Writing

Norway Data Center Power: 12 GW in the Queue, 8 TWh in Reality — Why the Queue Is an Options Book

10 min read

Lit data hall by a Norwegian fjord, fed by power pylons, with a queue of ghostly buildings fading into the mountains
One hall is lit; the rest of the queue is still an option on the grid. Norway’s 12 GW headline versus NVE’s 8 TWh forecast. Illustration: AI-generated.

Norway’s data centers used 3.3 TWh of electricity in 2025, and the energy regulator NVE projects about 8 TWh by 2030. The 12 GW “queue” cited in the Storting on 5 October 2026 is reserved plus queued grid capacity – an options book, not a demand forecast. Read it as demand and you get both the grid and the jobs debate wrong.

In August I argued in The Constraint Is Electrons Now that power, not chips, had become AI’s real ceiling, and that the hard question was allocation rather than generation. Norway is now having that allocation fight in public. Here is what the primary numbers say per October 2026, and what I would take from them if I were siting inference in the Nordics.

  • Actual use: 3.3 TWh in Norwegian data centers in 2025, up from under 1 TWh in 2020 (NVE, Elhub data).
  • Forecast: about 8 TWh in 2030 (NVE, May 2026) – roughly 913 MW of average load, by my arithmetic.
  • Reserved: 3,500 MW of grid capacity reserved for new data centers as of March 2026, out of 7,821 MW reserved for all new consumption in Q1 2026.
  • Reserved or queued: about 12 GW across roughly 120 projects, according to a Storting proposal by four Rødt (Red Party) MPs dated 5 October 2026.
  • North of Svartisen: Statnett has paused new reservations for loads above 5 MW since April 2026 – for all industry, not only data centers.
  • Tax: data centers pay the ordinary electricity tax of 7.13 øre per kWh; the finance minister advised against a higher rate in February 2026.

How much electricity do data centers use in Norway?

According to NVE’s method note on projecting data-center electricity use (11 August 2026), Norwegian data centers used 3.3 TWh in 2025, based on metering data from Elhub. In 2020 the figure was under 1 TWh.

The same note restates NVE’s projection from its May 2026 power-system report: about 8 TWh in 2030. Net growth is therefore about 4.7 TWh over five years. At NVE’s assumed 70 % utilisation, that is roughly 770 MW of new fully ramped load – my arithmetic, not NVE’s.

The Rødt proposal puts 2025 consumption at 3.14 TWh; I use NVE’s measured Elhub figure.

Where does the 12 GW number come from?

The 12 GW figure comes from Dokument 8:6 S (2026–2027), a representative proposal from Marie Sneve Martinussen, Bjørnar Moxnes, Geir Jørgensen and Terje Kollbotn, all Rødt, dated 5 October 2026. In my translation: “As of October 2026, Statnett has reserved or queued about 12 GW of grid capacity for data-center projects, spread over around 120 projects.” At NVE’s 70 % utilisation, the proposers write, that equals about 70 TWh a year – “more than half of Norway’s total power consumption today.”

The arithmetic holds: 12 GW × 8,760 hours × 70 % is 73.6 TWh. But two things matter. The number is the proposers’ compilation, not a figure Statnett has published in that form. And it merges two very different categories: capacity that is reserved and capacity that is merely queued.

What the queue looked like in 2025

The best official breakdown I have found is the energy minister’s answer to a written question from the Storting (8 May 2025). Statnett then had 28,188 MW of active consumption cases, of which 9,829 MW were data centers. Of those, 2,362 MW were reserved, 4,210 MW were in the capacity queue, and 3,257 MW were “other cases” – not yet assessed, or not assessed as mature.

So a third of the data-center volume in 2025 had not even cleared Statnett’s maturity screen. Whether the October 2026 figure includes such “other cases” is not stated in the proposal.

Why does NVE forecast only 8 TWh?

NVE forecasts only 8 TWh because it discounts every project by how far it has actually come. The method note is explicit: the lowest level at which a plan enters the projection at all is a grid reservation. Queued projects without a reservation count for zero.

  • Reserved capacity: 25 % probability of being built.
  • Final investment decision taken: 50 %.
  • Construction started: 75 %.

NVE also assumes an average of five years from reservation to the first construction phase, then a ramp to full operation of three years for AI data centers and five for cloud data centers. If all 3,500 MW reserved as of March 2026 were built, NVE calculates 21.5 TWh a year. Its 8 TWh projection for 2030 – which includes today’s 3.3 TWh – is about 11 % of the 73.6 TWh the full 12 GW would imply.

NVE has reasons for caution: 2019 also brought “high optimism” and many large data-center plans, and many were shelved. But it also writes, fairly, that “few data-center projects have been cancelled so far, but this may change.” And it singles out Stargate Narvik as a project that could use more than 1.5 TWh a year on its own.

Why the queue behaves like an options book

A grid reservation is an allocation of capacity in today’s or planned grid, tied to a connection date. A place in the capacity queue is a mature request for which no capacity is available yet. Neither is a commitment to consume.

The industry itself explains why developers file early. In its consultation response to Statnett (29 April 2026), Norsk Datasenterindustri wrote that a reservation of the requested capacity, together with a connection date, is an “essential element in an investment decision.” The reservation comes before the investment decision.

That is the definition of an option: pay a small premium now for the right, not the obligation, to build later. A developer who does not yet know which site will win financing, a tenant or a permit files at several. The sum of those filings measures how many doors are open, not how many buildings will stand.

The counter-evidence is real. Reserved data-center capacity rose from 2,362 MW (May 2025) to 3,500 MW (March 2026), about 48 % in ten months (two sources, possibly scoped differently). Data centers held about 45 % of Statnett’s 7,821 MW of reserved new consumption in Q1 2026. The reserved core is not fake. My claim is narrower: most of the 12 GW behaves like options, and the reserved core is a mix of real projects and hedges.

Why did Statnett pause reservations north of Svartisen?

Statnett paused new reservations north of Svartisen because the regional grid cannot carry the capacity already reserved plus growth in ordinary consumption. Statnett announced the hold in April 2026, citing an unacceptable risk of power shortage and voltage collapse in the event of faults. It applies to all consumers above 5 MW in the region – data centers, hydrogen, everything.

In its letter to the regulator RME (15 June 2026), Statnett confirms that existing reservations of about 1,150 MW are kept and new customers are placed in a queue. It adds that some customers take a long time to use reserved capacity, that “some fall away,” and that this applies especially to hydrogen and ammonia projects. Then the line for every Nordic infrastructure slide: “Kø vil være den nye normalen” – queues will be the new normal.

Tighter maturity rules for loads of 100 MW and above

Nationally, Statnett proposed stricter maturity requirements on 24 April 2026 for consumption connections of 100 MW or more, planned to take effect on 30 June 2026. Land must be zoned for the purpose. The maximum time to commissioning is cut from ten to six years. Financing must be documented with actual access to capital, “not just intentions.” Projects without sufficient progress lose their reservation.

This is option pricing done right: it raises the premium of holding a reservation without banning the instrument – and Norsk Datasenterindustri itself supports freeing reserved capacity from projects without real progress.

Is Norway restricting new data centers?

No – not as of October 2026. The Rødt proposal asks for a halt in allocating power and grid capacity to new data-center projects until a national framework is in place, a concession system for large new consumers, and that every new project of 25 MW or more be put before the Storting, modelled on the rules for petroleum investments. It is a proposal, submitted 5 October 2026; it has not been considered.

On tax, the direction runs the other way. In a letter to the finance committee (19 February 2026), Finance Minister Jens Stoltenberg noted that data centers already pay the ordinary electricity tax of 7.13 øre per kWh, after the reduced rate was abolished in 2023. He concluded that the industry should not pay a higher rate, citing no demonstrated market failure and possible illegal state aid under the EEA Agreement. The finance committee’s majority recommended against a special levy in March 2026.

What I would take from this for builders siting inference in the Nordics

I have argued before that falling token prices drive inference volume, and volume drives power. Norway is where that meets a queue. My practical reading:

  • The reservation is the asset. NVE notes that new sites in the cool Nordic climate can run a PUE close to 1; cooling is not the scarce input, a reservation with a date is. Partnering with a project that already holds one beats filing a new request.
  • Budget five years. NVE’s average from reservation to first construction phase is five years.
  • Treat the north as closed for new large loads until Statnett’s annual reassessment says otherwise, and read the Helgeland numbers carefully: about 1,000 MW already reserved and a large queue as of May 2026.
  • Price political tail risk. A 25 MW Storting review would make sub-25 MW modular inference sites the path of least resistance.
  • Expect rationing at every layer. Grid capacity is rationed the way high-bandwidth memory is; plan for both.

For policy, the lesson is symmetrical. A moratorium punishes the options and the real projects together, and it reads 12 GW of hedges as a threat to the record 2025 power surplus of over 22 TWh that the proposal itself cites. The proposal itself, citing Statistics Norway, puts data-center employment at 677 people in 2024 – a weak jobs case for traditional sites, but no basis for pricing the next generation. Maturity rules, deposits and connection charges price options. Bans destroy the information they carry.

Predictions

  • The Rødt proposal for a halt in grid allocation to new data centers will not win a Storting majority in the 2026–2027 session.
  • Elhub-measured data-center consumption for 2026 will come in below 5 TWh.
  • By the end of 2027, at least one data-center reservation of 100 MW or more in Norway will be withdrawn or returned under the tighter maturity rules.
  • NVE’s next projection for 2030 will stay below 12 TWh, even if the reserved-or-queued headline stays above 10 GW.

Frequently asked questions

How much electricity do data centers use in Norway, and how much is forecast for 2030?

Norwegian data centers used 3.3 TWh in 2025, according to NVE’s Elhub data. NVE projects about 8 TWh in 2030.

How much grid capacity have data centers reserved or queued in Norway as of October 2026?

About 12 GW across roughly 120 projects is reserved or queued, according to a Rødt proposal of 5 October 2026. Of that, 3,500 MW was actually reserved as of March 2026, according to NVE citing Statnett’s connection register.

Why did Statnett pause new grid reservations north of Svartisen?

Reserved capacity plus growth in ordinary consumption would create an unacceptable risk of power shortage and voltage collapse. The hold covers all new loads above 5 MW; existing reservations are kept.

Is Norway restricting new data centers?

Not as of October 2026. A proposed halt has not been considered; what is actually tightening is Statnett’s maturity rules and regional holds.

Do data centers pay a special electricity tax in Norway?

No. They pay the ordinary rate of 7.13 øre per kWh; the reduced rate was abolished in 2023.

Portrait of Håkon Berntsen

Written by

Håkon Berntsen

CTO and co-founder of Skjld Labs AS. Builds production AI agents, Microsoft 365 systems and the courses around them, from Sandefjord.

About LinkedIn GitHub