Nordic Region GPU Data Centers: Iceland, Norway, and Sweden Power Advantages

The Nordic region offers the lowest industrial electricity prices in Europe for GPU infrastructure, creating a compelling cost advantage for large-scale training workloads. Iceland leads with rates as low as EUR0.03-0.05 per kWh, powered exclusively by geothermal and hydroelectric sources. Norway follows at EUR0.04-0.07 per kWh with 98 percent hydroelectric generation. Sweden offers EUR0.06-0.10 per kWh, depending on grid zone, with a mix of hydro, nuclear, and wind. For comparison, a 1,000-GPU H100 cluster running at 70 percent utilization faces annual power costs of approximately EUR150,000-EUR250,000 in Iceland versus EUR1.6-1.9 million in Germany, representing a 6-10x cost differential.

This power cost advantage has triggered a wave of GPU infrastructure investment across the Nordic corridor. Total Nordic data center capacity reached 800 MW in 2026, with 400 MW additional under construction. The Nordics now host an estimated 20,000-30,000 H100-equivalent GPUs, concentrated in facilities designed specifically for AI training rather than general cloud compute. The emerging Nordic AI corridor stretches from Reykjavik through Oslo and Stockholm down to Copenhagen, connected by redundant fiber routes through the Baltic Sea and into continental Europe.

Iceland offers the world's lowest GPU operating costs when factoring in both power and cooling. The country's 750 MW of geothermal generation provides baseload renewable electricity at fixed rates of EUR0.035-0.05 per kWh under 10-15 year PPAs. The cooling advantage is equally significant: year-round ambient temperatures of 0-15 degrees Celsius enable free air cooling with PUE of 1.05-1.10, meaning that 90-95 percent of incoming power goes to GPU compute rather than cooling overhead. For a 1,000-GPU cluster, this translates to $1.2-1.5 million in annual savings versus a comparable Germany deployment.

The primary GPU facilities in Iceland are operated by Borealis Data Center (Reykjanes), Verne Global (Keflavik), and atNorth (Gardabaer). Borealis recently completed a 30 MW expansion dedicated to AI training, hosting 4,000 H100 GPUs with direct liquid cooling from geothermal hot water at 35 degrees Celsius. atNorth's Gardabaer campus runs 2,000 H200 GPUs for a European LLM training consortium, using Iceland's geothermal steam for their absorption chillers. Cross-border latency from Iceland to mainland Europe via the FARICE-1 and IRIS subsea cables is 15-25ms to the UK and 30-40ms to Frankfurt, making Iceland suitable for training but not for real-time inference serving to continental users.

Norway's GPU infrastructure leverages its hydroelectric dominance, with 98 percent of generation coming from hydropower stored in mountain reservoirs that provide natural power demand smoothing. Facilities in the Stavanger, Bergen, and Trondheim regions draw on regional hydro plants at EUR0.04-0.06 per kWh under PPAs. Bulk Data Centers' Kristiansand facility operates a 6,000-GPU cluster for European research institutions, and Green Mountain's DC1 Stavanger hosts 4,000 H100 GPUs for EU-based LLM training. The power is so cheap that Norway has become a destination for Bitcoin mining operations transitioning to AI compute.

The cooling advantage in Norway comes from cold fjord water and mountain tunnel locations. Green Mountain's Rjukan facility is built inside a mountain at 60 degrees north latitude, drawing 4-8 degrees Celsius tunnel ventilation air for year-round free cooling at PUE of 1.08-1.12. Lefdal Mine Datacenter, built in a former mine on the west coast, also offers PUE below 1.15 with fjord water cooling. These facilities are increasingly being marketed as GPU training centers for European clients who need GDPR-compliant data sovereignty combined with the lowest possible operating costs.

Sweden has the deepest GPU market in the Nordics by total capacity, anchored by its strategic position in SE3 and SE4 power zones. Stockholm's data center cluster around Kista (Equinix SK1/SK2, Bahnhof, Interxion) hosts approximately 8,000-12,000 GPUs across financial services, enterprise AI, and startup workloads. Swedish industrial power in the SE1 and SE2 zones (northern Sweden) is as low as EUR0.03-0.04 per kWh due to surplus hydro and wind generation, but transmission bottlenecks to the Stockholm region keep SE3 at EUR0.07-0.10 per kWh.

The most significant Nordic AI facility is the Lulea AI Factory in northern Sweden, a joint venture between H2 Green Steel (now H2GS AI), NVIDIA, and the Swedish government. The facility is designed for 200 MW total capacity with 100 MW operational as of mid-2026, hosting 16,000 H100 and 4,000 B200 GPUs. Lulea draws power directly from the Lule River hydroelectric system at EUR0.035 per kWh under a 25-year PPA, and the Arctic location with six months of sub-10 degrees Celsius ambient temperatures enables PUE of 1.08-1.10. The facility runs fiber direct to Stockholm with 6ms latency and to Helsinki at 8ms.

Procuring GPU capacity in the Nordics requires understanding the region's unique constraints: limited local inference demand means most Nordic GPU output is consumed remotely, so network connectivity to population centers is the most important metric after power cost. Facilities in Iceland and northern Norway offer the lowest power prices but add 20-40ms of latency to continental European end users. The optimal split is to run training workloads in Iceland or northern Sweden where power is cheapest, then serve inference from Frankfurt, London, or Amsterdam hubs.

Reserved pricing for Nordic H100 capacity ranges from EUR1.80-2.80 per GPU-hour, significantly below the EUR2.90-3.80 in FLAP-D markets. atNorth offers H100 at EUR1.90 per GPU-hour on 3-year reserved contracts. Borealis quotes EUR2.10 for H100 on 1-year terms. The caveat is that Nordic providers typically require larger minimum commitments (32+ GPUs) and longer contract durations (12+ months) compared to FLAP-D providers. Network costs can add EUR0.10-0.30 per GPU-hour for data egress to continental Europe, partially offsetting the power cost advantage. ClusterBid's Nordic-specific filters help buyers compare total cost of ownership including power, colocation, and network transport.