Why Nuclear Power Is Becoming Technology Infrastructure

For most of the modern technology era, nuclear power and computing belonged to industries that rarely appeared in the same strategic conversation. Nuclear generation was discussed through the language of utilities, electricity markets, engineering, environmental policy and national energy security, while the technology sector concerned itself with processors, software, telecommunications networks, cloud infrastructure and increasingly sophisticated digital services. Artificial intelligence is beginning to collapse that separation because the expansion of machine intelligence requires such large quantities of dependable electricity that the characteristics of the power system beneath computing are becoming relevant to the economics of computing itself.

This does not mean nuclear power has suddenly become the only rational energy source for artificial intelligence, nor does it mean every data center should be placed beside a reactor. Natural gas, renewables, hydroelectricity, geothermal resources, storage and transmission expansion will all contribute to meeting rapidly growing electrical demand according to their economics and geography. What is changing is the strategic category in which nuclear power belongs. As computational infrastructure becomes larger, more capital intensive and more continuously utilized, nuclear generation increasingly looks less like a distant utility-sector concern and more like part of the physical technology stack upon which the digital economy depends.

That shift can already be seen in the behavior of some of the world's largest technology companies. Microsoft entered a twenty-year power purchase agreement supporting the planned restart of the former Three Mile Island Unit 1, now called the Crane Clean Energy Center, with Constellation expecting the restored plant to return approximately 835 megawatts of generating capacity to the PJM grid. Meta subsequently entered its own twenty-year agreement with Constellation for the 1,121-megawatt Clinton Clean Energy Center in Illinois, while Google has contracted with Kairos Power around a planned fleet of advanced reactors and Amazon has invested in nuclear projects and small modular reactor development as part of a much larger effort to secure future electrical supply.

These are not isolated curiosities occurring at the edges of corporate sustainability programs. They reveal something more economically consequential about the emerging relationship between energy and computation: technology companies whose products exist at the frontier of machine intelligence are beginning to participate directly in the financing, preservation and development of electrical generation assets because access to sufficient power is becoming part of their ability to scale.

The Workload Has Changed the Energy Conversation

Cloud computing has always required electricity, but the growth of AI changes the scale and intensity of the requirement. The International Energy Agency projects global data-center electricity consumption rising from roughly 485 terawatt-hours in 2025 to approximately 950 TWh by 2030, while electricity consumption associated specifically with AI-focused facilities grows considerably faster and roughly triples over the same period. In the United States, the IEA expects data centers to account for nearly half of total electricity-demand growth through the end of the decade.

Those numbers help explain why the energy conversation inside the technology industry is broadening beyond procurement contracts for electricity already being generated. A rapidly growing computational sector cannot indefinitely treat electrical supply as though it were an external commodity that will automatically expand whenever demand appears. Once projected loads reach hundreds of megawatts at individual facilities and potentially several gigawatts across concentrated campuses, generation capacity, transmission availability, interconnection schedules and grid reliability become direct constraints on technology deployment.

Nuclear power becomes particularly interesting within this environment because its operating characteristics align unusually well with the demands of large computational facilities. U.S. nuclear plants operated at a capacity factor above 92 percent in 2024, according to Department of Energy data, meaning they produced near their maximum potential output through most hours of the year. Nuclear represented only about 8 percent of U.S. generating capacity in 2023 while producing approximately 18 percent of the nation's electricity, a difference largely explained by the consistency with which reactors operate compared with many other generating resources.

That characteristic becomes economically useful when the customer on the other side of the meter is a computational facility designed to operate continuously. Servers do not become valuable only when the wind is blowing or when electricity demand elsewhere happens to be low; expensive processors generate returns by remaining productive, and a data center filled with billions of dollars of computing equipment has an obvious incentive to maximize utilization across the day and throughout the year.

The value proposition therefore extends beyond the number of kilowatt-hours a reactor can generate. Nuclear provides large quantities of highly predictable power from a relatively compact physical footprint, with operating lives measured in decades and fuel requirements that are small relative to the amount of electricity produced. For technology companies making infrastructure decisions over long horizons, those characteristics begin to resemble features of the computing infrastructure itself rather than merely attributes of the utility supplying it.

A Data Center Is Becoming an Industrial Facility

The convergence becomes easier to understand once AI infrastructure is viewed as industrial infrastructure rather than as an unusually large version of an office technology system. A frontier-scale data center combines enormous quantities of fixed capital with continuously operating machinery, sophisticated cooling equipment, specialized electrical systems, telecommunications infrastructure and a requirement for exceptionally reliable power. Although the product leaving the facility is computational rather than physical, the underlying economics increasingly resemble those of other capital-intensive industrial operations in which utilization rates, energy costs, infrastructure reliability and long-term supply contracts have direct consequences for returns on invested capital.

This is one reason the language surrounding data centers is beginning to change. The industry increasingly speaks about “AI factories,” a term that can sound promotional until the underlying production process is examined closely. Electricity enters the facility, processors perform computation, models transform that computation into inference and reasoning, and the resulting machine intelligence becomes an input into software development, engineering, logistics, medicine, finance, scientific research, defense and countless other activities.

Within the Power → Compute → Intelligence framework, the data center is therefore not simply storing information or hosting websites. It is becoming a production facility in which energy and computational capital are combined to manufacture usable cognition. Once the facility is understood that way, the renewed technology-sector interest in nuclear power becomes less surprising. Industrial businesses have always cared deeply about securing the energy inputs required to operate their plants, and technology companies are increasingly discovering that the same logic applies when their most valuable factories happen to be filled with accelerators rather than assembly lines.

Why Existing Nuclear Plants Are Suddenly More Valuable

One of the most revealing developments is not the announcement of futuristic reactors but the reassessment of nuclear plants that already exist. The planned restart of the Crane Clean Energy Center provides a particularly striking example. The reactor ceased operations in 2019 for economic reasons, yet five years later Constellation announced a twenty-year agreement with Microsoft that created the commercial foundation for investing roughly $1.6 billion to return the plant to operation, with approximately 835 megawatts expected to return to the grid.

The underlying reactor technology did not suddenly become capable of producing a different kind of electricity. What changed was the economic environment surrounding the asset. Electricity-demand expectations strengthened, technology companies began placing greater value on large quantities of dependable generation, and a customer with an enormous prospective power requirement became willing to enter a contract long enough to support substantial capital investment.

Meta's agreement surrounding the Clinton plant illustrates the same change from another direction. Rather than restarting a closed reactor, the twenty-year agreement supports continued operation of an existing 1,121-megawatt plant beyond the expiration of an Illinois support program, while Constellation also plans an uprate adding another 30 megawatts of output.

This distinction between preserving existing nuclear capacity and constructing entirely new plants is economically important. New nuclear construction in the United States and other advanced economies has suffered from high capital costs, complicated project execution, long development timelines and, in several prominent cases, significant schedule and budget overruns. The IEA identifies scale, technical complexity, lengthy construction periods and financing risk among the principal difficulties surrounding new nuclear development, while noting that lifetime extensions for operating plants tend to be considerably easier to finance because the underlying assets and operating histories already exist.

The near-term nuclear response to AI demand may therefore come partly from extracting more value from the fleet already standing. License extensions, uprates, avoided retirements and technically feasible restarts can add or preserve firm electrical capacity without waiting for an entirely new generation of reactors to move from design through financing, licensing and construction. This creates an interesting reversal in the valuation of energy infrastructure. A nuclear plant that once appeared economically marginal in a period of stagnant electricity demand can become considerably more strategically valuable when several industries begin competing for large increments of reliable power at the same time. AI has not changed the reactor; it has changed the scarcity surrounding what the reactor produces.

Technology Companies Can Change Nuclear Finance

The emerging relationship between technology and nuclear power is especially significant because the principal historical constraint on new nuclear development has often been financial rather than scientific. Commercial nuclear reactors are well understood at the technological level, but constructing them requires enormous upfront capital that may remain committed for years before the plant begins generating revenue. Delays amplify financing costs, construction overruns increase the amount of capital at risk, and volatile wholesale electricity markets can make future cash flows difficult to predict. The IEA has consequently identified long-term power purchase agreements, contracts for difference and other mechanisms that stabilize future revenue as important tools for making nuclear projects more financeable.

This is where hyperscale technology companies can alter the economics in ways that smaller electricity customers cannot. Microsoft, Meta, Google and Amazon possess enormous balance sheets, high credit quality, rapidly growing electricity requirements and the ability to enter contracts extending across decades. A twenty-year commitment from a major technology company can provide an electricity producer with something particularly valuable in a capital-intensive industry: visibility into future demand and revenue.

Alongside increasing electricity consumption, AI expansion can introduce a new class of anchor customers whose long-term commitments reduce uncertainty for nuclear investment. This is potentially more consequential than a temporary increase in electricity prices because nuclear construction depends heavily upon confidence that expensive assets will have customers over very long operating lives. If technology companies conclude that securing power is essential to their future computing capacity, they can become participants in generation investment rather than passive purchasers in wholesale electricity markets.

The result begins to resemble vertical coordination between the technology and energy sectors. The technology company does not necessarily own and operate the reactor, just as an automobile manufacturer does not need to own the steel mill supplying it, but long-term contracts and direct investments can bind the two industries together closely enough that the expansion plans of one influence the capital formation of the other.

Small Modular Reactors Are Really a Business-Model Experiment

Much of the discussion around small modular reactors focuses understandably on reactor physics and engineering design, yet their potential significance to the technology sector may depend equally upon whether they create a different financing and deployment model. Conventional gigawatt-scale nuclear projects concentrate enormous amounts of capital into a single construction program. SMRs attempt to reduce the size of the individual investment, standardize designs and manufacture more components through repeatable industrial processes, allowing capacity to be added incrementally rather than through one enormous project. Whether those theoretical advantages translate into competitive commercial costs remains to be demonstrated at scale, and first-of-a-kind reactors will inevitably carry higher costs and execution risks than mature technologies produced through established supply chains.

The IEA nevertheless sees enough potential in the model to expect the first commercial SMRs around 2030, while estimating that investors already have plans associated with as much as 25 GW of SMR capacity primarily intended to serve data-center demand. Its nuclear outlook emphasizes that standardization, repeat construction and the development of deeper supply chains will be essential if modular reactors are to move down the cost curve rather than reproducing the project-by-project complexity that has burdened conventional nuclear construction in some advanced economies.

The emerging technology-sector agreements provide an early test of that model. Google's arrangement with Kairos Power contemplates electricity from multiple advanced reactors, with an initial unit targeted around 2030 and additional deployments through 2035; Google has described the broader agreement as potentially unlocking as much as 500 MW of new capacity. Kairos has already received Nuclear Regulatory Commission construction permits for its Hermes and Hermes 2 test reactors in Tennessee, meaning the company is progressing through actual regulatory and engineering milestones rather than remaining entirely at the conceptual stage.

Amazon's strategy reaches still further into the nuclear supply chain. Its agreements include development work with Energy Northwest that could begin with 320 MW of advanced SMR capacity in Washington and expand to 960 MW, while the company has invested more than $500 million in X-energy and has described an ambition to help advance more than 5 GW of new nuclear capacity by 2039.

It remains possible that some advanced-reactor projects will be delayed, redesigned or fail to achieve expected costs, which is precisely why the current period should be understood as a commercialization phase rather than evidence that every technical and financial problem has already been solved. The important shift is that technology companies with real power demand are willing to help fund the process through which those questions will be answered. That turns data-center demand into something approaching an early market for advanced nuclear technology.

Nuclear Is Not the Near-Term Answer to Every AI Load

The excitement surrounding the convergence of AI and nuclear power should not obscure a basic timing problem. A data center can often move from planning to operation much faster than a new nuclear plant can move through financing, licensing, supply-chain development and construction, which means that the enormous electricity-demand increases expected before 2030 cannot simply be met by announcing reactors that will not enter service until the following decade.

The IEA's energy-and-AI outlook reflects this reality. In the United States, natural gas and renewable generation are expected to supply much of the incremental electricity required by data centers through 2030, while nuclear's contribution becomes more important toward the end of the decade and especially thereafter as advanced reactors begin entering service. Globally, the IEA expects nuclear generation serving data centers to expand substantially, but its modeling does not treat nuclear as a substitute for the broader portfolio of resources needed to meet near-term growth.

This timing distinction helps separate a serious nuclear strategy from slogans. Existing reactors, uprates and some restarts can contribute sooner because much of the underlying physical and regulatory infrastructure already exists, while entirely new large reactors and first-generation SMRs belong primarily to the medium- and long-term supply response. Natural gas can often be deployed more quickly where pipeline and turbine capacity permit, while wind and solar can add generation relatively rapidly where transmission and grid conditions accommodate them.

The proper analytical question is therefore not which single technology should “win” the electricity system. An economy expecting sustained growth in computational and industrial demand needs a portfolio capable of expanding on different timelines while maintaining reliability and acceptable costs. Nuclear's significance emerges because it can eventually provide very large quantities of continuous generation within that portfolio, complementing resources whose economics or operating characteristics differ.

The Capital Cost Problem Cannot Be Wished Away

A serious analysis of nuclear power also has to acknowledge that its operating advantages do not erase the difficulties of construction. Recent large nuclear projects in advanced economies have demonstrated how damaging schedule overruns and escalating costs can become when design changes, supply-chain weaknesses, regulatory uncertainty and project-management failures accumulate. The Department of Energy's advanced nuclear commercialization work has estimated that first-of-a-kind advanced plants may require overnight capital costs around $6,200 per kilowatt under successful execution, while recent U.S. nuclear construction has exceeded $10,000 per kilowatt; the commercial thesis depends upon repeat deployment, standardization and industrial learning eventually reducing those costs substantially.

This is where the technology sector's involvement could be structurally important without magically resolving the underlying engineering problem. Deep-pocketed customers can strengthen project finance, provide predictable demand, help establish initial order books and support first-of-a-kind deployment, but capital alone cannot compensate indefinitely for poor project execution. A sustainable nuclear revival requires the industry to relearn how to build repeatedly rather than treating every reactor as an artisanal megaproject. Supply chains need sufficient order volume to invest in factories and specialized equipment, workers need repeated projects through which skills are retained and improved, designs need to stabilize before construction begins, and regulatory processes need to remain rigorous while becoming more predictable.

In that sense, the economic challenge resembles industrial manufacturing more than scientific invention. The physics works; the harder question is whether advanced economies can rebuild the institutional and industrial capacity to construct nuclear facilities at predictable cost and schedule. AI demand may supply something the industry has lacked for years: a customer base sufficiently motivated by electricity scarcity to help make that rebuilding economically worthwhile.

Electricity Reliability Is Becoming Compute Reliability

The relationship between nuclear and technology also changes how reliability should be valued. A computing facility can protect itself against short disruptions using batteries, backup generators and redundant electrical feeds, but these systems are designed primarily to bridge failures rather than substitute economically for continuous grid-scale generation over extended periods. For a data center operating around the clock, the quality of the surrounding electricity system therefore influences the amount of redundant infrastructure that must be built and the operational risk associated with the site.

Nuclear plants fit naturally into systems serving such loads because their output is highly predictable across long operating periods, interrupted principally by planned refueling and maintenance cycles. This does not mean individual reactors are incapable of outages, nor does it eliminate the need for diversified grids and redundancy; a resilient electrical system should never depend upon the assumption that any single generator is infallible. The value arises instead from the portfolio characteristics of firm generation. When large amounts of compute are added to a grid, supplying those loads through resources available across different weather conditions and hours reduces the amount of balancing, overbuilding, storage and transmission that would otherwise be required to achieve comparable reliability.

Technology companies increasingly have an economic reason to care about that distinction because electrical reliability is becoming inseparable from compute reliability. A processor costing tens of thousands of dollars creates little economic value while sitting idle, and the enormous fixed capital embedded in AI campuses rewards electrical systems capable of keeping those machines productive. The technology sector is therefore beginning to encounter an old industrial lesson: the cheapest nominal unit of energy is not necessarily the lowest-cost energy system once reliability, infrastructure and utilization of the productive asset are included.

Nuclear Plants Can Become Anchors for Computational Geography

The relationship may eventually influence where computational infrastructure develops. Existing nuclear sites possess several characteristics that are difficult to reproduce quickly: high-capacity grid connections, established transmission corridors, large generating assets, technically skilled workforces and land already associated with major electrical infrastructure. Some sites also contain room for expansion or additional generation, creating the possibility that existing nuclear locations could evolve into larger energy-and-compute clusters.

The Amazon relationship with the Susquehanna nuclear facility in Pennsylvania demonstrates why such configurations attract interest. Amazon has described its investment in a nearby data-center campus as supporting continued generation at the existing plant while also financing transmission upgrades associated with its operations, and the company has explored the potential for additional advanced nuclear generation at the site.

This model could create a new form of industrial geography in which computational facilities increasingly locate near large generation assets rather than assuming that unlimited power can always be transported to whatever location developers select. The economics will vary according to transmission rules, market design, regulatory treatment and local infrastructure, but the underlying physical logic is straightforward: when electricity becomes difficult to deliver to the computer, there is an incentive to move the computer closer to the electricity.

That possibility should interest regions that already host substantial nuclear infrastructure. An operating reactor is not simply producing electricity for the existing economy; it may represent a platform around which future computational, manufacturing or research investment can develop, particularly if the surrounding grid retains additional capacity and local institutions understand how to translate the energy asset into a broader economic-development strategy.

The Reactor Is Moving Up the Technology Stack

The growing connection between nuclear power and AI reflects the expanding physical requirements of machine intelligence. The production system begins with primary energy resources and generation, proceeds through transmission and electrical equipment into data centers and semiconductors, and continues through models and software to the businesses and institutions applying their output. A reactor belongs within that system alongside gas turbines, transmission lines, transformers and cooling equipment, each contributing to the infrastructure that makes computational intelligence possible.

Nuclear power stands out within that architecture because it can produce unusually large amounts of continuous electricity for unusually long periods from geographically compact facilities. An operating reactor can remain a productive asset for many decades, which aligns with an intelligence economy whose aggregate appetite for electricity appears more likely to compound than disappear.

The International Energy Agency now explicitly describes data centers and AI among the forces driving renewed interest in nuclear power, while projecting nuclear generation reaching new records and identifying more than 70 GW of reactor capacity already under construction globally. At the same time, the agency continues to emphasize the financing, construction, supply-chain and fuel-cycle challenges that must be overcome if current interest is to translate into sustained deployment. Those two realities belong together. Nuclear power possesses characteristics that increasingly fit the requirements of the intelligence economy, while the nuclear industry still has to prove that new projects can be delivered reliably enough for those characteristics to matter commercially.

AI Could Help Rebuild the Nuclear Industrial Base

The relationship between the two industries may eventually become recursive. AI creates electricity demand that improves the commercial case for nuclear investment, while AI itself can be used throughout nuclear engineering, operations, materials analysis, predictive maintenance, regulatory documentation, construction planning and scientific research. Greater computational capability can therefore help improve the very energy systems required to supply future computational capability.

More importantly, sustained demand from technology companies can help create the order volume necessary to rebuild manufacturing capacity around nuclear construction. Reactor suppliers cannot develop efficient production systems around sporadic projects separated by many years because suppliers, specialized workers and institutional knowledge dissipate between builds. A repeatable pipeline of projects changes that economic environment by allowing companies to invest against anticipated future orders rather than treating each plant as an isolated undertaking.

This is one reason the involvement of hyperscalers deserves attention even if their individual nuclear investments remain small compared with the total electricity system. They represent sophisticated private customers willing to commit capital to technologies that may not otherwise receive sufficient early orders to move from demonstration into repeat deployment. The technology industry could therefore become one of the demand anchors around which a renewed nuclear industrial base develops, particularly if utilities, manufacturers and other electricity-intensive sectors participate alongside it. That would make the AI-energy relationship much more consequential than the temporary question of how to keep today's data centers powered.

From Energy Procurement to Strategic Infrastructure

The first generation of corporate renewable-energy procurement allowed technology companies to purchase electricity and environmental attributes from projects developed largely within the existing structure of the power industry. The emerging nuclear relationships go further because technology firms are increasingly helping preserve existing plants, restart retired capacity, finance advanced reactor developers and support future generating projects whose commercial viability depends partly upon those companies becoming long-term customers. The distinction reflects the scale of the underlying demand. Once electrical availability begins constraining how much computational infrastructure a company can deploy, energy procurement stops being primarily a facilities-management function and becomes part of corporate capacity planning.

A technology company expecting to spend tens or hundreds of billions of dollars on data centers and computing hardware cannot treat the electricity required to operate those assets as an afterthought. Securing long-duration energy supply becomes analogous to securing semiconductor capacity, networking equipment or land: it is part of assuring that the rest of the capital program can actually function. Nuclear power consequently enters technology strategy through the back door. It does not become relevant because software companies have decided they want to become utilities, but because AI has increased the physical scale of computing enough that generation capacity itself affects how rapidly computing capacity can expand.