Norway Caps New Data Centre Power as AI Infrastructure Race Tests Grid Limits
A hard restriction on fresh grid allocations in Norway signals how power scarcity is reordering where AI infrastructure can be built.
Four of the largest technology companies are on track to collectively spend as much as $725 billion in 2026 on data centres, chips, and related power infrastructure, according to OilPrice.com on Saturday (2026-06-28), as McKinsey estimates $5.2 trillion will be deployed into AI infrastructure globally before the end of the decade. The scale of that commitment is now running into a physical limit that neither capital nor algorithms can dissolve: grid connection capacity.4
Norway has responded by capping new data centre operators without existing infrastructure at initial power allocations of just 5 megawatts, a ceiling that is barely sufficient to run a small facility, let alone the gigawatt-scale campuses that hyperscalers require. The restriction, reported by OilPrice.com on Friday (2026-06-26), formalises what is becoming true across the most attractive power markets: the right to connect at scale is no longer routine.3
Norway built its reputation as a data centre destination on cheap, reliable hydroelectric power and a cool climate that reduces cooling costs. Now that same reputation is drawing more capital than its grid can readily accommodate, and the regulator is rationing access rather than allowing unconstrained demand to destabilise supply. Existing operators with capacity already secured hold an advantage that is becoming measurable in contract terms.3
The 5-megawatt cap coincides with a deal that shows why incumbency is valuable. On May 5 (2026-05-05), Bitzero signed a binding letter of intent with Singapore-based OneQode for a 15-year, 110-megawatt lease at its Norway data centre site. The contract is expected to generate approximately $2.6 billion in total contracted revenue over its life, according to OilPrice.com.3
Among the largest spenders driving this demand, Amazon alone projects capital expenditure of $200 billion in 2026, with the majority directed toward data centres. Microsoft and Alphabet are each projected near $190 billion. Meta has outlined a $600 billion U.S. infrastructure plan running through 2028. The Economist noted in May 2026 that McKinsey's $5.2 trillion forecast draws comparisons to the 1990s telecoms investment cycle, though the physical power constraint distinguishes this build-out from that era's fibre overbuild: data centres burn watts, not spectrum.4,2
Grid access is what separates announcements from operations. More than 70% of interconnection requests in power markets are withdrawn before reaching operation, according to OilPrice.com, meaning the queue of announced projects substantially exceeds what transmission systems can absorb. Norway's cap is an explicit version of an implicit global constraint.4
Goldman Sachs estimates global data centre power consumption will surge by up to 165% by the end of the decade, according to OilPrice.com. A grid connection capable of delivering $100 million to $500 million worth of electricity annually to a new facility has become the scarcest input in the AI build-out.3
Analysis from Digiconomist's Alex de Vries-Gao, cited by The Guardian in May 2026, found that AI could account for nearly half of all data centre power consumption by the end of 2026. The IEA has separately projected significant additional demand from the sector. The Economist, writing in May 2026, raised the question of whether revenue projections can justify the investment scale in a market still defining its pricing dynamics.1,2
ICE Endex TTF front-month gas settled at €41.08 as of Saturday's close (2026-06-27). Norway's hydroelectric power costs remain well below European gas-fired generation at those hub prices, meaning cost is not the constraint — access is. The pressure on operators already holding Norwegian grid capacity is to lock in long-term leases before further regulatory tightening. The Bitzero deal, at $2.6 billion over 15 years for 110 megawatts, suggests that price floor is already visible.3,4