Data Center Heat Reuse and District Heating: EU Rules, Nordic Examples and U.S. Site Implications

Almost every watt delivered to servers ends up as heat that the cooling system must reject, usually to outdoor air through dry coolers, chillers or cooling towers. Heat reuse captures some of that energy and sends it to a customer that would otherwise burn fuel or run electric heat: a district heating network, a greenhouse, an office building, a pool, or an industrial process that needs warm water.

The obstacle is temperature. Server heat is plentiful but low-grade, often too cool to feed a conventional district heating network directly. IBM researchers noted years ago that typical liquid-cooled facilities run near 45°C, which they described as too low to heat homes, and built a prototype that returns water at about 60°C instead.¹¹ Most projects bridge the gap with large electric heat pumps, which use the data center heat as their source and deliver water hot enough for the network.⁴¹²

Heat reuse is reported with the energy reuse factor (ERF), the share of a facility’s energy that is reused outside the data center. Germany’s Energy Efficiency Act uses this metric for its targets.³ It sits alongside power usage effectiveness, covered in our guide to PUE and data center energy efficiency, and the two measure different things: a site can have a good PUE and reuse nothing.

The recast Energy Efficiency Directive, Directive (EU) 2023/1791, is what turned heat reuse from a sustainability project into a siting requirement in Europe.¹ Article 26(6) requires member states to ensure that data centers with a total rated energy input above 1 MW use their waste heat or other waste heat recovery applications, unless they can show it is not technically or economically feasible.¹ Article 26(7) requires an installation-level cost-benefit analysis when such facilities are planned or substantially refurbished, covering technical feasibility, cost-efficiency and local heat demand, including seasonal variation.²

The European Commission’s 2024 recommendation on Article 26 says the 1 MW obligation applies to operating data centers that do not already use their waste heat, and that the cost-benefit analysis is the test: if it shows reuse is not feasible, the facility does not have to reuse heat.² Member states may also set exemption thresholds based on the amount of useful heat, the heat demand, or the distance to a district heating network.² Separately, Article 12 requires data centers with an installed IT power demand of at least 500 kW to report energy performance and sustainability data each year.¹ Uptake is still low: a July 2025 Commission technical report estimated that only about 1.9% of the heat generated by EU data centers is reused.¹³

Germany goes further. Its Energy Efficiency Act (Energieeffizienzgesetz, or EnEfG) applies to data centers with a non-redundant nominal connected capacity of 300 kW or more.³ New data centers must meet energy reuse factors of 10% from July 2026, 15% from 2027 and 20% from 2028.³ An amendment adopted by the federal cabinet in June 2026 would ease several of these duties, including raising the threshold to 500 kW of installed IT power, and was still in the legislative process as of autumn 2026.⁸ Until it is settled, a heat customer, or a credible plan for one, remains part of German site selection for new capacity.

Denmark and Finland have the clearest working examples because their cities already run large hot-water district heating networks, so the customer, the pipes and the operator already exist.

In Odense, Denmark, Meta announced in 2018 that its data center would capture excess server heat and feed it to the local network run by Fjernvarme Fyn, recovering an estimated 100,000 MWh a year, enough for about 6,900 homes.¹⁴ Ramboll, which engineered the plant, describes a system in which district heating water absorbs the heat from server exhaust and electrically driven ammonia heat pumps raise it to 70–75°C, with total heat production of about 45 MW.⁴

In Finland, Microsoft and Fortum announced in 2022 that Fortum would build a large waste heat recovery unit connected to Microsoft’s planned data centers in the Helsinki region and feed the heat into its networks in Espoo, Kauniainen and Kirkkonummi. The partners estimated the heat would cover about 40% of the district heating needs of roughly 250,000 users in those three municipalities.⁵

In both cases the utility, not the data center operator, owns the heat pumps and the network and sells the heat. The data center’s role is to make the heat available at an agreed point, temperature and volume.

How much heat a site can sell depends heavily on how it is cooled. Air-cooled halls deliver heat in warm return air or a cool water loop, which needs a large temperature lift. Liquid cooling captures heat closer to the chip and at higher temperatures. An FEV industry presentation summarized by Fuels & Lubes reported direct-to-chip coolant outlet temperatures of 45–60°C and single-phase immersion at about 40–55°C.⁹ A 2026 ACEEE Summer Study paper on a “data-center village” heating network cites work showing direct-to-chip cold plates can capture up to 94% of server heat in the liquid.¹⁵

The gap between those numbers and what a network needs is closed with heat pumps, and the electricity they use is part of the economics. National labs are studying the step up: Oak Ridge National Laboratory modeled high-temperature heat pumps that would use waste heat from its Frontier supercomputer to supply hot water to the lab’s campus district heating, comparing cycle designs and refrigerants for efficiency and capacity.¹² Our guide to air vs. liquid cooling covers how the cooling choice affects the rest of the site.

The United States has no federal heat reuse mandate comparable to the EU’s, so a U.S. project happens only when someone nearby wants the heat and the numbers work. The bigger constraint is the customer base. An EIA-commissioned study drew on an industry database of 660 district energy systems and estimated they served about 5.5 billion square feet of heated floor space in 2012.¹⁰ The Department of Energy notes that U.S. systems are mostly on university, hospital and research campuses, military bases and airports, and in dense downtowns such as New York, Boston, Philadelphia, San Francisco, Denver and Minneapolis.⁶ Systems that distribute steam or high-temperature water are harder to serve from low-grade heat than modern low-temperature networks.

The best-known U.S. example is urban. In Seattle, heat from the Westin Building Exchange, a downtown building where about 70% of 400,000 square feet is data center space, is upgraded by heat pumps and piped under the street to heat Amazon’s nearby campus. The agreement allows up to 5 MW of heat to be supplied, and projected savings were about 4 million kilowatt-hours a year.⁷ Arrangements like this depend on density, which is the opposite of the large rural parcels where most new AI campuses are being built.

Smaller local uses, such as heating the site’s own offices, greenhouses, aquaculture or a neighboring industrial user, are more common than district heating in the U.S. and need no utility partner, but they absorb only a small share of a large campus’s heat.

For a European site, heat reuse is now part of feasibility: the cost-benefit analysis looks at local heat demand and distance to networks, so a parcel near a district heating system may carry an expectation, or in Germany, under current law, a requirement, to export heat.²³ For a U.S. site it is usually an option rather than a constraint, but it can matter in public approvals, where a credible heat partner is one of the few community benefits a data center can offer beyond taxes.

If heat reuse is on the table, check these items early:

• Offtaker and distance: who would buy the heat, how far away they are, and whether a pipe route exists across roads, rail and other owners’ land, which may need easements.
• Seasonal match: heat demand peaks in winter while data center heat is constant, so summer heat still has to be rejected on site. The EU analysis explicitly considers seasonal variation.²
• Space: room for heat exchangers, heat pumps and pumping, often operated by the heat buyer, near the cooling plant and equipment yard.
• Power: heat pumps add electrical load that may need to be included in the utility service request.
• Reliability: the data center must be able to reject all of its heat without the heat customer, so reuse equipment adds to, and does not replace, normal cooling.
• Contracts: who owns the heat recovery plant, the delivery temperature and volume, and what happens during outages on either side, as in the utility-owned Nordic models.

For most U.S. greenfield campuses the practical step is modest: confirm whether a heat customer exists, and if one might in the future, leave space and a route for it in the campus master plan. If you want help weighing that against power, water and land, you can get a site reviewed.