Google Bets €13B on Finland: Nuclear Power Meets AI Data Centers

Hyperscalers have spent 2026 racing to lock down power for AI data centers, and nuclear has become the preferred long-term contract structure. On September 9-10, 2026, Google announced its largest move yet in that pattern: a €13 billion ($15 billion) investment in Finland, anchored by a 22-year contract to buy up to half the output of a nuclear power plant.

TL;DR

The nuclear power purchase, not just the data centers

The headline figure is €13 billion, but the mechanism worth understanding is the power contract underneath it. Google signed a 22-year agreement with Fortum, the Finnish utility, to purchase up to 50% of the electricity generated by the Loviisa nuclear power plant. Loviisa currently supplies roughly 10% of Finland's total electricity — meaning a single hyperscaler's contract now claims a meaningful share of one nation's nuclear generation capacity.

This is a power purchase agreement (PPA), the same contract structure explainx.ai has tracked across Meta, Microsoft, and Amazon's nuclear deals in Every Hyperscaler Has a Nuclear Deal — but the Loviisa deal stands out for locking in an existing, operating plant's output at scale, rather than funding a new reactor build or an option on future advanced-reactor capacity. Fortum's statement frames the benefit explicitly both ways: Google gets long-term power certainty, and Fortum gets the revenue certainty to justify a roughly €1 billion investment program extending Loviisa's operating life to 2050 and potentially expanding its output.

What the money actually builds

Beyond the power contract, the €13 billion package funds:

• Three new data centers in Kajaani, Muhos, and Vaala
• Expansion of Google's existing Hamina facility — notably a converted paper mill Google first established as a data center in 2009, now getting new capacity
• Clean energy and grid projects, plus community funds Google describes as supporting "local biodiversity, education, research, and workforce development"

Google's own framing is careful to note the infrastructure serves "products such as its AI chatbot Gemini, alongside services including Search, Maps and YouTube" — this is general-purpose compute capacity, not a Gemini-exclusive buildout, consistent with how hyperscalers typically frame these announcements to avoid overstating the AI-specific share of a broad infrastructure investment.

Why Finland, and why now

Finland's pitch to hyperscalers rests on three structural advantages Google's own statement and Finland's government echo:

1. Cool climate — lower ambient temperatures cut the energy needed to cool server racks, a meaningful line item at data center scale
2. Low-carbon, plentiful electricity — Finland's grid mix, including nuclear, gives hyperscalers a path to net-zero power claims without the carbon offsets other regions require
3. Uncongested grid — unlike parts of the US and Western Europe where new data center interconnection requests face multi-year queues, Finland's grid has more available headroom

Google isn't alone in reaching that conclusion this month: TikTok separately announced a $1 billion investment in a new Kouvola data center the same week, citing nearly identical factors — "strong digital infrastructure, clean energy mix, robust data governance and skilled tech talent." Two major tech platforms making large, independent Finland bets in the same week is a strong signal the country has become a deliberate compute-buildout destination, not a one-off Google decision.

The bigger pattern: AI compute now runs on direct-to-generator contracts

This deal extends a 2026 trend explainx.ai has tracked across the industry: hyperscalers increasingly bypass general grid electricity purchases in favor of direct, long-term contracts with specific power generators — nuclear plants especially, because they offer 24/7 baseload output without the intermittency of solar or wind. Every Hyperscaler Has a Nuclear Deal tracked roughly 9.8 GW of hyperscaler nuclear commitments across Meta, Google, Microsoft, Amazon, and Nvidia as of August 2026 — a figure that mixes power purchases, reactor-development options, equity investments, and permitting partnerships that aren't all equivalent. The Loviisa deal is a clean example of the most concrete category: an actual power purchase agreement against existing, operating nuclear capacity, not a future-dated option on a reactor that doesn't exist yet.

Alphabet's broader capital picture makes the scale of this bet legible: the company raised its global spending plans to as much as $205 billion earlier this year specifically to expand AI computing capacity, per prior reporting. A €13 billion Finland package is one country-level slice of that much larger global buildout.

What "up to 50%" actually means for grid planning

A 22-year contract sounds like a fixed commitment, but "up to 50%" is a ceiling that leaves both sides flexibility year to year. In practice, long-duration power purchase agreements like this typically specify a baseline volume with the buyer able to flex up toward the cap during periods of high demand, while the seller retains the remainder for domestic grid needs, other industrial customers, or spot-market sales when prices are favorable. For Fortum, that structure converts a plant that has spent recent years facing uncertain long-term economics into one with a large, contracted anchor customer for over two decades — the kind of revenue visibility that makes a billion-euro life-extension investment underwritable in the first place.

For Finland's grid operators, the practical question is how much of Loviisa's output effectively gets reserved for one customer versus staying available for the broader market. A plant supplying roughly 10% of national electricity committing up to half its output to a single hyperscaler is a meaningful reallocation, even before accounting for the three new data centers' own direct electricity draw layered on top.

The compute-to-power ratio this deal implies

Data center power demand scales with the compute inside it, and Google's own framing — supporting "Gemini alongside Search, Maps and YouTube" — signals this is provisioning for years of anticipated AI-driven growth in query volume and model complexity, not a one-time capacity bump. The three new Finnish sites (Kajaani, Muhos, Vaala) plus the Hamina expansion effectively give Google a second major Nordic compute hub layered onto its existing Nordic and European footprint, positioning EU-based inference and training workloads closer to European users and inside the EU's own regulatory perimeter — relevant given the compliance considerations explainx.ai covered in Europe's sovereign compute push.

What this means if you build with AI

For practitioners, the Finland deal is less about picking a Google product and more about where inference economics are heading. When a hyperscaler signs a 22-year nuclear PPA, it is buying predictable baseload power at a price it can amortize across Search, YouTube, Maps, and Gemini — which means Google can keep Gemini API pricing and free-tier limits competitive longer than a provider renting spot grid power in a congested market. That does not guarantee cheaper tokens for you tomorrow, but it explains why Google, Meta, Microsoft, and Amazon are all racing for the same contract structure: energy cost is now a first-class input to model serving, not a facilities footnote.

If you run local or colo inference, Finland's pitch — cool climate, low-carbon mix, grid headroom — is the same reason TikTok announced a $1B Kouvola build the same week. EU teams evaluating sovereign compute should read this alongside explainx.ai's Europe AI landscape guide: Finland is positioning as a Nordic compute hub, not just a Google one-off.

Honest limitations

• "Up to 50%" is a ceiling, not a guarantee. The contract terms allow Google to purchase up to half of Loviisa's output; actual drawn volume may vary by year and by Fortum's other commitments.
• No public per-megawatt-hour pricing disclosed. As with most hyperscaler PPAs, the announcement doesn't include the price Google is paying — a meaningful gap for anyone trying to model the deal's actual economics.
• Construction timeline runs to 2028. Google states construction on the new sites takes place in 2027 and 2028 — this is a multi-year buildout, not capacity available immediately.
• Jobs and GDP figures are Google's own estimates ("more than 37,000 jobs during construction" and "€3.6bn a year" GDP boost) — these are company projections, not independently verified figures, and construction-phase jobs are typically temporary rather than permanent headcount.
• Nuclear PPAs do not eliminate grid bottlenecks everywhere. Finland's relatively open grid is part of the story; copying the contract structure in a congested US interconnection queue does not automatically deliver the same timeline.

The takeaway

Google's Finland deal is the clearest example yet of a pattern that now defines hyperscaler AI infrastructure spending: lock in decades of guaranteed nuclear power alongside the data centers themselves, so compute buildout and energy buildout happen as one contract rather than two separate bets. At €13 billion and a 22-year power agreement against an operating plant, it's also the largest and most concrete deal in that pattern to date — a bar the next hyperscaler nuclear announcement will be measured against.

Related on explainx.ai:

• Every Hyperscaler Has a Nuclear Deal — Here Is What Each Actually Bought — the full comparison table this deal now extends
• Elon Musk: 15GW AI Compute Power Bottleneck — why power, not chips, is the binding constraint on AI buildout
• RAM Prices and AI Demand — another resource squeeze from the same compute boom
• NVIDIA GB300, TSMC Arizona, Amkor Packaging — the chip-manufacturing side of the same buildout
• Europe AI Landscape: Sovereign Compute and the EU AI Act — where Finland's investment fits in Europe's broader AI infrastructure picture

Details in this post reflect Google's and Fortum's September 9-10, 2026 announcements as reported by the BBC and other outlets. Construction timelines, pricing, and final contract terms should be confirmed against Google's and Fortum's official statements.