Clarity for AI buildersSponsored by AI Routing
adoption

Sweden’s data centres link AI computing to local heat networks

Sweden’s data centres link AI computing to local heat networks

Sweden’s data centres show how cleaner power and heat reuse can reduce local impacts, according to a Guardian report.

The newspaper’s feature on EcoDataCenter’s facility in Falun examines an approach that combines timber construction, renewable electricity and recovered heat with municipal planning. It offers a counterpoint to disputes over data centres’ resource demands, rather than evidence that their environmental costs have disappeared.

At Falun, the relationship with the community predates the current AI boom. According to John Wernvik, EcoDataCenter’s chief external relations and sustainability officer, the municipality initiated the project in 2010. EcoDataCenter was formed in 2014, and the first part of the facility began operating in 2019. Further phases remain under construction.

That history matters to the model described in the report: the facility was planned around local infrastructure and needs, rather than treating community benefits as an addition after construction. Wernvik says 90% of its suppliers are sourced locally and 60% of its capital expenditure is spent locally. Those are company statements reported by the Guardian, not independently verified findings supplied with this draft.

Different measures address different costs

The Falun facility uses cross-laminated timber panels. Wernvik says this reduces the structure’s embodied carbon compared with conventional steel-and-concrete construction. Its electricity supply is reported as entirely renewable, comprising 75% hydroelectric power and 25% wind.

These measures address distinct parts of the environmental burden. Timber concerns the emissions associated with construction; the electricity mix concerns ongoing operation. Neither, by itself, establishes how efficiently the computing equipment uses power or how much energy the facility consumes overall.

For operational efficiency, the Guardian reports a power usage effectiveness, or PUE, of 1.15 for EcoDataCenter, compared with an EU average of 1.36. A value closer to 1 indicates less power used beyond the computing equipment itself. The report also gives a carbon usage effectiveness figure of 1.17gCO2/kWh, against 263gCO2/kWh for the EU average. These comparisons are reported figures; the supplied evidence does not establish identical measurement periods or accounting boundaries.

Absolute demand remains substantial. The Guardian describes Falun’s capacity as 80MW, while Wernvik says AI accounts for 70% of its computing power. Efficiency and cleaner electricity therefore reduce particular impacts without eliminating the need for a large, reliable power supply.

Turning cooling into a local resource

Sweden’s cold climate reduces cooling requirements. At Falun, the report describes air-based cooling alongside closed-loop liquid cooling for newer chips. Wernvik says the latter system does not draw external water into its cooling loop. That claim should not be read as saying the entire facility uses no water: the feature also describes water-assisted cooling elsewhere on site.

Recovered heat supplies wood-pellet drying and Falun’s district heating network. This changes the role of cooling: some energy that must be removed from servers becomes useful to nearby consumers. But it requires a destination for that heat, not simply equipment capable of collecting it.

The same dependency appears in Stockholm. According to the Guardian, roughly 20 data centres connected to district heating contribute about 1.5% of the city’s annual heat demand. The contribution is useful but limited; it does not make data centres a substitute for the wider heating system.

Expansion tests the approach

EcoDataCenter’s next site, in Borlänge, is being built on a former paper mill site with existing industrial infrastructure. The Guardian reports a planned capacity of 240MW across 200,000 square metres, with the first phase due to open in 2027. Wernvik projects employment of 200 to 300 people. These are plans and forecasts, not completed operating results.

Scale also creates a geographic constraint. atNorth chief executive Eyjólfur Magnús Kristinsson tells the newspaper that larger facilities tend to need more remote locations, where fewer nearby customers need their heat.

The practical lesson is therefore conditional. Sweden’s approach connects computing to electricity, industrial sites and heating networks developed over decades. Other locations may adopt individual elements, but the report does not establish that the complete model can be reproduced quickly—or that lower-impact construction resolves the separate question of how much additional computing demand should be served.