Helsinki AI Data Center Plans to Send Waste Heat Into the City

AI infrastructure developer OnZero and Finnish energy company Helen have signed an agreement to connect a planned Helsinki AI data center to the city’s district heating network. Helen says the facility could eventually supply more than 500,000MWh of heat per year, enough for roughly 70,000 apartments. Heat recovery is scheduled to begin in 2027.

The project turns a familiar data center problem into a potential energy input.

AI density makes heat more valuable

High density AI hardware produces large quantities of heat in a concentrated space. That creates a cooling challenge, but concentration can also make heat recovery more practical.

OnZero says its design uses liquid cooled AI infrastructure and is intended to capture compute heat at temperatures suitable for district heating. The company has said the architecture can recover a very high proportion of the heat generated by the facility.

Liquid cooling is important here because heat carried in a fluid can be easier to transfer into another thermal system than low grade heat dispersed through large volumes of air.

For operators evaluating similar architectures, liquid cooling in data centers introduces operational variables including flow, temperature, pressure, pump health, leak detection, and coolant condition. Heat reuse adds another layer because the data center becomes connected to an external energy network.

Waste heat changes the boundary of the data center

Traditional data center efficiency discussions often stop at the facility boundary. Electricity enters. Compute runs. Heat leaves.

District heating changes that model.

If recovered heat can replace part of the energy that a city would otherwise generate for buildings, the data center starts participating in a larger energy system. That can affect project economics, sustainability reporting, and the relationship between the facility and its surrounding community.

It also creates dependency.

The heat recovery system must coordinate temperatures, flow, availability, maintenance, and seasonal demand between the data center and the district network. A good thermal design therefore has to consider both the IT environment and the external heat customer.

The fundamentals still begin with data center cooling systems, but heat reuse changes what happens after the heat is removed from the equipment.

Not every data center can copy Helsinki

District heat reuse depends heavily on location.

A facility needs a nearby heat network or another customer that can use the recovered energy. The temperature of the recovered heat matters. So does the distance between the data center and the heat user. Seasonal demand can also affect how much heat is useful at different times of year.

That makes Helsinki particularly interesting because district heating is already established infrastructure.

The project therefore should not be read as proof that every AI data center can become a heating plant. It is evidence that site selection and cooling design can create opportunities that do not exist in every market.

The efficiency question is getting bigger

AI infrastructure is forcing operators to think beyond the traditional question of how efficiently a facility consumes electricity.

A more complete question is what happens to the energy after the compute work is done.

If heat can be captured and reused at meaningful scale, the answer changes the role of the data center within the local energy system.

Helsinki’s project is still a planned deployment, and its real performance will become clearer once operations begin. But the idea is straightforward: AI compute will always create heat. The strategic question is whether that heat is treated as waste or infrastructure.

*Originally published on the Sensaka blog.*