Google commits €13bn to Finnish AI data centers and buys up to half of Loviisa nuclear output

Northern locations make sense for cooling, but Texas and Arizona win on power access, permitting, available land, fiber connectivity, low population density, and friendlier regulation.

Politics matters too. Job-creating infrastructure gets placed in red states to appeal to Republican constituencies, and southern states make land sales and large contracts easier with less community backlash.

Union labor is another factor. Northeastern states can require it or stop projects, while Texas and Arizona are much more anti-union and pro-corporation.

Cooling savings may not outweigh higher northern power prices and other costs. Google says non-computing overhead is about 10% of power use, and power may be only one part of total data center cost.

Cold climates also need freeze protection and cannot always rely on evaporative cooling. Winter household demand can constrain power, while summer capacity is still needed.

Arizona has abundant solar, and Nevada has solar plus geothermal resources. Those are solid reasons to build data centers there.

Nuclear power is expensive and slow to build. Finland's Olkiluoto project began in 2005 for a 2009 completion, bankrupted Areva in 2016, and was not finished until 2023; US AP1000 projects also bankrupted Westinghouse.

Texas may also have better connectivity and latency to customers. That may matter less for AI workloads than it did for traditional internet services.

Finland's electricity emissions are low, about 71 gCO2e/kWh. France, Sweden, and Norway also look suitable for European data centers on that measure.

Sweden is already struggling to expand generation, and hydro has hard geographic limits. Norway has volatile prices, constrained transmission from its northern surplus, less reliable reservoir inflows, and political resistance to new hydro, wind, and nuclear.

France has announced €93bn in data center contracts. Countries that have substantially decarbonized electricity tend to depend on nuclear, hydro, or geothermal power.

Buying nuclear output does not mean homes receive dirtier power. Electricity is fungible, demand stays the same, and added demand should eventually bring more supply.

The Google deal may keep Loviisa running beyond 2030. Extending it to 2050 requires about €1bn in investment, so a long-term buyer helps justify keeping the plant open.

Finland is already among the greenest electricity markets, so the idea that households will simply switch to coal is misleading.

I wonder whether privacy-sensitive US data such as HIPAA, financial, or national-security data can be stored in Finland, perhaps encrypted. These facilities may instead mainly serve European customers, where local storage makes more sense.

I hope the data centers can send waste heat into district heating. Finland has good municipal infrastructure, though I do not know how its heating systems connect to these sites.

The grid is relatively uncongested for now. Google's contract is only for up to 50% of Loviisa's output, not necessarily half at all times.