A picture taken on September 10, 2011 shows the Google data center in Hamina, Finland during it's official opening. JARNO MELA/LEHTIKUVA/AFP via Getty Images

Google announced on September 9 that it would pour at least €13 billion (approximately $15.1 billion) into digital infrastructure in Finland over 2027 and 2028 — the company's largest single investment in Europe. The commitment spans four data center sites, a portfolio of clean energy contracts totaling 629 megawatts (MW) of new wind capacity, a grid-scale battery system, and a community investment fund worth €31 million (approximately $36 million). But the deal's most consequential element arrived almost without ceremony in the energy section of the press release: a 22-year power purchase agreement (PPA) with Finnish utility Fortum tied to the Loviisa nuclear power plant — an arrangement that, according to both parties, will keep a reactor complex that supplies 10 percent of Finland's electricity from shutting down in 2030.

That detail reframes the announcement entirely. Google is not simply buying computing capacity and sourcing power opportunitically. It is providing the long-term revenue certainty that allows Fortum to proceed with a roughly €1 billion (approximately $1.2 billion) modernization program and to justify operations at Loviisa through 2050. Without the PPA, Fortum stated, the plant could not have continued beyond 2030. A data center company's demand for AI compute has, for the first time on European soil, become the financial mechanism that sustains a nation's nuclear energy asset.

Finland Won the AI Infrastructure Race — Here Is the Technical Reason

The choice of Finland is not coincidental, and the competitive reasons are deeply embedded in the engineering of data center operations.

Data center power usage effectiveness (PUE) — the ratio of total facility power to power actually reaching computing equipment — is the primary operating efficiency metric in the industry. A PUE of 1.0 is theoretical perfection; most facilities run between 1.3 and 1.5, meaning at least 30 percent of all electricity consumed goes to cooling, not computing. At Hamina, Google has demonstrated for more than a decade that cold-climate, free-cooling architectures can approach PUE levels closer to 1.1. The mechanism it uses there — seawater cooling drawn directly from the Gulf of Finland — is technically unique and not replicable everywhere, but it illustrates the performance ceiling that cold geography enables.

The Hamina cooling system works as follows: Cold seawater is pumped from the Gulf through tunnels carved into the solid granite bedrock that once served the paper mill Google purchased in 2009. That water passes through a two-stage heat exchanger — seawater on one side, server cooling fluid on the other — transferring thermal energy without the seawater itself ever contacting the server cooling circuit. The separation is critical: seawater's dissolved salts and oxygen would corrode metal components in a matter of months if brought into direct contact with the equipment. After picking up heat, the warmed return water is mixed in a temperature-equalization building with incoming cold seawater before it is returned to the Gulf, minimizing thermal impact on the marine environment. Mechanical chillers — the energy-intensive refrigeration compressors that cooling systems in warmer climates depend on — are largely eliminated. The result is a data center that dissipates the heat of a large city's computing infrastructure using nothing more than gravity-fed seawater and a pair of heat exchangers.

That same heat, rather than being wasted, now flows somewhere useful. Google's heat recovery system at Hamina routes the thermal energy from the server cooling loop into the district heating network operated by Haminan Energia, the local municipal utility. When fully operational, this waste heat covers 80 percent of Hamina's annual residential and commercial heating requirements, at no cost to local customers. In a country where district heating covers roughly half of all buildings and winter heating costs are a significant household expense, the effect is material and visible. Google has stated that all new data center sites in Europe will be designed from the outset with heat recovery readiness built into the facility — not retrofitted after the fact.

The three new sites Google announced — Kajaani, Muhos, and Vaala, all in north-central Finland — were chosen in part because they sit inside what Finnish grid operator Fingrid describes as a production-dominated northern zone: a region that generates substantially more electricity than it locally consumes and whose surplus flows south to population centers. Building data centers at the point of excess generation, rather than at the point of consumption, reduces the need for new transmission infrastructure and lowers overall system costs for all Finnish electricity users — a point Fingrid's President and CEO Asta Sihvonen-Punkka explicitly endorsed: "These locations at strong grid points both help the whole electricity system and ensure cost-effective network development."

Finland's grid has another structural advantage that is becoming increasingly rare: it is already overwhelmingly carbon-free. Nuclear power generated approximately 38 to 40 percent of Finland's electricity in 2024 and 2025, with wind providing roughly 24 to 27 percent, hydropower 15 to 17 percent, and bioenergy another 11 to 13 percent. The country's overall carbon intensity runs at approximately 88 to 95 grams of CO₂ equivalent per kilowatt-hour, well below the European average and dramatically below markets still heavily dependent on gas or coal. For a company under investor and regulatory pressure to match AI compute load with genuinely carbon-free generation — not just offset credits — Finland offers a ready-made grid profile that most European nations cannot match.

How the Loviisa Nuclear Deal Actually Works

The Loviisa nuclear power plant, located on a small island on the southern Finnish coast near the town of Loviisa, houses two Soviet-designed VVER-440 pressurized water reactors that have been in continuous commercial operation since 1977 and 1981 respectively. Together they produce 1,014 MW of nameplate capacity, running at an availability rate of approximately 89 percent — a figure that places them among the more reliable nuclear generating assets in Europe. The plant accounts for a disproportionately large share of Finland's baseload power and, because it operates at high capacity factor around the clock regardless of weather, provides the stability that allows large volumes of variable wind and solar capacity to be integrated into the Finnish grid without threatening reliability.

The problem Fortum faced before the Google PPA was straightforward: the plant's existing electricity sales contracts expire around 2030, and without long-term revenue certainty, Fortum could not justify the roughly €1 billion (approximately $1.2 billion) in modernization investments — replacement of low-pressure turbines, turbine automation renewal, seawater pump and motor replacement — that are required to operate the facility safely through its Finnish government-granted license period ending in 2050. Each of those investment projects requires separate justification. Without a buyer committed to taking a substantial share of Loviisa's output for two decades, that justification did not exist.

The Google PPA provides it. Initial volumes under the agreement are modest when deliveries begin in 2028, scaling to approximately 50 percent of Loviisa's output across the 2030 to 2049 period — precisely the period when the modernization investments need to be in service and when the plant would otherwise have been shut down. Fortum's President and CEO Markus Rauramo described the arrangement as providing "a strong foundation for the continued development and reliable operation of our Loviisa power plant for decades to come." Fortum shares rose sharply on the day of the announcement.

The agreement goes further. Alongside the PPA, Google and Fortum signed a memorandum of understanding to explore cooperation on new power generation capacity, including potentially new nuclear reactors at Loviisa. The PPA is, in this sense, not just a bilateral transaction but a proof of concept for a new financing instrument in European energy. If a hyperscaler's AI compute revenue can sustain a legacy nuclear plant's economic life, the same template could in principle be applied to aging nuclear assets in France, Belgium, the Netherlands, and elsewhere in Europe — markets where operators have struggled to justify continued operation against lower-cost intermittent renewables that cannot themselves provide the 24/7 baseload profile AI data centers require.

Why AI Workloads Specifically Demand Nuclear Baseload

Understanding why the Google-Fortum structure makes engineering sense requires understanding what AI compute workloads demand from the grid, and why intermittent renewables cannot fully address that demand.

Large language model inference — the process of generating a response to a prompt, which happens billions of times per day across services like Google Search, Gemini, and Maps — runs continuously, around the clock, at near-constant power draw. Unlike traditional data center workloads, which surge during business hours and drop overnight, AI inference has no quiet period. The same is true of model training, which runs in intensive multi-week batches that require uninterrupted power for the duration. A data center whose power supply fluctuates with wind availability or solar irradiance cannot run these workloads without battery backup on a scale that is currently economically and physically impractical for meeting 100 percent of load.

Nuclear baseload is the natural complement: constant output, near-100 percent capacity factor regardless of weather, zero direct carbon emissions. The challenge until recently was that nuclear power's economics required decades-long revenue certainty to justify both construction and continued operation — certainty that electricity spot markets and short-term contracts could not provide. Hyperscalers, whose AI infrastructure roadmaps now run in 10- to 25-year horizons, can offer exactly that revenue certainty, and the Loviisa deal demonstrates the match is workable.

Google's commitment in Finland is the first European instance of this structural alignment. In the United States, the company had already committed to purchasing power from Kairos Power's Hermes small modular reactor in Oak Ridge, Tennessee, and signed a 25-year agreement with NextEra Energy to restart the Duane Arnold Energy Center in Iowa. Microsoft secured Three Mile Island power; Amazon contracted 1.92 gigawatts (GW) from Talen Energy's Susquehanna nuclear station. The US precedents existed. What the Loviisa deal adds is the first European example, and — crucially — the first deal in which the PPA is explicitly identified as the mechanism that prevented a plant from closing, rather than simply an additional customer for power a plant would have generated regardless.

Grid Insurance: The 94 MW Battery and 629 MW of Wind

The nuclear PPA resolves the baseload problem. Finland's remaining grid vulnerability in a heavily wind-reliant future is the cold, windless winter high-pressure period — days when wind generation collapses, temperatures drop, and heating demand peaks simultaneously. That is the scenario the 94 MW battery system near Kajaani is designed to address.

Battery systems at this scale function as frequency-response providers in the Finnish grid: they charge during periods of excess generation and discharge within milliseconds when generation drops, stabilizing the grid's frequency to the 50 Hz standard. Unlike batteries deployed solely to store energy for later use, a frequency-response battery earns revenues from Fingrid's ancillary services market independent of Google's own load management. The system, targeted for operation in late 2027, will be connected to the national grid and made available for this ancillary function — meaning its stabilization benefit accrues to all Finnish electricity users, not only to Google's campuses.

The 629 MW of new onshore wind capacity contracted through PPAs with Valorem (in the Ostrobothnia region) and Suomen Hyötytuuli (Ostrobothnia and Central Finland) adds new generation to Finland's grid rather than simply procuring from existing assets. Both the wind PPAs and the battery system are structured to participate in Fingrid's ancillary markets, allowing them to contribute to grid balancing beyond Google's direct demand.

What €13 Billion Means Inside Google's Broader Spending

The Finland commitment does not exist in isolation. Alphabet raised its full-year 2026 capital expenditure guidance at its second-quarter earnings call to between $195 billion and $205 billion, up from a prior range of $180 billion to $190 billion, driven by demand for AI infrastructure and cloud services. Google Cloud revenue rose 82 percent year-on-year to $24.8 billion in the second quarter; the company's cloud backlog stood at $514 billion; Gemini was processing 22 billion application programming interface tokens per minute; and approximately 90 percent of Fortune 100 companies were using Gemini Enterprise. The capital commitment to Finland, representing roughly $7.6 billion per year across 2027 and 2028, sits inside a total annual capex envelope that dwarfs it — but its concentration in a single country, oriented around a single energy thesis, distinguishes it from the more distributed spending behind quarterly capex numbers.

The thesis is legible. Finland offers a rare combination of ready nuclear baseload, abundant low-cost wind, a cold climate that reduces cooling energy costs, a stable regulatory environment, an existing 15-year operational footprint at Hamina, and a grid operator — Fingrid — actively partnering to site data centers at optimal grid connection points. Ruth Porat, Alphabet's President and Chief Investment Officer, said the investment reflects "Google's commitment to grow our presence responsibly, pairing the expansion of our technical infrastructure with new energy capacity, grid enhancements, and energy affordability initiatives."

For Finland, the economic implications are commensurately large. During the construction phase in 2027 and 2028, the investment is expected to contribute an average of €3.6 billion (approximately $4.2 billion) per year to Finland's GDP — equivalent to roughly 1.3 percent of the country's 2025 GDP — and to support more than 37,000 jobs nationwide, approximately 16,000 of them in construction. Once operational, the sites are projected to sustain 7,000 jobs annually at wages averaging 24 percent above Finland's median. Finnish Prime Minister Petteri Orpo said the investment was "a clear testament to our strengths."

What Comes Next

Data center construction at the four sites is scheduled to begin in 2027. The 94 MW Kajaani battery system carries the most near-term timeline, targeted for grid connection in late 2027. The nuclear PPA begins delivering initial volumes in 2028, scaling to its contracted level across the 2030–2049 window.

The architectural specifications of the compute facilities — GPU configurations, rack density, and whether individual sites will primarily serve training or inference workloads — have not been disclosed. Given the pace at which AI hardware generations turn over, Google is unlikely to lock those decisions in advance of commissioning. What is structurally defined, and what the Loviisa PPA makes possible, is the energy baseline: clean, firm, carbon-free baseload that can absorb continuous AI compute load without carbon compromise or grid-frequency risk for two decades.

The broader signal is what Fortum CEO Markus Rauramo articulated in his statement: "Long-term partnerships like the one between Fortum and Google are essential to making that happen, especially in today's uncertain market environment characterized by low visibility and highly volatile electricity prices." The deal structure answers that volatility with commitment — and in doing so, creates a template other European utilities and hyperscalers will be able to study. Whether it can be replicated in markets where regulatory frameworks are more complex, and where aging nuclear assets do not have the operational track record of Loviisa, remains to be seen. But the first European proof of concept now exists in Finland.

(Exchange rate as of September 9, 2026; all conversions are approximate.)

Frequently Asked Questions

Why did Google choose Finland for its largest European data center investment?

Finland offered a combination of factors that few other European markets can match simultaneously: nuclear baseload power generating roughly 38 to 40 percent of national electricity; a cold climate and coastal seawater access that enable near-eliminating mechanical cooling costs; a production-surplus northern grid that allows data centers to be sited at existing high-capacity connection points without triggering major new transmission investment; and a 15-year established operational footprint at Hamina that Google could expand rather than build from scratch. The Loviisa nuclear plant — whose 10 percent share of Finland's grid provides the continuous, weather-independent baseload AI data centers require — was the energy anchor that made the case compelling.

What is a power purchase agreement (PPA), and why does the Loviisa deal go further than a standard one?

A standard corporate PPA is a long-term electricity supply contract that gives a buyer price certainty and allows a generator to finance capital investment. Most corporate PPAs signed by tech companies to date have contracted electricity from wind and solar farms, providing those farms' developers with the revenue certainty they need to obtain project financing. The Loviisa PPA is structurally different in one critical way: Fortum has stated that without it, the plant could not continue operations beyond 2030, when its existing electricity contracts expire. Google's 22-year commitment provides the revenue floor that justifies Fortum's roughly €1 billion ($1.2 billion) modernization program — including turbine and pump replacements required to safely operate the plant through its government-granted license period ending in 2050. The deal is not just a procurement contract; it is a capital formation instrument for a national energy asset.

Could the same model save other aging nuclear plants in Europe?

That is the deal's most significant unexplored implication. Aging nuclear plants across Europe — in France, Belgium, the Netherlands, and beyond — face the same economic challenge Loviisa faced: existing revenue contracts expire, capital-intensive life-extension work is required, and electricity spot prices alone do not justify the investment. Hyperscalers with multi-decade AI infrastructure roadmaps could in principle provide the long-term revenue certainty those plants need, just as Google did at Loviisa. The model requires alignment of several conditions: a plant with a credible operational track record, an energy regulatory environment that permits bilateral long-term contracts, and a buyer whose AI compute load is large enough to underwrite a meaningful share of the plant's output. Whether those conditions converge elsewhere in Europe will depend partly on how clearly this deal demonstrates that the structure works — which makes the next two years of Loviisa's operating history unusually consequential for the European energy transition.

How does Google's seawater cooling system in Hamina work, and will new sites use it?

The Hamina system draws cold seawater from the Gulf of Finland through granite tunnels built for the former paper mill, passes it through direct water-to-water heat exchangers to cool the server cooling circuit, and returns it to the Gulf after mixing it with incoming cold water to minimize thermal impact. The two-stage heat exchange prevents the seawater's corrosive salt content from reaching the server hardware. Mechanical refrigeration chillers — which typically account for a large fraction of a data center's total energy consumption — are largely eliminated. The three new sites in north-central Finland will not replicate seawater cooling, which requires coastal access, but all new European data centers are being designed from the outset with heat recovery readiness, enabling waste heat at 40 to 60 degrees Celsius (104 to 140 degrees Fahrenheit) to feed into district heating networks rather than being dissipated to the atmosphere.