May 12, 2026

Nscale just put real money behind the “AI in the Arctic” narrative: an extra €670 million to add 115MW in Narvik, Norway, on top of a Microsoft deal that brings more than 30,000 NVIDIA Rubin GPUs to the site with compute expected by 2027. That’s not a press-release campus — it’s a financing stack chasing a specific GPU deployment plan. Pair it with Amazon lining up 700MW of carbon-free power in Nevada and a nuclear-powered Texas AI campus trying to hit first onsite power by end-2026, and you get the day’s theme: capacity is now being bought in three currencies at once — capital, grid access, and firm power.

The Big Stories

Nscale raises €670M for Narvik AI data centre expansion is the cleanest signal today that lenders are getting comfortable underwriting very large, very GPU-specific builds in non-traditional hubs. The additional €670m commitment (from ABN AMRO, DNB, Eksfin, Nordea and SEB) funds a further 115MW expansion and sits alongside an earlier €1.1bn debt facility. The interesting bit is the sequencing: financing follows an explicit hyperscaler tie-up — Microsoft’s agreement to add 30,000+ Rubin GPUs — which effectively turns “future demand” into something closer to an anchored offtake story.

Amazon invests 700 MW carbon-free Nevada projects for data centers shows how the biggest buyers are trying to de-risk power before the next wave of data halls lands. Amazon’s agreements cover 100MW of geothermal (Zanskar) and 600MW of solar plus 600MW of battery storage (Primergy) via NV Energy (Berkshire Hathaway Energy). Amazon also flagged this as its first data center partly powered by geothermal — notable not because geothermal is new, but because hyperscalers are explicitly reaching for firm-ish clean supply as a complement to the solar-plus-storage default.

Canada and TELUS advance sovereign AI data centre project is the government-side mirror image of the hyperscaler power land-grab: states want sovereign compute, not just “more cloud.” Ottawa and TELUS are advancing work on a proposed large-scale sovereign AI data centre project in British Columbia under the “Enabling large-scale sovereign AI data centres” initiative. The fine print matters: this follows Budget 2025’s plan to identify a limited number of projects and engage proponents via non-binding MOUs, and no funding has yet been committed — so treat this as momentum and direction, not a bankable capex announcement.

Fermi developing AI campus powered by nuclear reactors is a reminder that “direct-to-data-center nuclear” is still part infrastructure, part governance drama. Caddis Capital (which owns ~9.3% of Fermi) publicly backed the board and management, supported removing the CEO and CFO, and opposed a hasty sale — all while Fermi pushes Project Matador at Texas Tech University, planning to use Westinghouse AP1000 reactors to directly power data centres. The stated target of first onsite power by end-2026 is aggressive; the bigger takeaway is that these projects will be judged as much on execution discipline and control as on the attractiveness of firm power.

INNIO files S-1 for proposed Nasdaq IPO listing adds a capital-markets angle to the “power for AI” rush. INNIO has filed for a Nasdaq Global Select listing (ticker “INIO”) with Goldman Sachs, J.P. Morgan, and Morgan Stanley as joint leads, though timing/size/price are TBD. In a market obsessed with the next data center landlord, suppliers that sit closer to generation and on-site energy resilience may find investors suddenly willing to listen — especially if grid timelines keep stretching.

Behind the Headlines

Community acceptance emerges as key U.S. data center constraint is the least glamorous story today — and possibly the most predictive. Invera Energy’s Evelyn Carpenter argues community acceptance has become the primary constraint, driving delays, cancellations, and moratoriums, with the same three objections showing up everywhere: electricity strain, water use, and fuzzy local economic benefit. The investor read-through is simple: entitlement and public process risk is moving from “annoying” to “value-defining,” and it will start separating developers that can package credible resource plans and community benefits from those that show up with megawatts and a construction schedule.

Wet Compression Upgrades Gas Turbines to Meet AI Data Center Power highlights why “fast power” solutions are having a moment. Ironclad Energy added Mee Industries wet compression to two GE Frame 6B turbines at Morris Cogen, gaining ~5–6MW per unit, improving heat rate (~400 Btu/kWh), with a third unit planned for 2026; the reported cost of about $150/kW and 24-week installation timeline are the point. When large-load interconnections can take years, incremental upgrades that squeeze meaningful MW out of existing assets become an unusually practical lever — even if they’re not the end-state decarbonization answer.

Fast-Path to Affordability: Energy-Only (ERIS) Resources in PJM is a good example of how market design and interconnection policy are becoming de facto data center policy. RMI-commissioned Aurora Energy Research estimates that adding 10GW of ERIS (5GW wind + 5GW solar by 2028) could save ~$10.9bn (2025$) for PJM ratepayers over ten years, and urges PJM to create a separate, accelerated ERIS study pathway with minimal network upgrade requirements. For data centers, the lesson is that “more power faster” isn’t only about building plants — it’s also about the rulebook that determines what can connect quickly and under what upgrade burden.