The figures are public, but rarely read side by side. As of June 30, 2026, Svenska kraftnät’s application queue for the Swedish transmission grid held a total of 25,609 megawatts of new electricity consumption awaiting a decision. Of that, 7,014 megawatts were data centers and 10,201 megawatts were industry — two customer categories already standing, quite literally, in the same line.
The queues, country by country
In Norway, Statnett and the regional grid companies have reserved capacity for over 8,000 megawatts of new consumption. According to Teknisk Ukeblad’s review of connection data, 56 data centers hold reservations equivalent to roughly 21 terawatt-hours per year, while some sixty more with needs around 36 terawatt-hours wait in the queue. NVE’s system report from May 2026 expects actual data center consumption to grow from around 3 terawatt-hours in 2025 to roughly 8 terawatt-hours by 2030 — and calls data centers the forecast’s big wild card.
Finland is the extreme case. Fingrid has received connection inquiries from data centers alone exceeding 50,000 megawatts — more than three times the country’s peak consumption of roughly 15,000 megawatts. Fingrid’s forecast points to electricity consumption rising from 86 terawatt-hours in 2025 to 103–123 terawatt-hours by 2030, with new data centers as the main driver.
And on the horizon stands Denmark as a cautionary tale: in March 2026, Energinet paused new grid connections after the queue reached 60 gigawatts — nearly nine times the country’s peak consumption — of which roughly 14 gigawatts were data centers.
First come, or most mature first
So how does one decide who gets their megawatts? The short answer: no one decides — yet.
Norway is the clearest case. Statnett states explicitly that it “operates industry-neutrally and does not prioritize between customers.” Queue position is determined by the date an application is judged mature under standardized maturity criteria — permits, plans, financing. It is a reasonable arrangement, but it carries a built-in tilt that has been raised in the Norwegian debate: a well-capitalized data center company with land options in hand can produce a mature application faster than an industrial project that needs years of environmental review. Neutral rules do not produce neutral outcomes.
Sweden likewise has no explicit prioritization, but Svenska kraftnät’s April 2026 report — its answer to the government’s mandate to review the connection process — introduces tools pointing in a new direction: a mandatory three-party dialogue within six months, “business matching” between production and consumption, and a designation system under which Svenska kraftnät can identify capacity zones “where it is suitable to connect a certain customer category.” The agency stresses that the mandate is not about steering where businesses locate. But a zone for a certain customer category is, however you turn it, a soft form of prioritization.
Finland has chosen the wallet. Fingrid is consulting on new technical connection requirements for data centers and electrolyzers (KJV2026), but the sharpest decision was taken in parliament: from July 1, 2026, data centers were moved from the reduced electricity tax class to the general one — an increase of 2.19 cents per kilowatt-hour. The Finance Committee’s reasoning is worth quoting: data center operations are by nature services, not industrial production.
When the AI factory and the steel mill want the same megawatt
It is in the north that the question becomes concrete. In Boden, Stegra is building hydrogen-based steel production with a planned 740 megawatts of electrolysis capacity, of which the first 200 megawatts are installed — while data centers totaling 7,014 megawatts stand in the same national connection queue as industry. In Norway, Anthropic recently signed for 133 megawatts of AI capacity in Tydal — in price area NO3, where Statnett already has an industrial connection queue — and in Narvik, Stargate Norway is expanding beyond an initial 230 megawatts.
The point is not that any of these projects is wrong. The point is that a megawatt reserved in a constrained grid area is a megawatt someone else will not get for a decade: Fingrid states that a single transmission line takes seven to eight years on average to realize, including permitting and environmental assessment. A data center decision is made in months; a steel mill matures over years. The queue moves at one pace, the customers at another. This — not blackouts — is AI’s real stress test of the Nordic grid: it tests the design of the queue.
Public benefit — wallet or politics
Which brings us to the open question. In Norway, Høyre, SV and Rødt proposed licensing requirements for data centers; digitalization minister Karianne Tung rejected the demand, citing among other things EEA law. In Sweden, the question of a national data center plan has been raised in the Riksdag, while the government’s line via Svenska kraftnät so far is transparency rather than steering. Finland is so far the only country to act through economic instruments — the tax bill. And Denmark shows where the road ends when the queue outgrows the system: Energinet has abolished the first-come, first-served principle for large consumers and now sorts projects into pools by maturity and grid benefit.
Four countries, four answers to the same question: should society rank the value of a megawatt user — and if so, who decides? None of today’s frameworks has an answer for the day an AI gigafactory and a steel mill each hold a mature application for the same megawatt. Until someone formulates one, the answer already written into the queue order applies.
A back-of-the-envelope exercise: megawatts measured in person-years
What follows is an illustrative calculation, not a forecast — every step rests on public benchmarks or explicitly flagged assumptions.
A fully built AI data center costs, by industry estimates, 30–40 million dollars per megawatt including chips — that is, 30–40 billion dollars per 1,000 megawatts. Assume further that the investment must pay for itself over five years, roughly the lifespan of the hardware, and that the intermediary layer, following the logic of Sequoia investor David Cahn’s analysis “AI’s $600B Question,” needs a 50 percent gross margin. That implies required annual end-customer revenues of roughly 12–16 billion dollars — per gigawatt.
Electricity itself — the physical bottleneck — is a small line item in this calculation: one gigawatt running continuously draws 8.76 terawatt-hours per year, which at Nordic industrial electricity prices of roughly 5–8 euro cents per kilowatt-hour including grid fees and taxes (Eurostat, second half of 2025) comes to roughly 0.4–0.8 billion dollars a year — on the order of three to five percent of the revenue requirement, comfortably absorbed by the margin assumption above. The bottom-line figures are thus unchanged: it is the hardware, not the power, that drives the arithmetic.
If those revenues are ultimately to be paid for through the streamlining of human work, they can be expressed in person-years. At a Nordic full cost per full-time employee of roughly 80,000–110,000 dollars (Sweden’s average monthly wage of SEK 42,900 in 2025 per SCB/the Swedish Mediation Office, plus employer contributions and overhead — an assumption), that corresponds to on the order of 100,000–200,000 person-years. Per gigawatt.
Sum up the current Nordic data center queues — Svenska kraftnät’s 7,014 megawatts, the Norwegian reservations and queue totaling roughly 57 terawatt-hours per year (converted to continuous load, about 6.5 gigawatts; 57 TWh ÷ 8,760 hours), Fingrid’s more than 50,000 megawatts of inquiries and Denmark’s roughly 14,000 megawatts — and you land at on the order of 80 gigawatts, with the important addition that the categories differ (inquiries, reservations, queue) and that double counting and attrition are likely: this is an upper bound of expressed interest, not construction plans. Apply the same chain anyway, and it corresponds to an annual revenue requirement of roughly 900–1,250 billion dollars — or on the order of 8–16 million person-years.
Three caveats carry the whole exercise. Nordic data centers serve global markets — the person-years are not Nordic jobs. AI revenues are not solely labor substitution — new capacity and new services count too. And the benchmarks have wide ranges. But the order of magnitude says something: the megawatts now queuing in Nordic grids carry return requirements that sooner or later must be extracted from the real economy.
Neutrality is also a choice.
Sources: Svenska kraftnät, pågående anslutningsärenden (as of June 30, 2026) · Svenska kraftnät, rapport Svk 2025/5008 (April 2026) · Svenska kraftnät, May 4, 2026 · Statnett, March 31, 2025 · NVE, rapport 2026/17 (May 2026) · Teknisk Ukeblad · Fingrid, March 2025 · Fingrid KJV2026-remiss (juni 2026) · Yle on the electricity tax decision (HE 156/2025) · Energinet, March 2, 2026 · The Riksdag, interpellation 2025/26:414 · Energinyheter on Stegra · David Cahn/Sequoia, ”AI’s $600B Question” · SCB/Swedish Mediation Office, average monthly wage 2025 · Eurostat, electricity prices for industrial consumers (nrg_pc_205), second half of 2025 · Capex benchmarks for the calculation: JLL Global Data Center Market Outlook 2026 och Epoch AI (May 2026), via industry compilations (Axis Intelligence, Archdesk). The Tydal, Narvik and licensing details follow AI-skiftet’s news feed (Bloomberg, Nscale, Digi.no).