Norway exports energy. Virginia runs the models. The mismatch is the point.

An honest sentence about Norway and AI in 2026 reads as follows. Norway exports clean, abundant, low-carbon electricity to the grids that power the data centres that train and serve frontier AI models. Virginia runs the models. Norway exports the energy. The mismatch is the point — and as of mid-2026 it is the start of an industrial policy, not the end of an essay.

Most public-conversation framings of "AI and energy" treat energy as an environmental cost to be mitigated. That framing is not wrong, but it is incomplete and increasingly misleading for small open economies with abundant clean electricity. The right framing is industrial. AI training and inference are now, like aluminium smelting was a century ago, energy-intensive industrial processes whose location is partly determined by where electricity is cheap, clean, and abundant. Norway is one of a small number of places in the world where electricity meets all three conditions simultaneously and reliably.

That should change the conversation.

Why the cleanest electricity is now an industrial-policy asset

Frontier-model training is increasingly bottlenecked by power. Inference workloads, which now exceed training in absolute energy demand, are bottlenecked by power even more clearly. Hyperscale operators are willing to pay materially above-market industrial rates for clean, predictable, dispatchable electricity, with multi-decade contracts, in jurisdictions with stable governance and well-developed grid regulation.

Norway has the electricity. It has the governance. Through KI-fabrikken at Sigma2, opened in November 2025 around the Olivia supercomputer and connected to the European LUMI AI Factory network, it has an emerging credibility as an AI infrastructure operator ^[1]. What it does not yet have is an industrial policy that uses these three together as a deliberate proposition to allied AI infrastructure operators.

A workable Norwegian AI industrial policy in 2026 would have three planks.

Plank one: clean-energy AI zones with binding public-interest conditions

Designate two or three regions of Norway as priority zones for AI infrastructure investment. In each zone, offer multi-decade clean-electricity contracts at predictable rates, with two binding conditions in return.

The first condition is workload public-interest access: a defined slice of compute capacity, at cost or below, reserved for Norwegian public-sector users, Norwegian researchers, and Norwegian SMEs. This is not subsidy of the operator; it is a public-interest carve-out in exchange for the energy concession.

The second condition is allied-assurance transparency: where the operator has signed allied-assurance instruments with non-EU governments — the Microsoft–G42 Intergovernmental Assurance Agreement, signed in April 2024, is the template most likely to apply — the existence and scope of those instruments are disclosed to the relevant Norwegian authorities and, in summary form, to the Norwegian public ^[2].

Plank two: Article 4 literacy as a workforce condition

Tie eligibility for clean-energy AI zone benefits to operator commitments on Article 4-aligned AI literacy, both for the operator's own Norwegian workforce and for the public-sector users of the carved-out compute ^[3].

This is the place where Norwegian regulatory craft does the most work. Article 4 requires deployers to ensure literacy calibrated to context. A municipal hospital using zone-carved-out compute to run an AI triage service has literacy obligations that the operator can either help carry, ignore, or actively undermine. Tying zone eligibility to demonstrable literacy partnership keeps the operator on the right side of that line.

Plank three: Nordic-scale rather than Norway-only

Norway is too small to do this alone in a way that competes with Texas, Virginia, the Gulf, or southern France. Nordic-scale is more credible. Coordinated zoning between Norway, Sweden, and Finland — three countries with different but complementary clean-electricity profiles, similar regulatory cultures, and common membership in the EU AI Act regime via the EEA — would be a serious global proposition.

This is not a treaty. It is a coordinated industrial-policy stance, expressed in three national zoning regimes that read the same way to a hyperscale operator's site-selection team. The political coordination cost is modest compared to the upside.

The honest counter-argument

The honest counter-argument runs as follows. Norway should not host AI infrastructure at all because the marginal use of clean electricity for AI workloads displaces other uses — electrification of transport, hydrogen, industry, residential — and the resulting public-interest claim is vague.

This counter-argument is serious and deserves to be answered, not waved away.

The answer has two parts. First, the marginal-displacement argument is strongest where the public-interest claim is vague. With a binding public-interest carve-out and a literacy condition, the public-interest claim becomes specific and verifiable. Second, the alternative is not "Norway exports less electricity". Norway's electricity exports are not optional in any near-term policy regime; they are determined by transmission capacity and demand abroad. The choice is between exporting electrons that feed a hyperscaler in Texas and hosting the equivalent workloads inside Norway with Norwegian-shaped public-interest conditions attached.

The second answer is the better answer. It is also the harder argument to make in public, which is exactly why Nordic AI-policy discourse needs to make it.

What Norwegian institutions should do now

Three things, in order.

One. Name the proposition. Energy is industrial-policy leverage, not an environmental footnote. Joint Ministry of Energy and Ministry of Digitalisation ownership of this proposition is the precondition for everything else.

Two. Draft the zone framework. Two or three pilot zones, the public-interest carve-out, the literacy condition, the allied-assurance transparency clause. Publish the draft before negotiations begin.

Three. Coordinate Nordic-wide. Sweden and Finland will not move first; Norway can lead and harmonise after. The window in which this is possible — before a US administration, an EU presidency, or a hyperscaler announcement closes the optionality — is shorter than it looks.

Norway exports energy. Virginia runs the models. The mismatch can be closed in either direction. Which direction it closes is a policy decision Oslo can still make.