Can the heat from AI data centres warm our homes?

A data centre needs to get rid of heat. A nearby neighbourhood may be paying to produce it. Connecting those two needs can turn computing’s unwanted output into hot water for buildings. The connection has to work in the real world: heat must arrive at a useful temperature, travel through pipes and find customers when it is available.[1]

A heating system with a computer building at one end

The familiar warmth of a busy laptop has a much larger counterpart inside a data centre. Much of the electricity used by the facility ultimately becomes heat. The International Energy Agency identifies opportunities to recover that energy for district heating, especially where buildings are close together and networks can accept lower-temperature supplies. The principle applies to ordinary computing as well as AI. A heat-recovery project alone does not establish which share came from running AI models.[1]

District heating supplies several buildings from shared heat sources through insulated pipes. Homes, offices and public buildings can draw from the same network. For a resident, that means the useful connection is to a heating service. There is no need for a direct pipe from an individual server to an individual radiator. The network joins sources and customers across the neighbourhood.[2]

Warm water sometimes needs a temperature lift

Odense, Denmark, provides a physical example. Ramboll designed an energy centre that connects surplus heat from Meta’s data centre to the local district-heating system. Its project account describes electrically driven ammonia heat pumps raising the recovered heat to approximately 70–75°C. That temperature lift makes otherwise less useful warmth suitable for the network. The equipment linking the two systems is a substantial part of the project.[3]

Alfa Laval, which supplied heat-exchange equipment, describes the earlier stage of the journey: coils transfer heat from the data centre’s warm air into water, which carries it to the heat-pump installation. The local district-heating utility financed the heat pumps and operates the station. That division of work is revealing. The computer operator has heat to remove; the utility supplies equipment and expertise to turn it into a heating product.[4]

A heat pump transfers heat using a refrigerant circuit, a compressor and heat exchangers. Electricity drives the process. Because the pump moves energy from another source, the heat it delivers can exceed the electrical energy it consumes. The coefficient of performance, or COP, expresses that relationship. Its value depends on the equipment and operating conditions, including the temperatures involved.[5]

The US Department of Energy defines COP as the heating or cooling effect divided by the net work input, measured in matching units. In the example, a quarter of the delivered heat is accounted for by the pump’s electrical input. That matters when assessing a project’s energy use: ten units of delivered heat do not mean ten units of recovered server heat. Annual totals also depend on how many hours the installation runs.[6]

A working heating plant can still be waiting for server heat

Fortum’s projects alongside Microsoft data centres in the Helsinki region show why dates and stages matter. On 6 May 2026, Fortum announced that heat production had started at plants in Kirkkonummi and Espoo. The operating sources were ambient-air heat pumps and electric boilers. Recovery of heat from the neighbouring data centres was expected to begin in phases from 2027. Calling the heating plants operational therefore described a real milestone, while the data-centre connection was a later step.[7]

Fortum expects the recovered heat eventually to supply about 40% of the annual district-heating demand in Espoo, Kauniainen and Kirkkonummi. That is a share of a particular local heating system. The announcement also describes a large hot-water storage tank, giving the system another way to separate the moment heat is produced from the moment customers need it.[7]

Summer changes the calculation

A server can need cooling on a mild summer night when nearby homes need little space heating. Fortum’s project overview expects to use about 75% of the data centres’ waste heat over a year, explicitly linking the unused portion to low summer demand. This percentage answers a different question from the projected 40% contribution to the heating network: one measures how much available heat is used; the other measures how much customer demand it supplies.[8]

Those two measures can move independently. In an original simplified example, recovering half of a source’s available heat could still meet all the needs of a small customer. Recovering nearly all of it could supply only a modest fraction of a large city’s demand. A useful project description needs both the supply and the demand, with matching periods.[8]

Who promises to keep the heating on?

A 2025 report prepared for the Greater London Authority examined commercial arrangements as well as engineering. Its interviews and analysis distinguish supplying surplus heat when it is available, selling a guaranteed heating service, and arrangements in which the heat network guarantees cooling to the data centre. Those obligations create different businesses. A network receiving heat only when available carries uncertainty about its quantity and timing. A promise to heat homes reliably requires a way to cover gaps. The report also examines how ownership and separation of the systems affect responsibility.[9]

This explains why recovering unwanted heat still costs money. In Odense, the utility’s investment and operating role sit between the raw heat and the customer. Pumps, electricity, maintenance and the network have to be paid for. The relevant economic comparison is the cost of supplying dependable heat through this arrangement against the other ways the local system could supply it.[4]

For a neighbourhood considering a connection, the revealing question is where its winter heat will actually come from. A data centre can become one useful source within a shared service. The strongest projects make the entire journey clear: whose heat is captured, how it reaches the required temperature, which buildings can receive it and what supplies them when that source is unavailable.[9]

Report an error or suggest a correction