The Loviisa nuclear power plant, on Finland’s southern coast, was supposed to shut down in 2030. Its two reactors have been running since the late 1970s, and the arithmetic on keeping them going past their extended licence was the kind that usually ends with a decommissioning plan.
It will now run until 2050 instead, and the reason is a search engine.
On September 9, Google announced it would invest at least €13 billion, roughly $15.1 billion, in Finnish AI infrastructure over 2027 and 2028, the largest European investment in the company’s history. Bundled into that announcement was a 22-year power purchase agreement with the Finnish utility Fortum covering up to half of Loviisa’s output through 2049. The revenue certainty from that contract is what makes Fortum’s roughly €1 billion life extension programme work.
It is the first agreement of its kind in Europe and Google’s first nuclear deal outside the United States. It is also, depending on which Finnish politician you ask, either the best industrial news the country has had in a decade or a large unpermitted claim on the national grid.
Quick facts
- Google will invest at least €13 billion (about $15.1 billion) in Finland across 2027 and 2028
- Three new data centers in Kajaani, Muhos and Vaala, plus expansion of the existing Hamina site
- A 22-year power purchase agreement with Fortum for the Loviisa nuclear plant
- Deliveries start in 2028 at lower volume, reaching up to 50% of the plant’s capacity from 2030 to 2049
- Fortum’s roughly €1 billion investment extends Loviisa’s life from 2030 to 2050
- Construction is projected to add €3.6 billion a year to Finnish GDP and support more than 37,000 jobs
- Around 7,000 permanent jobs are projected, at wages averaging 24% above the Finnish median
- Google also signed onshore wind agreements bringing its contracted new Finnish wind capacity to 629 MW, plus a 94 MW battery system near Kajaani
What Google actually bought
It helps to separate the pieces, because the announcement folded several distinct commitments into one number.
| Component | Detail | Timing |
|---|---|---|
| New data centers | Kajaani, Muhos and Vaala in northern Finland | Built across 2027 and 2028 |
| Existing site | Expansion of the Hamina campus in the southeast | Same window |
| Nuclear PPA | Up to 50% of Loviisa’s output, via Fortum | 2028 start, full volume 2030 to 2049 |
| Onshore wind | New agreements with Valorem and Suomen Hyötytuuli | 629 MW contracted new-to-grid total |
| Storage | 94 MW battery system near the Kajaani site | Expected online late 2027 |
The geography is not incidental. Northern Finland is cold, politically stable, and sits on a grid with substantial existing hydro and wind. Cooling a data center is dramatically cheaper when the outside air does most of the work for much of the year, which is why the Nordics have been absorbing hyperscale capacity faster than almost anywhere else in Europe.
Why a power contract saved a reactor
This is the part of the story with implications well beyond Finland.
Extending the life of an ageing nuclear plant is expensive and front-loaded. You spend a billion euros on refurbishment now against electricity revenue spread over two decades, in a market where wholesale prices are volatile and policy can change twice in that span. Plenty of European reactors have closed not because they were unsafe or worn out, but because nobody would underwrite that arithmetic.
A 22-year contract with a creditworthy buyer for half the output removes most of that risk at a stroke. Fortum gets a revenue floor it can finance against. Google gets two decades of carbon-free baseload at a predictable price, which for an AI operator is worth more than the electricity itself. Power is now the binding constraint on AI expansion, and a contract that locks in supply through 2049 is a competitive asset, not just an expense line.
The pattern is becoming familiar. Google has been buying its way around grid constraints for a while, including a $1 billion commitment to Form Energy’s 100-hour battery technology. What is new here is a hyperscaler’s balance sheet directly determining whether a national grid asset continues to exist.
Northern Finland’s climate makes cooling cheap, which is why Nordic grids keep attracting hyperscale capacity. Photo via Pexels.
The argument happening in Helsinki
Prime Minister Petteri Orpo welcomed the investment, saying it will boost the economy and that electricity prices will stay under control. The economic case is not hard to make: €3.6 billion a year added to GDP during construction, more than 37,000 jobs supported at peak with roughly 16,000 in construction, and around 7,000 permanent roles afterwards at wages averaging 24% above the Finnish median. For northern municipalities that have spent decades watching young people leave, those are serious numbers.
The opposition’s objection is narrower and, on its own terms, fairly specific. Finland has no national permitting system for data centers. That means a single company can commit to a load of this magnitude without any centralized process weighing it against household and industrial demand. Opposition figures have warned this gap could strain supply and push consumer electricity prices up.
The framing that gives the objection weight
A Social Democratic lawmaker put the concern in terms that go beyond price, arguing that “the availability of electricity is also examined as a broader internal security issue, so that people have enough energy and that it is priced at a level they can afford to use.” Once electricity is treated as a security question rather than a market one, a private contract for half a reactor’s output looks less like a commercial deal and more like an allocation decision made without a public process.
Both sides have a point, and they are not really arguing about the same thing. The government is answering a question about supply, and on supply the deal genuinely adds rather than subtracts: it keeps a reactor alive that was otherwise closing, adds contracted wind capacity and a battery system. The opposition is asking a question about process, and on process they are plainly right that none exists.
The pattern this fits into
Finland is having a version of an argument that has now reached most places where AI infrastructure lands, and the Finnish version is unusually civilized by comparison.
In Texas, Amazon’s plans for a gas plant to power a data center campus put it on track to out-pollute every other power plant in the country. In Oklahoma, the disagreement ended with the Cherokee Nation banning hyperscale data centers outright, even as its own survey found most citizens used AI tools. Against those, a deal that extends a carbon-free plant, adds wind and storage, and brings well-paid permanent jobs to depopulating municipalities is close to the best-case version of this trade.
That is the uncomfortable thing about it. This is what a good outcome looks like, and it still arrives with no permitting framework, no public weighing of competing demands, and a 22-year commitment signed before the debate started. The Finnish opposition is not really trying to stop this deal. It is pointing out that the country has no mechanism for evaluating the next one, and that the next one is already coming.
What to watch
Three things will tell you how this actually goes. The first is whether Finland moves on a national data center permitting regime, since the political energy for it exists now in a way it did not a week ago. The second is Finnish household electricity prices through 2028 and 2030, when the contracted volumes step up, which is the concrete test of whether the government’s reassurance holds. The third is whether other European utilities follow Fortum, because if a hyperscaler power contract turns out to be a reliable way to finance nuclear life extensions, there are a good number of ageing European reactors whose owners would very much like a call from Mountain View.
