Google Fortum Nuclear Deal Turns Europe’s Data Center Boom Into a Power Test

Google has signed a 22-year nuclear power agreement with Fortum as Europe confronts a sharp increase in electricity demand from data centers. The Google Fortum nuclear deal covers up to half of Finland’s Loviisa nuclear plant capacity through 2050. It also supports an extensive life extension and power upgrade at the site.

That commitment changes the European nuclear debate. Technology companies are no longer merely buying certificates to match their electricity use with low-carbon generation. Google is helping create the financial certainty needed to keep a specific nuclear plant operating.

The immediate transaction concerns an existing facility, not a new reactor. Yet Google and Fortum also signed a memorandum covering new nuclear capacity, renewable energy, storage, and other flexibility measures. The central contest is now clear: AI infrastructure needs power within years, while new nuclear projects usually take much longer.

The Google Fortum Nuclear Deal Extends Loviisa’s Working Life

Google’s contract gives Fortum a long-term customer for an existing plant, turning data center demand into support for a nuclear life-extension program.

The agreement was announced on September 9, 2026. It contracts up to 50 percent of Loviisa’s generation capacity and runs through the end of the plant’s current operating licenses in 2050.

A power purchase agreement, or PPA, is a contract under which a buyer commits to purchasing electricity over a defined period. Long contracts give generators more predictable revenue, while buyers gain greater certainty over energy supply and related costs.

Fortum says the 22-year agreement provides the economic certainty needed for Loviisa’s life extension and power upgrade. Once half of the plant’s capacity is contracted, Fortum expects the PPA to raise its comparable return on net assets by about 1.4 percentage points.

The agreement does not mean that half of Loviisa’s electricity will travel directly into one Google facility. Both the plant and Google’s operations remain connected to the wider Finnish power system. The commercial contract links their economics even when electrons continue moving through the shared grid.

That distinction matters when discussing nuclear power for AI. The deal is a financial and procurement relationship, not a private cable connecting a reactor to a server campus. Its significance comes from the duration, contracted scale, and investment signal.

Loviisa already produces steady electricity, so the contract can support computing demand sooner than a new reactor could. Extending an existing plant avoids much of the site selection and first-build uncertainty associated with new nuclear construction.

Google’s requirements extend beyond nuclear generation. Fortum and Google also agreed to optimize a new 94-megawatt battery system beside Google’s planned data center in Kajaani. Google has contracted for that battery, which can help balance local demand and power-system conditions.

The broader memorandum covers possible new nuclear and renewable capacity. It also includes flexibility solutions and energy portfolio management. Those provisions make the relationship more than a single electricity purchase, although they are not commitments to build a reactor.

Google separately announced at least €13 billion of infrastructure investment in Finland during 2027 and 2028. The program includes data centers and supporting infrastructure in Hamina, Kajaani, Muhos, and Vaala.

The company estimates that its Finland investment will support more than 37,000 jobs during the initial construction period. It also projects an average annual contribution of €3.6 billion to Finnish gross domestic product during those two years.

Those estimates come from Google and depend on the final scale, timing, and local supply-chain effects of the projects. They should not be treated as verified economic outcomes before construction and spending occur.

Still, the commitments show why the electricity contract matters. Google is not buying nuclear power for a hypothetical expansion. It is pairing the agreement with a defined infrastructure program that places large computing loads across several Finnish locations.

The deal therefore creates a practical precedent. An AI infrastructure buyer is using a long-duration contract to help finance the continued operation of existing European nuclear capacity. That model is more immediate than waiting for an entirely new reactor fleet.

Europe’s Data Center Power Demand Is Becoming a Location Constraint

Europe’s data center race now depends as much on available electricity and grid access as it does on chips, land, or construction capacity.

Data centers consumed about 68 terawatt-hours of electricity across the European Union in 2024, according to the European Commission. One terawatt-hour equals one billion kilowatt-hours, making the total comparable to the annual consumption of a small country.

The Commission expects that use to almost double by 2030 and exceed 3 percent of total EU electricity demand. Its September 2026 data center standards proposal also seeks a common rating system for environmental performance.

The political ambition extends beyond managing existing facilities. The EU wants to triple its data center capacity within five to seven years as it pursues greater digital sovereignty. That expansion places computing policy and energy policy on the same planning horizon.

J.P. Morgan analysts, cited by Investing.com, estimate that European data center electricity demand will increase by another 89 terawatt-hours annually by 2030 compared with 2023. Iberia and the Nordic countries could account for about 45 percent of that increase.

The same analysis says Europe had 10.8 gigawatts of live data center capacity at the end of 2025. Its announced pipeline reached 66.1 gigawatts, although announced capacity does not always become an operating facility.

That gap is important. A project announcement can precede construction by years, and some proposed campuses never secure enough power to proceed. Grid connections, substations, generation, permits, financing, and equipment must all arrive in the right sequence.

The International Energy Agency estimates that servers account for about 60 percent of electricity demand in modern data centers. Cooling, storage, networking, power conversion, and other infrastructure consume the remainder.

AI workloads increase the pressure because accelerated servers use dense clusters of specialized processors. The IEA expects electricity use by accelerated servers to grow much faster than conventional server consumption through 2030.

Its global demand outlook projects data center electricity consumption rising from about 415 terawatt-hours in 2024 to roughly 945 terawatt-hours in 2030. That would remain below 3 percent of total global electricity consumption, but growth will be concentrated in particular regions.

Concentration creates local problems that global percentages can conceal. A national grid may have enough annual generation while lacking transmission capacity near a proposed campus. A region may also have surplus power during windy hours but face shortages during calm winter periods.

Finland offers several advantages for large computing projects. It has a cool climate, established electricity infrastructure, a growing supply of low-carbon generation, and access to Nordic power markets. Waste heat from data centers can also support district heating where networks and customers are nearby.

However, data centers compete with other new electricity users. Electrified industry, battery plants, hydrogen production, transport, and building heating all depend on additional supply. A new computing campus can therefore affect power prices and grid planning beyond the technology sector.

France and the Nordic countries have drawn particular interest because of their nuclear and renewable resources. Iberia offers abundant solar and wind potential, but moving electricity from generation sites to major loads still requires timely grid expansion.

This is why Europe data center power demand is pressuring utilities, regulators, and technology companies at the same time. Utilities must decide which generation investments are bankable. Regulators must allocate scarce connections without sacrificing reliability or industrial policy.

Technology companies face a different constraint. Their facilities can sometimes be designed and built within two or three years, while major transmission and generation projects need much longer. Securing chips and land accomplishes little if a campus cannot obtain a firm connection.

The Google Fortum nuclear deal responds to that mismatch by anchoring existing supply. It does not create all the additional electricity Google will need, but it helps prevent a large low-carbon source from disappearing during the company’s expansion.

AI Needs Power Faster Than Europe Can Build New Reactors

Existing nuclear plants can support the data center boom now, but new reactors cannot solve most of Europe’s near-term computing bottlenecks.

Nuclear power offers attributes that appeal to data center operators. It can generate electricity around the clock, produces low operational carbon emissions, and avoids the weather-driven variation associated with wind and solar facilities.

Those qualities do not make nuclear and renewable power direct substitutes. Wind and solar projects can often be deployed more quickly. Batteries, flexible demand, transmission, hydroelectricity, and interconnection can help manage their variable output.

The IEA expects renewable generation to meet nearly half of the worldwide increase in data center electricity demand through 2030. Nuclear generation becomes more important later in the forecast, particularly as reactors return to service or new capacity begins operating.

That timing defines the main tension behind nuclear power for AI. Data center developers want electricity during the present investment cycle. A conventional nuclear project proposed now is unlikely to supply that demand before the 2030s.

Existing plants offer a nearer-term route. Operators can pursue life extensions, maintenance programs, and power uprates, which increase a reactor’s maximum output. Each still requires safety reviews, investment, skilled workers, and regulatory approval.

Google’s contract shows how a large buyer can reduce some financial uncertainty around those decisions. A committed customer makes future revenue more predictable. It can also justify upgrades that might look less attractive under short-term wholesale pricing alone.

The model resembles developments in the United States, where Microsoft, Amazon, Meta, and Google have pursued nuclear power arrangements. European markets differ in regulation, ownership, grid structure, and national energy policy, so American agreements cannot simply be copied.

Finland is unusually suited to this first move. It operates nuclear plants, participates in a connected Nordic electricity market, and has maintained political support for the technology. The country also recently completed the long-delayed Olkiluoto 3 reactor.

Other European countries remain divided. France is planning new reactors and relies heavily on its existing fleet. Poland is pursuing its first nuclear plants, while the Czech Republic is exploring both large reactors and small modular designs.

Germany closed its final commercial reactors in 2023. Spain’s current schedule calls for its operating units to close between 2027 and 2035. Belgium has changed course on reactor closures, illustrating how security concerns can reshape policy.

These national differences limit the emergence of one European formula. A technology company seeking nuclear-backed electricity must navigate the rules, public attitudes, ownership arrangements, and physical grid conditions of each market.

Google and Fortum also appear willing to combine resources rather than bet on one technology. Their battery project addresses short-duration balancing. The memorandum covers renewables and other flexibility options alongside nuclear power.

That portfolio approach reflects the actual needs of a data center. A campus requires reliable electricity every hour, but reliability can come from a system of generation, storage, interconnection, and controlled demand. It does not require one generator to follow the load alone.

Power purchase agreements also have limits. They can support investment and match consumption financially, but they do not eliminate congestion. If a local network cannot carry enough power, an agreement with a distant generator will not create physical grid capacity.

A further question concerns price. J.P. Morgan estimates that Google’s nuclear agreement carries a premium of roughly 60 percent over forward electricity prices. The contract terms are not fully public, so outsiders cannot independently reconstruct that estimate.

A premium would not necessarily make the agreement irrational. Google may value long-term certainty, low-carbon supply, and faster access to scarce electricity. Avoiding a delayed data center opening can be more valuable than obtaining the lowest short-term power price.

This changes the commercial calculation for utilities. Data center customers can sign contracts longer than many industrial buyers would accept. They also possess strong balance sheets and need enormous volumes of electricity.

The result is a potential financing bridge. Utilities gain predictable revenue for nuclear investments, while technology companies secure credible access to firm low-carbon generation. Whether that bridge supports new construction remains the unanswered question.

The Nuclear Revival Starts With Old Plants, Not New Ones

Europe’s most credible near-term nuclear revival is a program of extensions and upgrades, while entirely new reactors remain a slower and riskier proposition.

The European Commission already expects substantial nuclear investment. Its current program estimates that about €241 billion will be needed through 2050 for new large reactors and lifetime extensions.

EU nuclear plants produced 23.3 percent of the bloc’s electricity in 2024. Installed nuclear capacity is projected to rise from 98 gigawatts in 2025 to about 109 gigawatts by 2050.

Those figures describe a revival in policy support, but not a rapid transformation of the power mix. The projected capacity increase is modest across 25 years. Much of the required spending will preserve existing output rather than create entirely new capacity.

The Commission’s nuclear investment needs also include more than reactor construction. Europe needs supply chains, trained workers, fuel services, waste arrangements, regulatory capacity, and long-term maintenance programs.

Extending existing plants can deliver greater value sooner because their sites, grid connections, and operating organizations already exist. These projects still require extensive safety work, but they avoid starting every planning process from zero.

The Google Fortum nuclear deal fits that pattern precisely. Its first measurable consequence is improved investment certainty for Loviisa. Any contribution to a new reactor would come later and depends on the memorandum producing a bankable project.

This distinction guards against an exaggerated headline. One corporate contract does not establish a continent-wide building boom. It proves that data center demand can influence an operating plant’s investment case.

Large new European reactor projects also carry sobering precedents. Finland’s Olkiluoto 3 and France’s Flamanville 3 both experienced extensive delays and cost escalation. Those outcomes make investors cautious about construction schedules and financing exposure.

Flamanville 3 was connected to the French grid in December 2024, about 12 years later than originally planned. France’s Court of Auditors later estimated its total cost at €23.7 billion and criticized the project’s expected returns.

Olkiluoto 3 entered regular production in 2023 after construction began in 2005. It now provides substantial low-carbon power, yet its delivery history demonstrates how difficult first-of-a-kind European reactor projects can become.

Supporters argue that repeating standardized designs can improve performance. A stable project pipeline may retain skills, strengthen suppliers, and prevent teams from relearning the same tasks at each site.

Critics respond that those benefits remain conditional. Governments must sustain policy across election cycles, regulators must preserve safety standards, and developers must control complex construction programs. Data center customers may not want to carry those risks.

Corporate PPAs could absorb part of the uncertainty, but they cannot erase it. A technology company can commit to purchasing power once available. It cannot guarantee that a reactor receives permits or reaches operation on schedule.

The contract length also creates strategic risk for the buyer. Computing demand may keep rising, but hardware efficiency, AI economics, and facility locations can change. A long energy commitment must remain useful under several technology scenarios.

Another issue is public value. Electricity systems serve households, hospitals, transport, and industry as well as data centers. Policymakers will face questions if scarce low-carbon power appears reserved for foreign technology companies while local prices rise.

The Google Fortum nuclear deal is defensible partly because it supports an existing generator connected to the public grid. Loviisa continues serving the Finnish system rather than becoming an isolated corporate plant.

Future projects could generate sharper conflicts. Governments may provide guarantees, regulated returns, or other public support for new reactors. Voters will ask whether the resulting power primarily benefits domestic users or hyperscale computing campuses.

A European nuclear revival therefore needs more than corporate demand. It requires credible construction performance, clear allocation of financial risk, public acceptance, and a convincing role within the wider electricity system.

Small Modular Reactors Arrive After the Immediate AI Crunch

Small modular reactors match the scale and siting needs of some data centers, but Europe has no operating commercial fleet ready for this decade’s expansion.

A small modular reactor, or SMR, is designed for lower output than a conventional large reactor. Its components are intended to use factory production and repeatable designs, potentially reducing construction complexity.

That concept appeals to data center developers. Capacity could be added in smaller increments, and reactors might sit nearer to major industrial loads. A modular program could also expand as computing demand grows.

However, Europe has not yet demonstrated commercial SMR deployment at scale. Designs still face licensing, financing, supply-chain, fuel, waste, and site-development requirements. Claimed cost advantages depend heavily on producing multiple units.

The European Commission adopted an SMR strategy in March 2026. It targets the first European projects for the early 2030s, which places them outside the most urgent period for data centers planned before 2030.

The SMR strategy calls for coordinated licensing and stronger supply chains. It also promotes a fleet approach, meaning repeated construction of standardized designs instead of isolated custom projects.

That approach addresses a genuine problem. Building one demonstration unit cannot establish low costs. Factories and suppliers need a sustained order book before repetition can produce meaningful economies.

The EU has added financial support, including a €200 million guarantee intended to attract private investment in innovative nuclear technologies. In September 2026, the European Investment Bank also announced support of up to €40 million for Finland’s Steady Energy.

Those measures can advance research, testing, and licensing. They remain small compared with the capital required to build a reactor fleet and the electricity infrastructure surrounding it.

Google’s memorandum with Fortum could eventually support a new nuclear project, including a modular design. Neither company has announced a selected reactor, site, capacity, budget, or construction decision.

It would therefore be premature to present the agreement as an SMR order. The immediate commercial instrument concerns Loviisa’s existing units. The new-build language establishes a route for further work, not a final investment commitment.

The timeline problem remains decisive. A data center entering service during 2028 cannot wait for a first European SMR in the early 2030s. Developers must rely on available generation, grids, efficiency, storage, and faster renewable additions.

SMRs could affect the following investment cycle if initial projects operate reliably. Successful licensing and construction would give technology companies a new option for campuses planned in the 2030s.

Failure would have the opposite effect. Delays, rising costs, or design fragmentation would reinforce dependence on existing reactors and broader grid portfolios. Developers might then favor markets with abundant renewable generation, gas backup, hydroelectricity, or established nuclear fleets.

The sector should watch actual milestones rather than partnership announcements. A signed customer agreement, completed safety review, financed construction plan, and binding delivery schedule reveal more than a general memorandum.

Nuclear power for AI will also compete with other applications. SMRs may serve district heating, industrial steam, hydrogen production, or remote grids. Data centers will not automatically receive the first units.

The technology’s eventual role depends on whether developers can align standardized production with Europe’s national regulatory systems. Fragmented requirements would weaken the manufacturing logic behind modular construction.

For now, SMRs strengthen the long-term nuclear narrative but do little for immediate connection queues. Europe’s near-term answer remains less dramatic: retain safe reactors, expand grids, build renewables, add storage, and improve computing efficiency.

Three Signals Will Show Whether Europe’s Nuclear Shift Is Real

The next phase will be measured by binding investments and delivered power, not by the number of companies expressing interest in nuclear energy.

The first signal is Fortum’s execution at Loviisa. Investors should watch the scope, regulatory progress, timing, and cost of the plant’s life-extension and power-upgrade program.

Successful execution would strengthen the case that hyperscale contracts can preserve existing nuclear output. Delays or rising costs would weaken claims that life extensions offer an easy answer to Europe data center power demand.

The second signal is whether Google and Fortum convert their memorandum into a defined new-generation project. A meaningful step would identify a technology, site, capacity, regulatory route, and financing structure.

Another broad cooperation statement would not establish a revival. A final investment decision backed by a long-term customer would provide much stronger evidence that corporate electricity demand can unlock new nuclear construction.

The third signal is the relationship between announced data center capacity and completed grid connections. Europe’s pipeline is much larger than its live capacity, so delivery rates matter more than headline totals.

If Finland and other Nordic markets connect new campuses without major price or reliability problems, their combination of nuclear and renewable power will gain credibility. Persistent delays would show that generation contracts alone cannot overcome transmission and local network constraints.

These signals should be evaluated together. A successful Loviisa extension preserves supply, a new project expands it, and completed connections prove the grid can deliver it where needed.

The Google Fortum nuclear deal has already crossed one important threshold. An AI infrastructure buyer is supporting a named European plant through a multidecade electricity commitment. That is more concrete than a corporate pledge to investigate nuclear power.

It still falls short of proving that Europe can build new reactors at the speed AI companies demand. The continent’s recent nuclear projects warn against assuming that political support or customer interest guarantees timely delivery.

The most likely path combines several technologies. Existing nuclear plants provide firm low-carbon output. Renewables add capacity more quickly, batteries manage short fluctuations, and stronger grids share electricity across regions.

Efficiency must remain part of that mix. More capable chips and better cooling can reduce electricity used for each computing task. Yet lower unit consumption can be offset when companies deploy far more servers and offer more AI services.

For developers and enterprise technology buyers, electricity availability is becoming a product constraint. It can determine where cloud capacity appears, how quickly services expand, and whether regional computing costs remain competitive.

The critical question is no longer whether AI companies are interested in nuclear power. They clearly are. The test is whether their contracts finance safe capacity that arrives before power shortages slow the infrastructure they were meant to support.

Watch Loviisa’s upgrade, the first binding new-build proposal, and actual data center connections through 2030. Together, those results will show whether Europe is entering a nuclear revival or extending existing assets while the harder construction challenge remains unresolved.