What lives on the hall floor.
The tenants on GreenGridLabs sites run high-performance computing and artificial intelligence workloads — foundation-model training, large-scale inference, scientific and industrial simulation, enterprise research clusters. They come in through framework agreements with the local operating entity, bring a well-defined workload profile, and run the jobs they said they would run. That is the shape of the hall, and that is who the building was designed for.
Picking this lane is not a constraint — it is the reason the economics work. The best sites in the Nordics for clean power, grid services and heat recovery are the ones that know their workload in advance. The rack layout, the cold-plate choice, the cooling loop and the waste-heat offtake all depend on knowing what the hardware is going to do every hour of every day.
Density, cooling and flexibility follow the workload.
Three things fall out of a knowable workload profile, and all three are the reason a customer chooses a site like Dalsbruk over a generic colocation hall:
- Density. When we know the hardware, we know the thermal envelope, and we can push rack density much higher than a mixed-tenant hall can. More FLOPS per square metre, fewer square metres per megawatt, a cleaner build.
- Heat recovery that actually works. A compute hall whose temperatures are designed around the local district heating network can feed real thermal energy into it. That only works if the workload cooperates. Ours does.
- Grid-side flexibility. A managed workload with a defined response envelope can be co-dispatched with the on-site battery to deliver flexibility services to the Nordic grid. The tenant sees no impact; the site earns a second revenue line.
Onboarding the way it should be.
A new tenant comes in through the local operating entity, not through a remote sales desk. The process is short and boring: a counterparty screening against the consolidated EU sanctions list and the relevant national lists, a workload-profile review with the local engineering team, and a contractual schedule that sets the envelope for the hosted capacity. Documented, kept for the duration of the contract plus the statutory retention period, and handled by the same people who run the hall on shift.
Short-notice capacity requests go through the same process. We would rather say "not this quarter" than compress it — and we have found that serious customers prefer that answer to the alternative.
Where this goes next.
HPC and AI are the workloads that are hardest to fit into a generic hosting footprint and easiest to fit into a site that was engineered around them. As the next generation of Nordic sites comes online, this is the lane we are building for — the customers whose compute demand is growing faster than anyone in the sector can build power for, and who value a site that treats grid services and heat recovery as part of the product, not as marketing.