In a series of landmark agreements that have reshaped the global energy landscape, the world’s largest technology conglomerates have pivotally shifted their strategy to secure the massive amounts of electricity required to fuel the artificial intelligence (AI) boom. Within a single week, two of the three dominant cloud computing providers—Amazon and Google—solidified 20-year contracts for nuclear power, signaling a historic reversal in the fortunes of an industry that had been in decline for decades. Amazon has secured 690 megawatts from Constellation Energy’s Calvert Cliffs plant in Maryland, a deal that includes 190 megawatts of additional generating capacity slated to come online between 2030 and 2032. Simultaneously, Google has entered an agreement designed to bring 890 megawatts of new nuclear capacity into the United States grid. These moves follow Microsoft’s high-profile deal to restart a shuttered reactor at the infamous Three Mile Island site, effectively turning the tech sector into the primary financier of a nuclear revival.
The scale of this shift is underscored by the financial commitments involved. Constellation Energy, the largest operator of nuclear plants in the U.S., is investing more than $4.3 billion to upgrade 11 reactors across Illinois, Pennsylvania, and New Jersey. Notably, these agreements do not currently focus on the construction of new, large-scale traditional reactors. Instead, the tech giants are paying a premium to "uprate" or restart existing facilities—some of which were built before the founders of these tech companies were born. This strategy reflects an urgent need for "firm" power—electricity that is available 24/7, unlike intermittent solar and wind—as data centers evolve from simple storage hubs into energy-intensive AI processing factories.
From Skepticism to Necessity: The Historical Context of Nuclear Energy
To understand the magnitude of these deals, one must look at the historical trajectory of nuclear power. For over forty years, the industry was defined by stagnation and public apprehension. Following the partial meltdown at Three Mile Island in 1979, the catastrophic explosion at Chornobyl in 1986, and the 2011 Fukushima Daiichi disaster in Japan, nuclear energy became a political third rail in many democratic nations. Germany famously committed to a total phase-out of its nuclear fleet, a process completed in 2023, while Japan idled its reactors for years before cautiously beginning restarts to meet climate goals.
In the United States, the industry faced an existential crisis as aging plants struggled to compete with the shale gas revolution, which drove down the price of natural gas. Between 2012 and 2022, a dozen U.S. nuclear reactors were retired prematurely due to economic pressures. However, the narrative shifted abruptly with the emergence of generative AI. The training of large language models (LLMs) requires thousands of specialized chips running at high capacity for months, while "inference"—the process of an AI answering a user query—consumes significantly more power than a standard Google search. This sudden surge in demand coincided with aggressive corporate and national carbon-neutrality targets, leaving nuclear as the only carbon-free energy source capable of operating at the necessary scale and reliability.
The Efficiency Paradox: Why AI Demand Refuses to Shrink
In the early stages of the AI revolution, industry leaders argued that technological innovation would solve the energy problem. The focus was on efficiency: developing more powerful chips that consumed less energy per calculation, optimizing cooling systems, and creating smaller, more specialized AI models. These advancements were significant; modern AI chips are vastly more efficient than their predecessors on a per-watt basis.
However, the industry has encountered a phenomenon known as the Jevons Paradox. Named after 19th-century economist William Stanley Jevons, the paradox states that as the efficiency with which a resource is used increases, the total consumption of that resource often rises rather than falls. In 1865, Jevons observed that more efficient steam engines led to increased coal consumption because cheaper power made steam engines viable for a wider range of industrial applications.
A contemporary example of this occurred when the AI startup DeepSeek demonstrated that a competitive model could be trained at a fraction of the cost previously estimated by industry giants. Rather than using these savings to reduce overall spending or energy use, the industry doubled down. Laboratories and cloud providers redirected those resources toward even larger training runs, models that reason for longer periods, and autonomous "agents" that operate continuously in the background. According to the International Energy Agency (IEA), global data center electricity demand is projected to more than double to approximately 945 terawatt-hours by 2030. This growth is equivalent to adding the entire electricity consumption of a country like Germany to the global grid in just six years.
A Chronology of the Nuclear Resurgence
The current wave of nuclear agreements is the culmination of a multi-year shift in policy and corporate strategy:
- 2022: The Palisades nuclear plant in Michigan is closed, seemingly marking the end of its operational life. However, the Biden-Harris administration’s Inflation Reduction Act begins providing federal support for carbon-free energy.
- March 2024: The U.S. Department of Energy (DOE) announces a conditional loan of up to $1.52 billion to restart the Palisades plant. This marks the first time a decommissioned commercial reactor in the U.S. has been slated for a restart.
- September 2024: Microsoft announces a 20-year power purchase agreement (PPA) with Constellation Energy to restart Unit 1 of the Three Mile Island plant (now renamed the Crane Clean Energy Center).
- October 2024: Amazon and Google announce their respective deals with Constellation and various partners to secure nearly 1.6 gigawatts of nuclear capacity, focusing on both existing plant upgrades and future Small Modular Reactors (SMRs).
- Present: The DOE is now providing a total of $4.4 billion in loans to support the restart of three shuttered plants, aiming to restore 2.25 gigawatts of carbon-free power to the grid.
The Economic Burden on Public Utilities and Consumers
While the tech industry frames these deals as a win for the climate, the financial implications for the average consumer are becoming a point of intense political friction. Most of the recent nuclear deals are concentrated within the PJM Interconnection, the largest regional transmission organization in the U.S., serving 67 million people across 13 states and the District of Columbia.
PJM operates "capacity auctions" to ensure there is enough electricity generation to meet future demand. Because data centers require massive, guaranteed amounts of power, they drive up the prices in these auctions. According to the grid operator’s independent market monitor, data center demand contributed approximately $29.4 billion in capacity costs across the four most recent auctions. In the latest auction alone, roughly $6.3 billion of the costs were attributed to data center demand, including forecasts for facilities that have yet to be built.
These costs are passed directly to households. The Natural Resources Defense Council (NRDC) estimates that by 2028, the typical household within the PJM territory could see their electricity bills increase by $70 per month due to these capacity charges. This has sparked a populist backlash. In Virginia—home to "Data Center Alley"—the state has established a separate electricity rate class for data centers to shield residential consumers. Ohio and Illinois are following suit with minimum billing requirements and legislative debates aimed at forcing tech companies to pay a larger share of the infrastructure costs.
The Small Modular Reactor (SMR) Gamble
Recognizing that the supply of existing reactors is finite, tech companies are also betting on the future of Small Modular Reactors (SMRs). Unlike traditional reactors, which are massive, bespoke engineering projects, SMRs are designed to be factory-built and transported to sites, potentially reducing costs and construction timelines.
Google’s agreement includes a focus on SMR technology, and Amazon has invested in X-energy, a leading SMR developer. However, the timeline for SMRs remains a significant hurdle. Most experts do not expect these reactors to deliver electricity to the grid until the early-to-mid 2030s. This creates a "timing mismatch": AI hardware cycles move in 18-to-24-month increments, while nuclear energy cycles move in decades. There is a looming risk that the chips being installed today will be obsolete long before the 20-year nuclear contracts intended to power them have even reached their midpoint.
Strategic Implications for the Gulf Region
The U.S. experience serves as a critical case study for Gulf countries—including Qatar, the United Arab Emirates, and Saudi Arabia—as they position themselves as the next global hubs for AI infrastructure. These nations possess a distinct competitive advantage: abundant energy resources and sovereign wealth capable of funding massive capital expenditures.
The scale of ambition in the region is unprecedented. Qatar’s sovereign wealth fund has entered a $20 billion joint venture with Brookfield for AI infrastructure, starting with 300 megawatts and targeting 1.5 gigawatts. Abu Dhabi is pursuing up to five gigawatts of capacity, while Saudi Arabia’s HUMAIN project targets nearly two gigawatts.
However, the "opportunity cost" in the Gulf is high. If Qatar completes its 1.5-gigawatt build-out, it would represent roughly 12 percent of the nation’s total installed generating capacity. Electricity diverted to AI cannot be used for desalination, industrial cooling, or other domestic needs. The Gulf nations have a unique window to avoid the mistakes made in the U.S. market by setting clear terms early. This includes establishing industrial tariffs that reflect the true cost of infrastructure, determining who funds new generation, and ensuring that the economic benefits of hosting data centers extend beyond energy revenue to high-value job creation and domestic AI sovereignty.
Analysis: The Future of Voluntary Commitments
The current trend of Big Tech "voluntarily" bringing new power to the grid—as seen in the Google agreement—is a constructive attempt to mitigate public and regulatory backlash. By funding additional capacity rather than just consuming existing supply, these companies hope to avoid being blamed for rising utility bills.
However, the sustainability of this voluntary model is in question. In October 2026, U.S. federal regulators rejected a proposal from PJM that would have legally required large new consumers to provide their own generation capacity. This leaves the industry in a delicate position: they are offering commitments that regulators have not yet mandated. If the cost of nuclear upgrades and SMR development continues to climb, the financial pressure on tech companies may test their commitment to these 20-year deals.
Ultimately, the marriage of Big Tech and nuclear power represents a high-stakes bet on the permanence of the AI revolution. If AI demand continues its exponential growth, these nuclear contracts will be seen as visionary moves that secured the "green" backbone of the future economy. If demand falters, the tech industry may find itself tethered to the most expensive and long-term energy obligations in corporate history, while the public continues to grapple with the infrastructure costs left in the wake of the AI gold rush.
