the variable flow and return temperatures of a district heating network. Heat exchangers, filtration, isolation valves, metering and controls must be arranged so that heat recovery can be isolated without compromising the primary cooling loop. Legionella risk management and water chemistry matter, particularly where existing cooling towers are involved. Retrofit projects face tighter plant-room constraints than new-build halls, which is why skid-mounted heat-recovery modules are attracting attention.

There is also a contractual engineering problem. Heat networks need anchor loads and predictable offtake; data centres need predictable cooling. A heat-recovery system that depends on a single network customer is a reliability risk. The more robust projects involve thermal storage, backup heat sources and clear operating envelopes. These are unglamorous details, but they determine whether a project reaches financial close.

Policy is catching up

UK policy has moved from encouragement to structure. The Energy Act 2023 introduced heat network zoning in England, and by 2026 local authorities are working through designation and delivery models. Ofgem's heat network regulation is intended to give customers and investors more confidence in pricing and service standards. Planning authorities increasingly ask data-centre applicants to demonstrate that waste heat has been considered, even if the resulting obligation is not yet uniform.

The missing piece is bankable offtake. Heat network developers need long-term price certainty; data-centre operators need flexibility. A zoning framework can help by identifying where heat demand is dense enough to justify a network, but it cannot manufacture a contract. That is why the most credible schemes pair a data-centre operator with a heat-network developer, a local authority and an engineering, procurement and construction partner at an early stage. The project is as much an infrastructure transaction as a technology sale.

What investors should watch

For venture investors, the opportunity sits in components and controls rather than in another data-centre developer. High-temperature heat pumps, sorption systems, thermal stores, heat exchangers and control software all have a role. The strongest propositions will have measured performance data from real installations, not just laboratory coefficients of performance. Investors should ask for levelised cost of heat, capex per megawatt of recovered heat, maintenance intervals and evidence that the system can operate for thousands of hours without derating.

For infrastructure investors, the signal is contracted heat offtake. A project with a creditworthy network operator, a defined temperature envelope and a clear allocation of construction and performance risk is more investable than a technical pilot with a memorandum of understanding. The UK's heat-network market is still small compared with those in Denmark or the Netherlands, but it is developing. The export case rests on British firms learning to package design, financing and operation into repeatable offers.

An export category in waiting

Britain is not going to out-manufacture every heat-pump producer. It can, however, export engineering services, controls, modular heat-recovery plant and project-delivery expertise to markets where data-centre growth and district-heating ambition overlap. The UK's strengths in process engineering, standards and financial services are relevant here. So is the country's experience of retrofitting constrained urban sites.

The risk is that heat reuse remains a collection of worthy pilots. The prize is a thermal-engineering sector that treats data-centre heat as a product. In 2026, the AI build-out is often framed as a power problem. It is also a heat opportunity, and the UK should be building the capability to sell it.