AI Data Centers Look to Norway and Finland as US Grid Falls Short
Power-first developers are securing European hydro and nuclear capacity as American interconnection delays strand billions in AI investment.
Bitzero, a Nasdaq-listed data center developer, has secured 110 megawatts of immediate hydroelectric capacity at its Namsskogan site in Norway at an electricity price of 3 to 4 cents per kilowatt-hour, roughly 70% below the US average of 12 cents, with a second site near Pori in Finland holding potential capacity of up to 1 gigawatt, the company said in late June (2026-06-30).5
The cost differential reflects the primary constraint in AI infrastructure buildout. The White House warned in July 2025 that without $1.4 trillion in new infrastructure investment, US electricity prices could surge as much as 58% by 2030, driven mainly by AI data center and cryptocurrency demand.5 Locking in 3-cent Norwegian hydro against a US average of 12 cents, on a trajectory that Washington itself regards as steeply rising, is not a marginal pricing decision.
The standard development playbook — secure land, file interconnection requests, wait out a queue that can run several years — has struggled to keep pace with the capital being deployed. Bitzero chief executive Mohammed Bakhashwain said his firm reverses the sequence: power access and grid positioning come first, infrastructure follows. The company has signed a deal worth up to $2.6 billion, with as much as 85% projected as net income, a margin that illustrates how scarce contractable low-cost power has become.4
Goldman Sachs estimates up to a third of incremental US data center capacity, roughly 25 gigawatts, will be built off-grid over the next five years as on-grid interconnection timelines become unworkable for operators trying to meet AI deployment schedules.2 xAI completed a $20 billion fundraising in early January 2026 and has been pursuing off-grid power solutions alongside that raise.2
The Finland site near Pori would add scale. At up to 1 gigawatt of potential, the facility is large enough to serve multiple tenants, backed by hydro and nuclear, with 10 megawatts of capacity available beyond the initial Namsskogan 110 megawatts.5
Demand projections give the shift urgency. Within two years of ChatGPT's launch in 2022, around 40% of US and UK households had used an AI chatbot, according to IEA data.1 Efficiency gains have provided little relief. The improvements DeepSeek introduced in its V3 model were more than offset by the additional compute time required by reasoning-oriented successors, The Economist reported in May (2026-05-17).3 More capable AI generates more AI use, which generates more load, not less.
European power markets have not spiked in response to the buildout. ICE Endex TTF front-month traded at €49.53 on Friday (2026-07-10) and ICE NBP at €44.66, stable enough to make long-term European offtake agreements structurally attractive. The URA uranium exchange-traded fund rose 3.14% to $42.35 on Friday (2026-07-10), reflecting market interest in nuclear as a credible continuous-baseload option for AI-scale loads in markets like Finland that have the installed capacity.
What remains open is not whether hyperscalers will find power outside the US grid. Norway and Finland can absorb significant new load. The exposure is in the US itself: whether the $1.4 trillion infrastructure investment gap will be closed, or whether the interconnection backlog will persist long enough to make the offshore shift structural rather than transitional.