BIG TECHCOMPUTE POWER

In Finland, GOOGLE secures nuclear power for its AI expansion

To contact us: editorial@fw.media

Google plans to invest at least €13 billion in its Finnish infrastructure in 2027 and 2028. But the US group’s most consequential move is an agreement with Fortum covering up to half the capacity of the Loviisa nuclear power plant through 2049. Microsoft, Amazon and Meta had already begun shifting their data centres towards regions with abundant low-carbon electricity. Google is taking the logic one step further.

From 2030, Google will be able to purchase up to half the capacity of the Loviisa nuclear power plant, around 100 kilometres east of Helsinki. The plant’s two pressurised-water reactors have been operating since the late 1970s and each has a capacity of 507 MW. Together, they generate around 8 TWh a year, close to 10% of Finland’s electricity consumption. Once fully ramped up, Google’s contract could therefore represent about 4 TWh annually, equivalent to an average capacity of nearly 450 MW.

That figure reveals what lies behind the announcement of three new data centres in Kajaani, Muhos and Vaala, alongside an expansion of Google’s longstanding Hamina site. The company is no longer merely looking for land, processors and a grid connection. It is assembling the industrial system on which its computing capacity will depend: nuclear power, wind energy, storage, the grid and the ability to shift certain workloads when electricity becomes scarce.

Google plans to invest at least €13 billion in Finland during 2027 and 2028 alone. This is the largest investment it has ever announced in Europe, at a time when Alphabet expects capital expenditure of between $195 billion and $205 billion in 2026.

A 22-year contract to extend Loviisa’s operating life

The power purchase agreement, or PPA, signed with Fortum will begin in 2028 with a limited volume. It will rise to as much as 50% of Loviisa’s capacity between 2030 and 2049, covering almost the entire period during which the utility will need to operate the plant beyond its original lifespan.

Fortum plans to invest around €1 billion between 2023 and 2050 to modernise the two reactors. Nearly €700 million relates to projects that have yet to receive a final investment decision. According to the company, Google’s commitment will provide the financial visibility needed to continue the programme. It will also support a planned 38 MW capacity increase and enable a further 10 MW. Fortum estimates that the agreement will gradually improve the comparable return on its assets by 1.4 percentage points once half of the plant’s capacity is under contract.

Google will not own the plant or directly pay its construction bills. It is instead guaranteeing demand for more than two decades under financial terms that remain confidential. By making future revenue predictable, the agreement allows the utility to justify expenditure today that will be amortised over several decades. The mechanism could inspire other energy suppliers seeking to finance the development of their own facilities. Without the lifetime-extension programme, the reactors could not have continued operating beyond 2027 and 2030 in their original configuration.

The agreement also opens another avenue. Google and Fortum have signed a memorandum to examine new renewable capacity, flexibility resources and potential new reactors at the Loviisa site.

The data centre now comes with its own energy portfolio

The scale of the nuclear agreement should not obscure the broader architecture being built around the future campuses. Google says it has increased the capacity of the new onshore wind projects it supports in Finland to 629 MW. The total includes agreements with Valorem and Suomen Hyötytuuli.

In Kajaani, a 94 MW battery is due to enter service at the end of 2027. Fortum will optimise the asset and may deploy it in the ancillary-services markets operated by Fingrid, Finland’s transmission system operator. The battery will absorb some short-term fluctuations, release energy when wind generation falls and help stabilise grid frequency.

Google is therefore building an energy mix that combines continuous nuclear generation, variable renewables and storage.

The fourth component sits inside the data centres themselves. Google is examining with grid operators whether it can temporarily reduce consumption or shift some computing tasks to other sites during periods of strain. The company had already tested this form of flexibility at Hamina during the 2022–2023 energy crisis.

Not every computing workload can be moved in this way. A search made by a European user or a critical cloud service cannot always wait for the wind to pick up. Training certain models, running simulations and processing batches, by contrast, offer some temporal or geographical leeway. As campuses multiply, the orchestration of computing workloads could itself become a grid-balancing tool.

Microsoft put power before users

Google is not the first company to understand that the geography of cloud computing has become a geography of energy. As early as 2025, Microsoft described the evolution of its strategy as “power first”. The earliest data centres were built close to users. The next generation followed anticipated demand. With AI, some workloads can be directed towards regions where abundant low-carbon electricity is available.

Finland is also serving as a laboratory for Microsoft. The group is developing around a dozen data centres across three sites. In Espoo and Kirkkonummi, two campuses were designed with Fortum to feed waste heat into the district heating network. Around 75% of the heat generated over a year should be recoverable. Eventually, it could cover nearly 40% of the needs of Espoo, Kauniainen and Kirkkonummi. Heat pumps will raise its temperature before it is distributed to homes and businesses across an area with roughly 250,000 residents. Fortum describes the scheme as the largest data-centre heat recovery project of its kind.

The two groups are acting at opposite ends of the chain: Microsoft is finding value in what leaves its server rooms, while Google is securing what must enter them. Microsoft is also working on using its facilities’ backup batteries to help regulate the grid. Google is now combining the same flexibility with a contract capable of supporting the lifetime extension of an entire nuclear plant.

At the same time, Microsoft is pursuing another model by leasing capacity from specialist operators. In Narvik, northern Norway, it has expanded its agreement with Nscale to deploy more than 30,000 additional NVIDIA Rubin GPUs in 2027. The 230 MW hydro-powered campus was originally presented as the first European site in OpenAI’s Stargate programme. Nscale and Microsoft are now assembling one of Europe’s largest concentrations of computing capacity without requiring Microsoft to own the building, the connection and every machine itself.

The coexistence of these two models is revealing. Microsoft builds its own facilities in Finland when they can be integrated into district heating systems. It reserves compute in Norway when a local developer can convert surplus hydropower into GPU capacity more quickly. In both cases, electricity comes before the end customer.

Amazon racks up gigawatts, then runs into the grid

Amazon’s geographic push is broader still. AWS operates infrastructure in more than 20 European countries and is concentrating new investment in Spain, Germany, France, Ireland and Sweden. Its largest recent announcement concerns Aragón, where €33.7 billion is due to be invested by 2035 to expand the Spanish region’s cloud and AI capacity.

The programme extends beyond the data centres in Huesca, Zaragoza and Teruel. Amazon also plans to build a facility for assembling and testing servers, a repair centre for AI equipment and a logistics hub dedicated to its European infrastructure. By adding this supply chain, the group is responding to one of the criticisms directed at large digital campuses: billions may be spent locally while the machines themselves are manufactured elsewhere. Amazon estimates that the programme will support nearly 29,900 full-time-equivalent jobs a year and contribute €31.7 billion to Spain’s GDP through 2035. These company-commissioned projections include direct, indirect and induced employment.

In France, Amazon also plans to invest more than €15 billion between 2026 and 2028. That figure, however, combines cloud infrastructure, AI, logistics and operating expenditure. It cannot be compared directly with the €13 billion announced by Google for its Finnish digital infrastructure.

Amazon has industrialised the use of renewable power agreements. At the beginning of 2025, its European projects represented about 9 GW of wind and solar capacity. Worldwide, the group claims more than 700 low-carbon projects with a combined capacity exceeding 40 GW. Its nuclear investments, notably in X-energy and small modular reactor projects, remain concentrated in the United States.

The limit appears at the point of connection. The time required to strengthen electricity networks is slowing several of AWS’s European programmes. Amazon, Google and Meta have joined GIGA, an association calling on European policymakers to accelerate investment in electricity transmission and distribution. The challenge is no longer simply to sign a PPA guaranteeing that an equivalent quantity of renewable electricity will be generated somewhere on the network. That power must reach the servers where and when they consume it.

Meta went north before the current rush

Meta chose northern Europe long before every new campus was presented as a national AI asset. Luleå, in northern Sweden, offered hydropower and a climate suited to cooling. Odense, in Denmark, combined several hundred megawatts of renewable power contracts with a network capable of recovering heat. Clonee, in Ireland, completed the footprint serving European users.

The group says it has invested more than DKK 10 billion in Odense, €1.4 billion in Clonee and over SEK 8.7 billion in Luleå. According to Meta, each campus supports around 300 operational jobs. The ratio between the capital committed and permanent employment is a reminder of the nature of this industry: large quantities of concrete, cables and processors during construction, followed by highly automated facilities.

Meta thus anticipated the first Nordic generation of cloud infrastructure, built around renewable energy, lower cooling costs and heat recovery. Its largest new investments and nuclear commitments, however, are now concentrated in North America. In Europe, its footprint appears older and less aggressive than the programmes currently being developed by Google, Microsoft and Amazon.

The difference is also one of scale. The first Nordic campuses sought competitive low-carbon electricity. The new generation must reserve several hundred megawatts for GPU clusters, absorb fluctuations in their loads and guarantee their operation even before the grids have been reinforced.

Nuclear power becomes insurance against intermittency and delays

In the United States, the four large hyperscalers have already taken different routes towards nuclear power. Microsoft is backing the restart of a reactor at Three Mile Island. Meta has signed an agreement designed to extend the operation of the Clinton plant in Illinois. Amazon is financing SMR projects, while Google has reserved output from future advanced reactors developed by Kairos Power.

The Finnish agreement brings this movement to Europe, with one important difference: Google is not relying solely on a technology that still has several industrial and regulatory hurdles to clear. Loviisa is already operating. The risk concerns the cost of modernisation, the length of the contract and changes in the market, not whether a first-of-a-kind reactor can be made to work.

Google is not, however, the first data-centre operator to contract nuclear electricity in Europe. In France, Data4 has signed a 12-year nuclear allocation agreement with EDF covering 40 MW. The Loviisa agreement is Google’s first nuclear contract outside the United States and one of the first direct commitments by a major hyperscaler to extending the life of a European nuclear plant.

The return to nuclear power does not mean renewables are being abandoned. It reflects the difficulty of matching continuous demand with variable generation hour after hour. Wind PPAs add low-carbon electricity to the grid, batteries correct short-term imbalances and the modulation of computing workloads reduces certain peaks. Nuclear power supplies the continuity around which these components can be organised.

Compute follows dams, power plants and high-voltage lines

The map of Google’s Finnish programme reflects this new logic. Hamina lies in the south-east, around 55 kilometres from Loviisa. Kajaani, Muhos and Vaala are farther north, close to strong points on the grid and regions where new low-carbon capacity can be installed.

Google says it has worked with Fingrid, Business Finland and local authorities to avoid concentrating all its demand in the south. Consuming electricity closer to where it is generated should reduce the reinforcements needed to transport it across the country.

A company-commissioned study compared a hypothetical 1 GW increase in demand in the Oulu–Kajaani region with an equivalent development in the south. According to the model, choosing the north would save Finnish consumers €520 million over 20 years. The gigawatt remains a study scenario, not the declared capacity of Google’s future sites. The estimate must also be read for what it is: a result produced at the request of the company advocating for these locations.

The principle nevertheless extends beyond Finland. Microsoft and Nscale are converting Narvik’s hydropower into GPU capacity. Amazon is deploying its Spanish AWS Region in an Aragón that offers both land and renewable energy. Meta chose Luleå and Odense. In France, SoftBank plans to invest €45 billion over five years in three campuses representing 3.1 GW, with EDF and Schneider Electric among its partners. Part of the infrastructure is due to occupy a former industrial site. SoftBank explicitly presents France’s nuclear capacity and its status as a net electricity exporter as decisive advantages.

The announced French project is financially larger than Google’s programme. The €13 billion represents Google’s largest investment in Europe, not Europe’s largest AI infrastructure project. The two projects are also at different stages. Google already has its Hamina site, land, a signed nuclear PPA and identified energy contracts. SoftBank’s campuses are due to come online progressively by 2031, and the precise terms of their power supply remain less detailed.

Europe’s compute map therefore no longer overlaps only with its financial capitals and major telecommunications hubs. Workloads that are less sensitive to latency are moving north towards Scandinavian dams, closer to Finnish nuclear plants or in search of capacity from France’s nuclear fleet.

Finland repurposes its old industrial geography

Google acquired a former paper mill in Hamina in 2009 and converted it into a data centre. The site uses cold seawater for cooling. A heat recovery project is intended to cover up to 80% of the local district heating network’s annual needs.

The shift from paper to compute is more than a convenient metaphor. The decline of the forestry industry and other power-intensive sectors has left behind large sites, electricity connections and local authorities looking for new investment. Finland adds a nuclear fleet, substantial wind generation, a cold climate, engineering expertise and the networks of Nokia and Elisa.

This combination helps explain why proximity to Russia, long perceived as a risk, is no longer enough to deter operators. Finland’s accession to NATO in 2023 reduced some security concerns. More importantly, regions capable of accommodating several hundred megawatts quickly have become scarce enough to change the trade-offs.

The investment also comes as the Finnish economy struggles with weak demand, the loss of some trade with Russia, tighter public spending and unemployment above 10%. According to estimates published by Google, the programme would contribute an average of €3.6 billion a year to GDP during construction in 2027 and 2028. It would support more than 37,000 jobs over the period, including around 16,000 in construction, followed by 7,000 direct, indirect and induced jobs each year once the facilities are operating. Google says the programme also includes €31 million for local communities, including €10 million for research and innovation.

These figures do not mean Google will hire 7,000 people to work in its data centres. They combine on-site jobs, suppliers, services and the effects of local consumption. Meta’s European campuses, which each support a few hundred operational roles despite investments often exceeding €1 billion, offer a more realistic order of magnitude for direct employment.

The gap between the €13 billion in spending and the estimated €3.6 billion in annual GDP contribution over two years is also a reminder that some equipment will be imported. Accelerators, servers and network components represent a significant share of the bill.

The real outcome will depend on what remains after the construction companies leave: engineering teams, specialist suppliers, laboratories, training programmes, tax revenue and local companies’ access to the installed capacity.

Hosting the servers without becoming their electricity supplier

Finland is not backing data centres without conditions. The country is preparing a mandatory register to track projects, their consumption and their security implications. It has also introduced a temporary aid scheme tied to site registration, heat recovery or energy efficiency above existing requirements. The Finnish government is seeking to distinguish facilities that deliver measurable economic value from those that merely consume subsidised electricity.

Google joins Microsoft, Nscale, Nebius, atNorth, Pure Data Centres, DayOne and Polarnode in a country where the pipeline already represents several gigawatts. Not all the announced capacity will be built. Permits, customers and, above all, grid connections will make the selection.

Google is trying to answer the criticism in advance. Rather than simply drawing on available power, it is linking its growth to the continued operation of Loviisa, new wind projects, a battery and flexibility mechanisms. The balance will nevertheless have to be assessed using data the company has yet to disclose: the four sites’ combined capacity, their annual consumption, the battery’s storage duration, the PPA’s pricing terms and the additional investment required on the grid.

Another question lies beyond the electricity system. The facilities will be located in Europe, powered by European producers and connected through Nokia and Elisa networks. They will increase the Google Cloud capacity available on the continent. Yet the processors, orchestration software, models, resource allocation and commercial relationship will remain under Google’s control.

More compute in Europe does not automatically mean greater European sovereignty.

Finland will capture the construction work, energy contracts, some of the jobs and business for local suppliers. Google will retain the platform where margins and decision-making power are concentrated. The same imbalance runs through the projects of Microsoft, Amazon, Meta and OpenAI: the territory supplies the electricity and hosts the machines, while the US company chooses the workloads and bills for the services.

Finland’s success will therefore depend less on the number of buildings that rise from the ground than on its ability to create AI companies, laboratories, skills and intellectual property around them. Otherwise, the country will have replaced one part of its power-intensive industry with another — cleaner and better connected, but with its economic centre of gravity in the United States.

EDITORIAL TEAM

To contact the editorial team: editorial@fw.media Our Editorial Policy on Artificial Intelligence : Our analyses and articles are written by journalists. AI may be used as an assistive tool for translation, summarisation, research or stylistic improvement. All facts, figures and analyses are systematically checked and approved by our editorial team. Illustrations generated or modified using AI are clearly labelled.
Back to top button