Energy efficiency

Reading order

Start with the metric name. PUE is facility energy divided by IT energy for a stated period and boundary. Other sentences — renewable share, fossil-free mix, matching products, design watts per rack — are not PUE. Write the metric, then the boundary: IT, hall, building or campus. Then write the period: a design value, a test day, or a measured year. Then label design versus measured. Only after those four cells exist may you look at a second hall.

Keep electricity-system claims in a side column. A national mix from Energimyndigheten, a bidding-area price in SE1 to SE4, and a Skatteverket energy-tax condition do not make two PUE figures comparable. The European Commission page on data-centre energy performance explains reporting work. It does not publish every Swedish hall’s PUE and it is not a census you can scrape into a ranking.

• Same metric.
• Same boundary (IT, hall, campus).
• Stated period.
• Design versus measured labelled.

How efficiency claims sit in the Swedish system

PUE moves when the numerator or the denominator moves. Cooling plant, transformers, lighting and other facility loads sit in the numerator. IT energy sits in the denominator. A hall that rejects heat to outdoor air and a hall that pumps heat toward a district network can both publish a low operator PUE and still not be comparable, because the boundary and the year may differ. Heat sold to a utility is a heat-reuse claim. It is not automatically a better PUE.

Swedish power institutions do not compute hall PUE. The network owner answers a connection request. Svenska kraftnät publishes an indicative transmission-capacity map that does not replace formal connection and does not show underlying-network capacity. Energimarknadsinspektionen’s bidding areas show transmission constraints in the wholesale price. A cheap SE1 hour in Luleå does not prove that Meta’s operator PUE can be copied onto another plot. A dear SE4 hour in Falkenberg does not prove that GleSYS’s operator PUE is worse.

National mix and hall efficiency remain separate. Energimyndigheten’s system pages describe generation and networks at country scale. An operator sentence about hydropower or 100 percent renewable energy is a supply or contract claim. It can be true in the same year as a mediocre measured PUE, and it can be silent in the same year as a strong design PUE. Skatteverket’s energy-tax rules follow statute, not a marketing efficiency badge. Boverket and Naturvårdsverket do not issue PUE certificates.

Reporting frameworks are not rankings. EU energy-performance work for data centres is about how operators report. It does not fill a missing Swedish cell. ISO wording on an operator page, as on EcoDataCenter’s Falun site, is the operator’s own citation of a method. DataCenterSweden does not reproduce copyrighted standard text and does not treat that citation as a third-party audit unless a dated audit document is cited. Design PUE and measured annual PUE stay different claims even when both say 1.2.

What named primary sources show

atNorth’s SWE01 page states a design PUE of 1.2 together with 12 MW site power capacity, 6,400 square metres, liquid cooling and heat recovery with Stockholm Exergi. The PUE cell is labelled design. Occupied IT load is not published. The figure is an operator figure for one Kista hall. It is not placed in a table against other operators on this page.

GleSYS’s current Falkenberg page states PUE 1.2, 5 MW total capacity, 100 percent renewable energy and district-heat recovery. The 1 October 2020 release stated 3 MW utility power and a heat-delivery start to Falkenberg Energi. The PUE on the later page is an operator figure. The renewable-energy sentence is a supply claim, not a measured PUE. The two megawatt rows stay dated and separate. Annual heat volumes after 2020 are not on those pages.

EcoDataCenter’s Falun site page states PUE 1.2 according to ISO standards and 80 MW available power. The 2025 annual report states 41 MW after commissioning C and D, and 65 MW once E and F are operational. Those megawatt sentences are not PUE, and they are not added together. Meta’s cited Luleå page reports PUE 1.09 with outdoor-air cooling and local hydropower. Current megawatts are not on that page. The 1.09 is an operator figure from that source’s method and year, not a national benchmark.

The European Commission energy-performance page is cited as reporting context. It does not supply a Swedish league table and it does not fill Meta’s missing current MW or GleSYS’s missing post-2020 heat volume. Microsoft’s 2021 region opening and later community PDF name Gävle, Sandviken and Staffanstorp without facility megawatts and without a PUE in the materials used here. Absence is recorded as absence.

Common misreads

The first misread is to sort operator PUE figures into a league table. 1.09 and 1.2 look comparable on a slide. They are not comparable when year, method, boundary and design-versus-measured labels differ. This page will not rank Meta against atNorth, GleSYS or EcoDataCenter. The second misread is to treat a renewable-contract or hydropower sentence as a PUE. Meta’s hydropower description and GleSYS’s 100 percent renewable sentence are supply claims.

The third misread is to use PUE as a planning score or a connection argument. Boverket does not grant bygglov on a PUE. A network owner does not reserve megawatts because a brochure says 1.2. The fourth is to treat EU reporting as a Swedish results table. The fifth is to let campus megawatts — Falun’s 41, 65 and 80, SWE01’s 12, GleSYS’s 3 and 5 — masquerade as efficiency metrics. They are capacity rows with their own gaps.

• Comparing two PUE numbers without period and boundary.
• Building a league table from operator pages.
• Using PUE as a planning-permit or grid-capacity score.
• Treating renewable contracts as measured hall efficiency.
• Using EU reporting pages as a Swedish census.

What to ask next

Ask for the metric definition, the boundary drawing, the period, and whether the number is design or measured. Ask who calculated it and whether a third-party assurance document exists. If the page says ISO, ask for the dated method statement rather than assuming an audit. If the page says renewable or hydropower, move that sentence to a supply cell and keep PUE empty until a PUE document appears.

Then ask the network owner how facility load was defined in the latest written position. Open the cooling page for the plant that sits in the numerator, the heat-reuse page for sold heat that is not PUE, and the capacity-definitions page before you compare two MW cells. If any of those answers is missing, do not compute a winner and do not invent a national average.

• Metric, boundary, period, design versus measured.
• Who calculated the figure, and is an assurance paper cited?
• Which supply claim is being confused with PUE?
• How does the facility-load definition meet the grid paper?
• Which cell is still empty — and therefore blocks a comparison?

What this does and does not prove

This page proves that efficiency claims in the reviewed Swedish set cannot be ranked from the cited operator pages. It proves that atNorth, GleSYS and EcoDataCenter each state 1.2 with different surrounding words, that Meta states 1.09, and that those figures remain operator figures from different years and methods. It proves that EU reporting context is not a Swedish results table.

It does not prove which Swedish hall is most efficient. It does not publish a national average PUE or a national energy-use total for data centres. It does not convert hydropower, 100 percent renewable wording, or heat recovery into a PUE. It does not use PUE to award spare megawatts in any bidding area or to settle a detaljplan. Operator figures from different years and methods stay unranked.

If a later primary source publishes a measured annual PUE with a drawn boundary, store it as a new dated row. Do not overwrite the older design figure. Until that row exists, the comparison stays blocked, the empty cell stays visible, and a slide that sorts 1.09 against 1.2 remains a misread rather than a league table of Swedish halls.