February 28, 2026

Equinix is buying its way deeper into the Nordic AI buildout. In a $4.0bn enterprise-value deal, CPP Investments and Equinix to acquire atNorth for $4B gives Equinix a ~40% stake in a platform that already controls 1GW of secured power across eight sites in Denmark, Finland, Iceland, Norway, and Sweden. The subtext: in power-constrained markets, “secured power” is increasingly the only currency that matters.

The Big Stories

The atNorth acquisition is a clear signal that the next phase of data centre consolidation will be led by buyers who want ready-made power positions, not just real estate. CPP will own ~60% and Equinix ~40%, and the buyers have provisionally lined up a $4.2bn financing package underwritten by European and Canadian lenders. atNorth’s footprint is explicitly pitched at AI/HPC facilities, and the 1GW of secured power is the headline number investors will care about—because it’s hard to replicate quickly in the Nordics without running into permitting, grid, and community friction.

Korea just put a big, concrete marker down on “AI campus as national industrial project.” Hyundai’s MoU-backed plan for Saemangeum includes a $4bn AI data center designed for up to 50,000 GPUs, with construction targeted across 2027–2029, alongside a 200MW-class PEM electrolyzer and gigawatt-scale solar ambitions by 2035 (Hyundai to build integrated AI, hydrogen and solar innovation hub). It’s a reminder that in some countries the AI data centre story is being fused directly into energy policy and industrial strategy—data halls, hydrogen, and renewables packaged as one employment-and-export narrative.

Google and Xcel’s Minnesota long-duration storage deal is the kind of power story data centre operators have been waiting for utilities to make real. Under the agreement, they’ll deploy a 300MW / 30GWh Form Energy iron-air battery in Pine Island, paired with 1,400MW of wind and 200MW of solar, plus a $50m investment into Xcel’s Capacity*Connect (Google and Xcel to deploy 300MW/30GWh iron-air system). The scale matters: 30GWh is multi-day territory, not “shave the peak” territory, and it points toward grid solutions that can support 24/7 clean power narratives without hand-waving.

In Sweden, the intersection of batteries, cheap power, and data centres just got more explicit. Lyten completed its acquisition of Northvolt’s Swedish operations and restarted manufacturing at the Northvolt Ett plant in Skellefteå, picking up assets valued around $5bn including 16GWh of capacity; it also floated a “Lyten Industrial Hub” concept powered by nearby hydro with co-located AI data centres, where EdgeConneX intends to take a site with potential for a 1GW campus (Lyten completes Northvolt Sweden acquisition, restarts Skellefteå plant). This is the industrial symbiosis pitch in its purest form: energy-heavy manufacturing and energy-heavy compute anchored to the same power story.

On the connective tissue side, FiberLight is placing a sizeable bet on West Texas as an AI corridor, not just a place with cheap land. The company committed $350m more—taking its total regional commitment to $500m—to add ~1,400 route miles, 1.2m new fiber miles, and a third route into Abilene to support AI and data center deployment (FiberLight commits $350M to West Texas AI and data-center infrastructure). If you’re tracking where “next markets” become real markets, the presence (and redundancy) of long-haul and regional fiber is usually the tell.

Behind the Headlines

Arizona is a case study in how quickly the “AI + fabs” combination turns grid planning into a physical bottleneck. A warning from Obodo Energy Partners’ John Mitman is blunt: the state’s growth in semiconductor fabs and data centers could require roughly two- to threefold generation increases over the next decade, with interconnection, substations, and transmission already strained (Arizona power constraints strain semiconductor fabs and data centers). The timelines are what should make operators nervous—pipeline expansions targeting late 2029, equipment lead times stretching into the early 2030s—because they collide directly with the industry’s typical build cadence.

The UK is showing what happens when data centre growth becomes a mainstream political issue, not a planning footnote. Community groups and Global Action Plan are protesting hyperscale AI data centres, while the Environmental Audit Committee has opened an inquiry into energy and water impacts; the cited context is stark: capacity growth could drive 8–10% of electricity-demand increase by 2035, and a single 100MW facility can use as much water as 80,000 people (Campaigners raise environmental concerns over UK data centres). This isn’t just reputational noise—formal scrutiny changes timelines, mitigation costs, and the “social licence” calculus for where hyperscale can expand.

India’s nuclear policy shift is worth watching through a data-centre lens, even if the story isn’t explicitly about data halls yet. The SHANTI Bill (assented in December 2025) opens India’s civil nuclear sector to private firms and targets 100GW by 2047—an ambition that implies roughly $214bn of investment—while the article argues India must also open nuclear R&D to build indigenous SMRs and avoid strategic dependency (India’s SHANTI Bill opens nuclear sector to private firms). If India is simultaneously projecting major data centre load growth and looking for firm, scalable clean generation, nuclear policy becoming “investable” is a structural development, not a niche legislative tweak.