Technology explained

Choose the technical branch

Ask four questions before comparing technologies: who owns the IT load, who operates the building, whether retail customers share the hall, and whether high-density compute is a current requirement or a future design target. Then ask which evidence is available: a design description, a contract, a measured period or only a marketing claim.

Follow the branch that can change the site decision. Cooling can change water, heat rejection and neighbours. Connectivity can change carrier entry and resilience. AI/HPC can change density and power assumptions. A PUE without a date and boundary cannot choose a Swedish site.

• Hall type and customer model.
• Workload density and power evidence.
• Cooling, water and heat-rejection path.
• Connectivity, security and physical resilience.
• Energy and heat-reuse metrics with date and boundary.

What the Swedish file actually changes

A technical choice lands in several Swedish files at once. The network owner cares about requested import capacity, voltage and the point of connection, not about whether the hall is marketed as colo or AI. Svenska kraftnät’s indicative transmission map does not replace that conversation and does not show underlying-network capacity. Energimarknadsinspektionen’s four bidding areas — SE1 Luleå, SE2 Sundsvall, SE3 Stockholm, SE4 Malmö — show transmission constraints in the wholesale price. They do not allocate spare megawatts to a cadastral unit.

Boverket’s planning guidance and the municipal detaljplan speak to land use and building rights. They do not certify a cooling plant or a PUE. Naturvårdsverket’s environmental guidance is the starting point when water, noise or other Environmental Code questions open. Skatteverket’s energy-tax pages are a tax file, not an efficiency league table. A national generation mix published for Sweden is context. It is not a local connection offer in Falun, Falkenberg, Kista or Luleå.

The European Commission page on data-centre energy performance is a reporting context. It does not publish every Swedish hall’s numbers, and it is not a Swedish capacity census. Use it to understand that reporting frameworks exist. Do not use it to fill a missing operator megawatt, a missing annual heat volume, or a missing PUE period. A reporting rule is not a measured hall result.

Named examples already in the register

Meta describes the Luleå campus as operational from 2013, with outdoor-air cooling and local hydropower. Current megawatts are not on the cited Meta page. GleSYS published a heat-delivery start to Falkenberg Energi on 1 October 2020 and a 3 MW utility-power figure in that release; a later operator page states 5 MW. Those are two dated rows, not a silent replacement.

atNorth’s SWE01 page describes direct liquid cooling, a design PUE of 1.2, 12 MW site power and heat recovery with Stockholm Exergi. Conapto’s south-campus materials mention heat recovery toward Stockholm Exergi. EcoDataCenter markets Falun for HPC and AI, states a PUE of 1.2 according to ISO standards, and reports campus megawatts in separate annual-report and marketing sentences that must not be added together. evroc’s 3 February 2025 notice designed for up to 16,000 GPUs and did not state megawatts.

What this hub does not prove

This hub does not rank Swedish cooling plants, does not publish a PUE league table, and does not invent a national heat-market total. Operator efficiency figures come from different years and methods. A hydropower sentence in Luleå does not answer a connection request in Staffanstorp. A GPU design envelope is not a commissioned IT load. Read the leaf pages for the decision questions, then keep the gaps visible.