Product1 publisher3 min readPublished
Google buys 22 years of Finnish nuclear output to get four data centres onto the grid
Google will spend 13bn euros on four Finnish sites in 2027 and 2028. To secure the power for them, it has contracted Loviisa's life extension plus 629MW of new wind.
The Product Desk · Product desk

What happened
- Google says it will spend 13bn euros on data centre infrastructure across four Finnish sites during 2027 and 2028, calling it its largest single investment in Europe.
- The company claims more than 37,000 jobs supported, including some 16,000 in construction and 7,000 continuing roles at wages it puts 24 percent above the Finnish median.
- thenextweb.com reports that nearly two-thirds of Europe's new data centre capacity is now landing outside Frankfurt, London, Amsterdam, Paris and Dublin, pushed out by grid queues and land constraints.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint If a Nordic connection now needs contracted generation behind it, the entry price is a document only a buyer with Google's balance sheet can sign, and a colocation operator sizing a 20MW hall cannot.
- exposure Alphabet has tied 22 years of its power position to a reactor commissioned in the 1970s, which outlasts the economic life of the buildings it was signed to supply.
- decision With no published load figure, any regulator, municipality or rival bidder weighing this against local supply is doing that arithmetic with one side of the equation missing.
- contradiction The same reporting that makes grid access look decisive names land constraints in the same breath, so treating the connection queue as the only binding limit overstates what the evidence carries.
In a site selection deck, the power line used to read "available, subject to connection". In the Nordics that line now needs a named counterparty and a fixed term, a different document going to a different committee than the one that used to sign off on power. Google is contracting generation for buildings that open in 2027 and 2028 [1], and the wind it named is new-to-grid rather than existing supply reallocated [4]. thenextweb.com reads the Fortum agreement as a condition of entry rather than a climate posture, and points to Denmark pausing new connections outright after finding that Europe's cleanest grid could not absorb the demand [8], plus a year of EU messaging asking households to move their electricity use around the load [9].
The economics deserve a slower read than the headline number gets. Spread over two years, 13bn euros is 6.5bn a year [1]. Against that, the claimed 3.6bn euros of annual Finnish GDP during construction [10] is about 55 percent of the annual spend [2]. The jobs figure breaks down into 16,000 in construction and 7,000 continuing, out of more than 37,000 supported [11], which leaves roughly 14,000 in the category nobody audits two years later [4]. Per permanent job, 13bn euros buys 7,000 of them, about 1.9m euros of capital each [3]. That level of capital intensity is simply what a data centre is, and it only misleads someone who reads 37,000 and stops there.
Google published 629MW of nameplate onshore wind through Valorem and Suomen Hyotytuuli [4] and a 94MW battery near Kajaani from late 2027 [5], and no figure at all for what the four sites will draw [6]. That is the checkable part and the gap in one line: nameplate wind is not delivered energy, and the 94MW battery figure is a power rating with no stated duration attached. Without the load number, thenextweb.com notes, nobody outside the company can tell whether the new capacity covers the new demand or merely softens it [6].
The nuclear piece carries the exposure. A 22-year offtake supporting Loviisa, which currently supplies about 10 percent of Finland's electricity [3], attaches Alphabet to the operating economics of a reactor built in the 1970s [12] for longer than most data centres stay economically current [13]. The publisher's read is unsentimental: extending an existing plant is faster than building one, and much faster than waiting for small modular reactors [14].
Grid access has not replaced land as the constraint, on this evidence. thenextweb.com puts nearly two-thirds of Europe's new capacity outside Frankfurt, London, Amsterdam, Paris and Dublin, and attributes that to grid queues and land constraints together [15]. What changed is the assumption sitting underneath the connection, and the queue in Finland is already filling: TikTok is building a second site there for 1bn euros [16] and Nebius has been tripling its Nvidia capacity in the country [17].
For anyone taking a site to approval, that reduces to two questions. Can you state the load in megawatts, and can you name the counterparty and term of the firm power behind it? Answering both gets you an energy plan. Answering only the second gets you what Google published, a document that works when you can write a 22-year offtake and does not work when you cannot.
What to watch
- A disclosed load figure for the four sites, from Google or from a Finnish grid connection agreement, which would let anyone test the 629MW against demand.
- Whether other hyperscalers follow with life-extension offtakes on ageing reactors rather than waiting for new build.
- Whether Finland or another Nordic grid follows Denmark in pausing or queueing new connections.