- Meta signs 20-year agreements to purchase up to 6,6 GW of nuclear power in the US and secure supply for its AI data centers.
- The company will combine electricity from existing Vistra reactors with new small modular reactor projects from Oklo and TerraPower.
- The contracts aim to provide low-carbon, stable and continuous energy to support the rise of artificial intelligence until the 2030s.
- The move could pave the way for other technology companies and revive the role of nuclear energy in the global digital ecosystem.
The expansion of artificial intelligence has brought to the forefront a problem that until recently was mostly discussed behind closed doors: There are not enough reliable sources of electricity to power the growth of data centersMeta, the parent company of Facebook, Instagram and WhatsApp, has decided to tackle this energy bottleneck with a major move: relying heavily on nuclear energy.
In recent weeks, the company has closed a series of long-term contracts with operators of traditional nuclear power plants and with startups specializing in modular reactors. The objective is clear: to secure several gigawatts of nuclear capacity for at least two decadesguaranteeing a continuous, low-carbon and sufficiently predictable supply to support its AI ambitions, including the development of advanced models and superintelligence projects.
A change of stage: from renewables to nuclear as the basis of AI
For years, Big Tech has boasted about fueling its operations with wind and solar energyThis model relies on large renewable energy purchase agreements and ambitious climate commitments. However, it begins to show its limitations when it comes to AI data centers that operate 24/7 and cannot depend on whether the sun is shining or the wind is blowing.
In this context, nuclear energy has returned to the forefront as stable base load sourcewith low emissions and continuously available. Meta de facto recognizes that to scale AI at high speed it needs more than just photovoltaic parks and wind turbines: it requires a reliable electrical backbone, although it can later supplement it with renewables where it makes sense.
The consulting firm Grid Strategies estimates that Electricity demand in the United States could grow by at least 30% by 2030And that a large part of that increase will come precisely from data centers. This mismatch between the speed at which digital facilities are being built and the much slower pace of new electricity generation construction has led companies like Meta to secure their power supply years in advance.
Faced with this scenario, Meta's nuclear bet is not only a technological decision, but also a strategic business move: linking the future of its AI infrastructure to multi-decade energy contractseven at the cost of assuming higher initial costs than those of natural gas or some renewables.
For the European and Spanish sector, this move sends a direct message: if you want to attract large data centers and AI projectsIt is no longer enough to talk about fiber optics and talent; the stability of the electricity mix, firm generation capacity and the regulatory framework for nuclear and other dispatchable technologies are gaining prominence.
Key agreement with Vistra: more than 2,5 GW of existing nuclear capacity
The most immediate pillar of Meta's plan involves a long-term contract with Vistra Corp., one of the largest US energy operators. Through this agreement, Meta will purchase electricity from the Davis-Besse and Perry nuclear power plants, both in Ohio.in addition to taking advantage of improvements at another facility in Pennsylvania.
In total, these plants will contribute more than 2,1 gigawatts of nuclear generation already operationalto which will be added an additional 433 megawatts derived from investments to expand capacity at both the Ohio and Beaver Valley (Pennsylvania) power plants. That power flows through the network managed by PJM Interconnection, one of the largest electrical interconnections in the country, serving more than 67 million people in the Midwest and Mid-Atlantic.
The agreements last for a period of about 20 yearsThis is unusual in other industries, but increasingly common at the intersection of technology and energy. With this structure, Meta pursues two objectives: to guarantee supply for its AI data centers in the region and to contribute to extending the lifespan of nuclear power plants that might otherwise face early closures.
The announcement had an immediate impact on the markets. Vistra's shares registered double-digit increases after contracts were made publicThis is a sign that investors perceive Meta's commitment as a very long-term source of income and as a clear endorsement of the economic viability of the existing nuclear plant.
For the European energy debate, where several power plants face decisions about their future, these types of long-term corporate agreements point to a possible model: Large digital consumers are committing to nuclear power plants to ensure both supply and profitability of the facilities..
Oklo: the bet on small modular reactors
Beyond utilizing existing reactors, Meta has decided to strongly support a new generation of nuclear technology: small modular reactors (SMR)designed to be more compact, flexible and potentially cheaper thanks to mass production.
The company has signed an agreement with the startup Oklo, backed by Sam Altman, to secure up to 1,2 gigawatts of electrical capacity sourced from the reactors the company plans to build in Ohio. Each unit of its Aurora Powerhouse design is intended to produce about 75 megawatts, so more than a dozen units would be needed to fulfill the commitment to Meta.
According to current plans, Oklo's first reactor could become operational around 2030, subject to regulatory approvalNo SMR is yet in commercial operation in the United States, so these projects face rigorous scrutiny from the Nuclear Regulatory Commission and other agencies.
The agreement includes a significant financial component: Meta will make a prepayment largely intended for the fuel acquisition already the first phases of developmentThis helps the startup secure resources to advance engineering, permits and construction, in exchange for preferential rights to the energy generated.
From a cost perspective, Oklo aims to position the price of its electricity in the region of $80 to $130 per megawatt hour once their designs are mass-produced. These figures are still theoretical, and the first units are likely to be more expensive, but these projects will serve as a test to see if modular manufacturing can truly lower the costs of nuclear energy in the medium term.
TerraPower and thermal storage to strengthen the grid
Meta's third major nuclear partner is TerraPower, a company co-founded by Bill Gateswhich works on advanced designs for sodium-cooled reactors with integrated thermal storage systems. The idea is to have not only firm nuclear electricity, but also a certain flexibility to respond to peak demand.
With TerraPower, Meta has agreed to support the development of two reactors with a combined capacity of about 690 megawattswith deliveries planned from 2032. In addition, the technology company reserves rights to the energy from up to six additional projects in the future, which could add around 2,1 gigawatts more.
TerraPower's design combines a reactor capable of generating about 345 megawatts continuously with a thermal storage system that would allow the injection of an additional 100 to 500 megawatts for several hours when the grid demands it. This approach aims to ensure that nuclear power is not only a baseload source, but can also provide some buffering capacity.
The company, which is collaborating with GE Hitachi on its first project in Wyoming, has made relatively smooth progress through the regulatory processes, although it will still be years before these reactors are in commercial service. TerraPower estimates that, once its technology is deployed, It could offer electricity in the range of $50 to $60 per megawatt hourcompetitive levels with some renewables and well below the estimated costs of other advanced nuclear proposals.
For Meta, this alliance not only brings additional power in the 2030s, but also diversify the basket of technologies with which to support your AI data centers, distributing the risk between traditional reactors, SMRs and advanced designs with integrated storage.
A 6,6 GW package and a new role for energy in Meta's strategy
Adding together the contracts with Vistra, the projects with Oklo and TerraPower, and a previous agreement with Constellation Energy for a plant in Illinois, Meta estimates that it could secure approximately 6,6 gigawatts of nuclear energy by 2035It is a figure equivalent to the capacity of more than six standard nuclear power plants and enough, according to the company's estimates, to supply a city of around 5 million homes.
Joel Kaplan, head of global affairs at Meta, argues that these steps will transform the company into one of the largest corporate buyers of nuclear energy in US historyThe financial details have not been made public, but analysts such as Jefferies and Bloomberg Intelligence estimate that the average cost Meta could be paying is above $100 per megawatt hour, well above gas or some renewables.
For the technology company, the additional cost is offset by several factors: 24/7 supply guarantee, nuclear fuel price stability versus gas volatilityand a direct contribution to their emissions reduction targets. At a time when some industries are toning down their climate rhetoric, large digital companies like Meta insist on keeping the low-carbon energy label at the heart of their decisions.
In parallel, the company continues to develop other energy projects, such as HyperionA gigantic data center campus in rural Louisiana, potentially reaching 5 gigawatts of power, will initially rely primarily on natural gas-fired power plants operated by Entergy. This demonstrates that Meta's strategy is not purely nuclear, but rather a pragmatic combination of energy sources to ensure an adequate electricity supply.
The result is that energy has become as strategic a component as digital infrastructure itself. At the same time, it announces investments of hundreds of billions of dollars in data centers, chips and networks for AI Until the end of the decade, Meta is working to close the other leg: ensuring there is enough reliable electricity to power everything.
Impact on the global AI race and signals for Europe
Meta's move comes amid a race between tech giants like Amazon, Alphabet, and Microsoft, who have also signed agreements to gain access to long-lasting clean energy, including nuclear, and Google is also investing in nuclear energyHowever, Meta's commitments in terms of volume and timeframe have partially overshadowed the announcements of its competitors.
As generative AI-based services become more widespread and increasingly powerful models multiply, energy infrastructure is becoming a differentiating factor between regions and economiesCountries capable of providing large blocks of firm, low-carbon energy can attract massive investments in data centers and capture some of the added value of the digital economy.
For the European Union and Spain, which are debating the role of nuclear energy in their decarbonization strategies, these agreements serve as a warning: if the digital industry perceives that there will be no nuclear energy, it will be forced to abandon the system. Sufficient stable energy at reasonable pricesLarge AI projects are likely to concentrate in areas with a more ample combination of generation, interconnections, and favorable regulatory frameworks.
In the Spanish case, where the electricity grid is robust and renewables are playing an increasingly important role, the dilemma lies in deciding whether to complement this deployment with tools that provide additional stability—either by extending the life of existing nuclear power plants, investing in new combined cycle plants with CO₂ capture, or exploring, in the future, technologies such as SMRs or nuclear fusion— to reinforce its appeal as data center and AI services hub for Europe.
The figures used by consulting firms like McKinsey & Company regarding the potential of AI —between 2,6 and 4,4 trillion dollars The annual impact on the global economy—which will only be achievable if there is an energy base commensurate with that ambition—is what Meta, with its portfolio of nuclear contracts, is suggesting: that the true limit of AI lies not so much in the algorithms themselves, but in the capacity to generate electricity in a robust and reliable way.
Costs, risks and public debate surrounding the new nuclear wave
Meta's decision also reopens the debate on costs, timelines and risks of nuclear energyBuilding new conventional capacity can cost around $13 per watt installed, while advanced reactor designs like those at Oklo or TerraPower could cost up to $24 per watt, according to BloombergNEF estimates. At the high end, developing 6 gigawatts of new advanced nuclear power would require more than $120.000 billion in capital investment.
In terms of the final price of electricity, analysts consulted indicate that The nuclear energy that Meta contracts can range between $141 and $220 per megawatt hourThis is compared to the typical $50-60 cost of gas, wind, or solar in certain markets. Even so, the company is willing to absorb this cost because it prioritizes security of supply and emissions reduction over minimizing short-term costs.
At the same time, doubts persist as to whether the small modular reactors They will be able to achieve economies of scale comparable to those of current large-scale power plants. Some analysts believe that factory production of standardized modules will reduce costs over time; critics, however, point out that regulatory complexity and safety requirements could diminish some of these promised advantages, and fears stemming from past incidents such as Chernobyl They fuel part of the public debate.
On the political and social front, the shift by Meta and other major technology companies towards nuclear power is interpreted as a shift towards a more pragmatic approach to the energy transitionprioritizing stability and security of supply over certain slogans. The discussion is no longer focused solely on how many megawatts of renewables can be installed, but on how to guarantee that the system can sustain—without blackouts or runaway prices—the massive computing demands of AI.
This debate, already very much alive in the United States, is expected to intensify in Europe as cloud and data center projects grow and it becomes clear whether the electrical grid can absorb them without strain. Meta's experience can serve as a reference point, both for proponents and opponents of nuclear power, regarding What types of agreements and what risks do companies that opt for this route assume?.
Taken together, Meta's steps toward nuclear energy show how the heart of the digital economy increasingly relies on heavy infrastructure decisions: reactors, high-voltage power lines, and 20- or 30-year supply contracts. In the race to lead in artificial intelligence, securing sufficient clean and constant energy is emerging as a key factor. essential conditionAnd nuclear power, for better or for worse, is once again at the center of that game.
