Google’s Finland Plan Highlights the Search for Power- and Cooling-Advantaged AI Data Center Sites

Google’s planned Finnish expansion underscores the growing importance of energy procurement and thermal management in AI data center development. The company says it will invest at least €13 billion over the next two years in data centers and related infrastructure in Finland. Its 22-year agreement with Fortum provides for up to 50% nuclear power supply.

The Nordic market is also represented by OpenAI’s Stargate Norway project in Narvik. The project is planned to start with a 230 MW initial phase, followed by a further planned 290 MW expansion. Its initial target is deployment of 100,000 Nvidia GPUs. These figures describe planned phases and an IT deployment target, rather than operational capacity.

In Canada, Meta is building a 1 GW data center in Sturgeon County, Alberta. The facility is intended to use closed-loop liquid cooling and dry cooling without water use for cooling. The project illustrates the industry’s interest in designs that reduce dependence on water while supporting high-density compute loads.

Google has also disclosed Project Suncatcher, an experimental satellite project incorporating a TPU capable of AI inference. The concept extends the search for new infrastructure environments beyond terrestrial locations, but its cost has not been disclosed. Commercial viability would depend on future economics for launches, satellite replacement and data transmission, while heat rejection in vacuum, radiation, connectivity and orbital debris remain unresolved constraints.

Marine concepts remain earlier-stage. Hanwha Ocean has presented a 60 MW-class floating data center model. HD Hyundai Heavy Industries and Schneider Electric have signed an MOU to jointly develop a floating data center and 100 MW-class power equipment. Separately, a subsea data center in Ulsan is planned to move to sea trials in autumn 2027 after land-based testing. These developments represent models, development work, agreements and planned trials—not evidence of operating commercial capacity.

The broader implication is that grid capacity, access to lower-carbon generation and cooling design could become constraints on AI data center growth alongside server supply and land availability. The reviewed data indicate that more than 40% of US data center electricity supply came from natural gas in 2024, compared with about 24% from renewables, about 20% from nuclear power and about 15% from coal. Northern land-based projects with access to hydroelectric, nuclear or other stable generation may be nearer-term options; marine, subsea and space-based formats currently offer technological optionality but carry materially higher execution and economic uncertainty.