Nscale, one of the largest independent AI infrastructure operators in Europe, is deploying facilities in Iceland, Norway, and the UK. Its Icelandic operations at Blonduos run on 100% renewable energy from geothermal and hydropower sources. Verne, the Nordic data center operator, signed a 15 megawatt (MW) agreement in November 2025 to host approximately 4,600 Nvidia Blackwell Ultra graphics processing units (GPUs) for Nscale at its Icelandic campus, one of the largest liquid-cooled GPU installations in Europe. Iceland's data center market, valued at $425 million in 2024, is projected to reach $812 million by 2030 at an 11.4% compound annual growth rate, driven by the combination of 100% renewable grid power, natural cooling from ambient temperatures that rarely exceed 59 degrees Fahrenheit (15 degrees Celsius), and submarine cable connectivity to European and North American networks.
Microsoft committed more than $1.5 billion to Nordic data center infrastructure in 2024 and 2025 combined, with significant capacity expansions in Finland and Sweden. AtNorth, the Nordic data center operator backed by Apax Partners, surpassed 200 MW of operational capacity across Iceland, Sweden, and Denmark in 2025 and announced a further 300 MW expansion pipeline. GlobalData senior analyst Beatriz Valle noted in early 2026 that hyperscalers will focus on sustainability as a core operational requirement in 2026 as AI workloads continue to push power density higher and as cooling costs become a more significant variable in total cost of ownership calculations. These are production capital decisions by organizations with global infrastructure portfolios and sophisticated cost modeling, not pilot investments.
Why Cooling Geography Is Becoming a Capital Allocation Variable
Sites selected for power access and network connectivity in the 2010s were engineered around cooling as an afterthought. That assumption broke down around 2023 when GPU-dense AI workloads entered production deployment at scale. A rack of Nvidia H100 servers draws 10 to 14 kilowatts (kW) per server, meaning a standard 42-unit rack can reach 60 to 80 kW of heat load. Air cooling infrastructure designed for conventional server densities cannot manage that thermal profile in warm ambient environments without significant additional energy consumption.
The power usage effectiveness (PUE) data illustrates the gap directly. Nordic data centers routinely achieve PUE of 1.1 to 1.15, meaning that for every watt consumed by compute, only 0.1 to 0.15 additional watts are consumed by cooling and facility overhead. U.S. and Southern European data centers in heat-stressed markets typically operate at PUE of 1.4 to 1.6 during summer months. For a 10 MW compute deployment, that PUE difference translates to 2.5 to 4.5 MW of additional annual power consumption that serves no computational purpose. At scale, the cooling inefficiency of warm-climate locations is a material operating cost line, not a rounding error. Most new hyperscale builds specified in 2025 and 2026 require direct liquid cooling (DLC) infrastructure as a base design requirement, further disadvantaging existing facilities in warm climates that were built before DLC became standard.
What Enterprise Technology Leaders With Legacy Infrastructure Should Take From the Migration
Most large enterprises do not operate at hyperscale and cannot relocate infrastructure to Iceland on a capital planning cycle. That is not the point. The point is that the same physics driving hyperscaler geography decisions apply to on-premise and co-location infrastructure in heat-vulnerable markets, just at a smaller scale and with a longer degradation curve before the consequences become visible.
On-premise data centers in facilities located in markets that regularly experience summer ambient temperatures above 95 degrees Fahrenheit (35 degrees Celsius) are carrying a cooling cost and reliability exposure that is likely understated in current technology budgets. Co-location contracts in warm-climate markets that do not include service level commitments for cooling capacity during heat events are holding an availability assumption that the ambient environment may not support every summer going forward. The hyperscaler migration is communicating something specific about the direction of costs and risks in heat-vulnerable infrastructure environments. The decision for enterprise leaders is whether to translate that signal into an asset review before the next summer season puts pressure on cooling systems that were not designed for the temperatures they are now running in.
