Google puts 13 billion euros into Finland: you pay for it, here is how

The investment covers an expansion of digital infrastructure together with clean energy projects meant to secure stable supply over the long term. It is not only about buildings and servers, but about where their electricity will come from years ahead.

One Google data centre has operated in Finland since 2009, using cold seawater from the Gulf of Finland for cooling. The new investment continues the same logic at a larger scale.

The company signed a 22-year agreement supporting a nuclear power plant in Finland. The term matters more than the sum: contracts of that length are signed when a predictable price and volume have to be fixed for decades.

Google also plans to increase wind generation capacity and invest in a new battery energy storage system. Nuclear, wind and storage solve different problems: base load, cheap generation and smoothing the peaks.

In a data centre's cost structure, cooling sits next to powering the equipment itself. The colder the outside air and water, the more hours per year the system runs without active chillers, and the less it spends.

For AI workloads this is felt more sharply than for ordinary hosting: racks of accelerators release more heat per square metre, and the cost of removing that heat grows faster than the cost of the floor space.

The news looks distant from everyday work, but it explains what anyone costing AI inside their product runs into.

• The price of a model request is ultimately the price of electricity and cooling. That is why tariffs move and will keep moving.
• When planning an AI feature, cost it over a year rather than at today's price list: you are buying capacity, not software.
• Data centre location affects both latency and data requirements. Ask a provider about the region, not only the price.
• Caching answers and reducing the number of calls saves more than picking a cheaper model: you pay for computation, not for a line of text.

Thirteen billion euros in a cold country with a nuclear plant is a bet that demand for computation will grow for a long time. For a business the conclusion is more modest: the cost of AI is set by infrastructure rather than by a subscription, and should be planned accordingly.