What if the giant AI data center near your city could help heat your shower?
It sounds strange at first.
We usually talk about data centers as electricity-hungry machines. They buy huge amounts of power, run thousands of processors, and then spend even more energy trying to keep those processors cool.
But there is another way to look at the same system.
A data center does not simply consume electricity.
It also produces something continuously:
heat.
Almost all of the electricity used by a data center eventually ends up as heat. The International Energy Agency says roughly 70–80% of that heat can potentially be recovered with heat pumps.
That leads to a surprisingly simple question:
If we are already paying to remove the heat, why not sell it to someone who needs heat?
That idea is beginning to move from small demonstration projects into much larger energy systems.
Every AI Calculation Eventually Becomes Heat
Think about what happens inside a server.
Electricity enters a GPU.
The GPU performs billions of calculations.
But the electricity does not disappear after the answer is generated.
It ends up mostly as heat inside chips, memory, power supplies, networking equipment, and the cooling system.
That is why cooling is such a big part of data-center design.
In the previous article, Smaller AI Models Could Change the Data Center Boom, we looked at how more efficient AI can reduce the electricity needed for each task.
Here we are asking a different question.
What can we do with the energy after the computer has already used it?
Normally, the Heat Is Just a Problem
A conventional data center has one main goal after the servers create heat:
get it out of the building.
Cooling systems move the heat from the chips into air or liquid and then release it to the outside environment.
From the data center’s point of view, that is success. The servers stay within safe temperatures.
From an energy-system point of view, however, something valuable may have been thrown away.
Imagine a city only a few kilometers away burning gas, biomass, or using electricity to heat buildings while a huge data center nearby is simultaneously throwing heat into the air.
Those two systems do not have to remain separate.
District Heating Is the Missing Pipe
This idea works best in places with district heating.
A district-heating system is basically a city-scale network of insulated pipes.
Hot water moves through the network to homes, offices, schools, hospitals, and other buildings.
Instead of every building producing all of its own heat, many heat sources can feed one shared network.
Traditionally, that heat might come from a power plant or boiler.
But a modern network can also accept heat from:
- industrial processes,
- wastewater,
- large heat pumps,
- geothermal systems,
- and data centers.
This changes the role of the data center.
It is no longer only an electricity consumer.
It can become a heat supplier.
But Server Heat Is Often Not Hot Enough
There is one catch.
The heat leaving a data center is often relatively low-temperature heat.
Warm water from a cooling loop may be useful, but it may not be hot enough to feed directly into an older district-heating network.
This is where a large heat pump enters the story.
A heat pump works like a temperature elevator.
It takes a large amount of low-temperature heat and uses additional electricity to raise that heat to a more useful temperature.
So the flow becomes:
Servers → cooling water → heat pump → district-heating water → buildings
The International Energy Agency notes that lower-temperature district-heating networks make this kind of waste-heat recovery easier, while heat pumps can upgrade lower-temperature sources when necessary.
Finland Is Building This at City Scale
One of the most interesting examples is now taking shape around Espoo and Kirkkonummi in Finland.
Microsoft is building large data centers there, while Finnish energy company Fortum is building the infrastructure needed to capture their waste heat and feed it into the local district-heating network.
There is an important update as of 2026.
In May 2026, Fortum announced that large heat-pump plants at Hepokorpi in Espoo and Kolabacken in Kirkkonummi had started operating.
For now, those plants can produce heat using ambient air and electric boilers.
The Microsoft data-center waste heat is expected to be connected gradually beginning in 2027 as the data centers are commissioned.
Once the system is fully developed, Fortum expects waste heat from Microsoft’s data centers to cover roughly 40% of district-heating demand in the network area.
The IEA describes the project in even more intuitive terms: the cluster could eventually provide enough recovered heat for around 100,000 homes.
That is no longer a tiny experiment.
It is energy infrastructure at city scale.
The Interesting Part Is the Business Model
The environmental story is easy to understand.
If a city can reuse heat that would otherwise be discarded, it may need less heat from other sources.
But for The Contexta, there is another question:
Can that heat become money?
In some places, yes.
Stockholm already operates a system called Open District Heating.
Companies such as data centers, supermarkets, and other businesses can deliver excess heat into Stockholm’s district-heating network.
And they can be paid for it.
Stockholm Exergi says its payment is linked to what equivalent heat would have cost to produce in its own plants. As an example, it says a supplier delivering about 1 MW of heat can receive roughly SEK 2 million per year, although the actual payment varies with factors such as outdoor temperature.
This turns the basic energy flow into something new:
Electricity → Compute → Heat → Heat Sale
The data center’s first product is still computing.
But in the right city, heat may become a second product.
This Does Not Mean Waste Heat Is Free Money
If it were that easy, every data center would already be heating a city.
The problem is that heat has very little value if nobody nearby needs it.
The IEA emphasizes that heat-recovery projects need to be evaluated site by site.
Distance matters.
Temperature matters.
Existing infrastructure matters.
And someone has to pay for heat exchangers, heat pumps, pipes, pumps, controls, and maintenance.
Condition 1: Someone Must Be Close Enough to Buy the Heat
Heat is not like software.
You cannot send it around the world through the internet.
The farther hot water has to travel, the more infrastructure is needed and the more heat can be lost.
That is why proximity to a city or an existing heat network matters so much.
A recent 2026 engineering study showed that longer-distance transport can be technically and economically possible for very large heat sources, but the result depends heavily on system design, scale, storage, and local economics.
For most real projects, the simple rule remains useful:
Waste heat is most valuable when the heat customer is nearby.
Condition 2: The Network Temperature Has to Work
Older district-heating systems often operate at relatively high temperatures.
Data-center waste heat may be much cooler.
That temperature gap means a heat pump has to work harder.
Newer, lower-temperature district-heating networks are generally a better match for low-grade waste heat.
This is one reason the design of the city’s heating system matters almost as much as the data center itself.
Condition 3: Winter and Summer Are Different Businesses
A data center produces heat all year.
A city does not need the same amount of heating all year.
In northern Europe, winter demand can be enormous.
Summer demand can be much lower.
Fortum expects that about 75% of the Microsoft data centers’ annual waste heat could ultimately be used for district heating, with low summer heat demand limiting how much can be absorbed.
That mismatch creates another infrastructure question:
Can the system store heat?
Can it serve hot-water demand in summer?
Can other industrial users take the heat?
Or does some of it still have to be rejected?
This is why thermal storage is becoming part of the waste-heat conversation too.
Condition 4: The Contract Has to Make Sense
Even if the engineering works, somebody still has to answer uncomfortable business questions.
Who owns the heat pump?
Who pays for the pipe connecting the data center to the city network?
Who guarantees the heat supply?
What happens if the data center changes its computing load?
What price is paid for the heat?
Who benefits from lower cooling costs?
A 2025 review of data-center waste-heat projects found that technical feasibility is only one part of the challenge. Infrastructure costs, regulation, market design, and long-term contracts can determine whether a technically good project is financially attractive.
In other words:
There must be a business model connecting the server rack to the radiator in someone’s home.
Europe Is Starting to Measure This More Seriously
The European Union is also putting more attention on data-center energy reuse.
Its data-center reporting framework includes an indicator called the Energy Reuse Factor, or ERF.
The idea is straightforward.
Instead of measuring only how much electricity a data center consumes, ERF asks how much energy is actually reused outside the data-center boundary.
That matters because two data centers with similar power consumption may have very different relationships with the surrounding city.
One may dump almost all of its heat into the environment.
Another may send a large share into homes and buildings.
As data centers become larger, that difference becomes harder for cities and regulators to ignore.
This Could Change Where Data Centers Are Built
Traditionally, a good data-center location might be described with a short list:
- cheap and reliable electricity,
- strong grid connection,
- fiber connectivity,
- enough land,
- acceptable cooling conditions.
Waste-heat reuse adds another item:
a nearby heat customer.
Fortum says the Microsoft sites in Finland were selected with waste-heat recovery in mind, including access to the district-heating network, electricity, land, and data connectivity.
That is a subtle but important shift.
The city next to the data center is no longer only a source of electricity demand, construction permits, and employees.
It can also become a customer for one of the data center’s by-products.
The New Data-Center Equation
The old picture looked like this:
Electricity → Compute → Heat → Waste
The circular version looks different:
Electricity → Compute → Heat → District Heating → Useful Energy
And in a functioning heat market:
Electricity → Compute → Heat → Revenue
That does not eliminate the enormous electricity demand of AI.
It does not make data centers automatically green.
And it does not work everywhere.
But it changes the economics of one part of the system that used to be treated only as a cost.
The Simple Lesson
A data center is basically a machine that turns electricity into two things:
computing and heat.
We already know how to sell the computing.
The interesting question is whether we can also find a buyer for the heat.
In cities with the right district-heating infrastructure, the answer is increasingly yes.
That leads to one of the most useful ideas in this entire series:
Waste heat has value only when someone nearby needs it.
That one sentence explains why Finland and Stockholm are interesting, why heat pumps matter, why summer demand matters, and why future data-center location decisions may look different from today’s.
What to Watch Next
We have now looked at several very different answers to the same problem.
Put some computing in space.
Move some AI onto phones and PCs.
Make models smaller and more efficient.
Reuse the heat from the data centers that remain.
None of these solutions wins everywhere.
That suggests a final question for this series:
Cloud, Edge, or Orbit: Where Should AI Computing Actually Happen?
Key Vocabulary
Waste heat
Heat produced as a by-product of another process. In a data center, electricity used by IT equipment ultimately becomes heat that must be removed.
District heating
A network that distributes hot water or steam from central or distributed heat sources to many buildings.
Heat pump
A machine that uses electricity to move heat from a lower temperature to a higher, more useful temperature.
Low-grade heat
Heat available at a relatively low temperature. It may be useful directly in some systems or may need a heat pump before use.
Energy Reuse Factor (ERF)
A metric that compares energy reused outside a data center with the data center’s total energy consumption.
Related Articles
- Can We Put AI Data Centers in Space?
- Space Is Cold. So Why Is Cooling a Data Center in Orbit So Hard?
- Will Edge AI Reduce the Need for Giant Data Centers?
- Smaller AI Models Could Change the Data Center Boom
- How Much Power Is 100 MW? An AI Data Center Compared with Entire Cities
Sources
- International Energy Agency — Opportunities for District Heating in the Changing Energy Landscape, 2025
- International Energy Agency — Renewables in District Energy, 2026
- Fortum — Heat Production Starts at Two Large Data-Centre Sites in Finland, May 2026
- Fortum — Fortum and Microsoft Data-Centre Heat-Recovery Project
- Stockholm Exergi — Open District Heating / Heat Recovery
- European Union — Delegated Regulation (EU) 2024/1364: Data-Centre Sustainability Indicators
- Renewable and Sustainable Energy Reviews — Data Center Waste Heat for District Heating Networks: A Review, 2025
- Energy — Large-Scale Long-Distance Data Center Waste Heat District Heating System, 2026
Published: August 2026 · Sources checked through: August 2026 · Actual heat-recovery performance and economics depend on cooling design, heat temperature, network temperatures, distance, local electricity and heat prices, seasonal demand, infrastructure cost, and contractual structure.