Summary
- Canaan will supply Avalon A1566HA hydro cooled units for an 8MW Nordic district heating deployment.
- The project includes 2MW already operating and a 6MW follow on order, with claimed heat supply for about 2,800 homes.
- The deployment tests whether high temperature compute heat can be integrated into local energy infrastructure at useful scale.
Canaan will supply hydro cooled Bitcoin mining equipment for a Nordic district heating project, using compute waste heat as a source of hot water for local homes.
The deployment uses Canaan’s Avalon A1566HA hydro cooled units with a combined capacity of about 8MW.
Around 2MW, comprising 228 units, is already operating in the region, while a follow on order placed in March 2026 adds a further 6MW through 692 additional units.
Canaan says the equipment can produce hot water at approximately 80°C, allowing direct integration with existing district heating systems.
Once fully deployed, the company expects the project to provide reliable heating for about 2,800 homes.
The customer, exact location, commercial terms, heat offtake arrangements, power source, and baseline heating fuel being displaced have not been disclosed.
Those details will decide the eventual energy and emissions value of the scheme, because heat reuse depends on the electricity source and the local heat system it replaces or supports.
The project nevertheless gives a useful view of how liquid cooled compute can be connected to thermal infrastructure where district heating networks already exist.
High temperature heat improves the case
Many data centre heat reuse schemes are limited by temperature, because low grade rejected heat may require heat pumps or additional plant before it can serve a district heating network.
Canaan’s claim of approximately 80°C hot water changes that equation, since higher temperature output can be more useful to network operators and easier to integrate into existing systems.
The architecture differs from a boiler plant, because the heat source is a fleet of parallel compute units rather than a single thermal generation asset.
Canaan says the heating nodes can be dynamically overclocked or underclocked, giving operators some flexibility in how heat and compute output are managed.
Seasonal demand still has to be managed carefully, because heat demand rises in winter and falls in summer while compute economics and electricity prices can move on different cycles.
The electricity source behind the load will be central to the environmental assessment, since low carbon power and displacement of fossil heat would produce a stronger case than grid constrained or high carbon supply.
Nordic district heating networks give this type of project a better starting point than many other markets, as pipe infrastructure, heat customers, and colder weather can support more sustained offtake.
AI data centres may face similar questions as liquid cooling becomes more common, since denser compute can produce more concentrated heat streams that are easier to capture but not automatically easy to sell.
Canaan’s project does not prove that all compute waste heat can be reused profitably, but it offers a practical example of how digital infrastructure can be designed around a local energy system rather than treated only as an electrical load.
