Over the past two years, Google has committed to three different nuclear power strategies: buying electricity from an existing plant in Finland, supporting the restart of a shuttered reactor in Iowa, and backing the development of a new generation of advanced reactors in the United States. Together, the agreements show how the rapid growth of artificial intelligence is changing the way technology companies think about electricity.
For years, power was largely treated as an operating requirement for data centers. That assumption is becoming harder to sustain. AI systems require increasingly large amounts of computing capacity, and the data centers running them require a steady supply of electricity. The result is a planning problem that extends well beyond servers, chips and buildings.
Google’s nuclear commitments illustrate one response: secure power years in advance, sometimes through contracts lasting decades.
The pressure behind that strategy is growing. The International Energy Agency estimates that electricity consumption by data centers worldwide will more than double to about 945 terawatt-hours by 2030, with AI expected to be the biggest driver of that growth. In the U.S., data centers are expected to account for nearly half of electricity-demand growth through the end of the decade.
Power Generation Is Only Part of the Challenge
Adequate electricity generation is just one aspect of the problem. A data center can be built relatively quickly, but connecting it to the electricity grid can take considerably longer. In the U.S., Lawrence Berkeley National Laboratory estimates that more than 2,000 gigawatts of generation and storage capacity were seeking interconnection at the end of 2025, highlighting the scale of the country's grid bottleneck.
That makes transmission and grid access increasingly important to technology companies. A long-term power contract has limited value if the electricity cannot reach the facility when the data center is ready to operate.
The issue is particularly important for AI infrastructure because technology companies are planning data-center capacity years ahead, while energy infrastructure often requires much longer development timelines. This creates a potential mismatch between the speed of data-center expansion and the time needed to build generation and transmission capacity.
Three Ways Google Is Securing Nuclear Power
Google’s three nuclear agreements represent three distinct approaches to securing future electricity.
In Finland, the company is buying output from a plant that is already operating. In Iowa, it is helping support the return of a reactor that has been offline since 2020. In the U.S., it is also backing the development of advanced reactors that are still moving through construction and commercialization.
The three arrangements carry different levels of development risk, but they share the same objective: securing a source of long-term electricity for future demand.
Buying Into an Existing Reactor
Google made its first nuclear transaction outside the U.S. in September 2026 when it signed a 22-year power purchase agreement with Finnish utility Fortum for electricity from the Loviisa nuclear facility.
The agreement begins with smaller deliveries in 2028 and reaches the equivalent of up to 50% of the plant’s capacity from 2030 through 2049. The contract gives Fortum greater financial certainty as it invests about €1 billion to extend Loviisa’s operating life to 2050. Without the investment, the plant would not be able to continue operating beyond 2030.
The deal is part of a wider €13 billion Google investment program in Finland covering data centers and supporting infrastructure. The broader energy strategy also includes renewable generation, battery storage and grid-related investments, indicating that nuclear is being added alongside other sources of electricity rather than replacing them.
That distinction is important. Google is not simply buying nuclear power; it is building a broader energy strategy around the infrastructure required to support its data-center expansion.
Restarting a Shuttered Plant
The company has a 25-year contract to purchase electricity from the Duane Arnold Energy Center in Iowa, a 615-megawatt reactor that was shut down in 2020 and has remained offline since then.
NextEra Energy, the utility behind the project, is preparing to restart the plant in early 2029, subject to regulatory approval from the Nuclear Regulatory Commission. The company has also secured a U.S. Department of Energy loan of up to $1.9 billion to help finance the restart.
For Google, the agreement provides a long-term commitment to purchase electricity from a plant that does not yet have that capacity available. For NextEra, the contract provides a long-term customer commitment as it undertakes the costly process of bringing the facility back into operation.
The distinction matters: Google is securing future nuclear capacity, rather than receiving electricity from the reactor today.
Betting on New Nuclear Capacity
Google’s most forward-looking nuclear commitment is with Kairos Power, a startup developing advanced reactors.
Under a 2024 agreement, Kairos is developing a fleet intended to reach 500 megawatts of capacity by 2035. Its first deployment, the Hermes 2 project in Tennessee, is designed to provide up to 50 megawatts and is targeted to begin operations in 2030.
The project has moved beyond the purely conceptual stage, with construction and development activity underway. But the broader commitment remains a bet on a new generation of nuclear technology moving from early deployment to commercial scale.
That makes the Kairos partnership fundamentally different from the Loviisa agreement. Google is supporting a technology whose commercial deployment is still developing, rather than purchasing electricity from an established operating plant.
Google Is Part of a Bigger Shift
Google is not pursuing this strategy alone.
Microsoft and Constellation Energy have a 20-year contract linked to the planned restart of a reactor at the Three Mile Island site in Pennsylvania. Amazon has secured nuclear power agreements, including one connected to Talen Energy’s Pennsylvania facility. Meta has also announced agreements supporting up to 6.6 gigawatts of new and existing nuclear capacity by 2035.
The arrangements differ, but the broader pattern is becoming clearer: major technology companies are increasingly planning their energy requirements alongside their data-center expansion.
They are signing long-term power contracts, supporting the extension or restart of existing plants and, in some cases, helping finance new generation. Their involvement is beginning to extend beyond computing infrastructure to the energy infrastructure needed to support it.
That does not mean nuclear will become the dominant source of electricity for AI. Renewables and natural gas are expected to meet much of the near-term growth in data-center electricity demand, while nuclear is expected to become increasingly important later in the decade and beyond.
Nuclear is therefore better understood as one part of a broader strategy to secure reliable, low-carbon electricity for large and growing data-center loads.
The Real Test Is Delivery
The commitments are significant, but contracts alone do not create electricity.
Duane Arnold still requires regulatory approval before its planned 2029 restart. Kairos must move from its early deployments toward a larger commercial fleet. Google’s Loviisa supply reaches its full contracted level only from 2030.
That gap between signing an agreement and actually receiving power is becoming increasingly important as AI infrastructure expands.
For technology companies, securing chips and data-center sites remains essential. But electricity is moving into the same category of strategic planning. The companies building the next generation of AI infrastructure increasingly have to think years ahead about where their power will come from, how it will reach their facilities and whether the generation capacity will be ready when the computing demand arrives.
Google’s three-track nuclear strategy offers a clear example of that shift: use existing capacity, bring retired capacity back online and help develop new capacity for the future.
The bigger question is whether the energy system can move at anything close to the speed of the AI industry.
