Restarting Retired Nuclear Plants for AI Workloads: Powering the Next Computing Revolution

Neil L. Rideout

10/7/20266 min read

Restarting Retired Nuclear Plants for AI Workloads: Powering the Next Computing Revolution

The artificial intelligence boom is rewriting the rules of energy demand. Training large language models, running inference at scale, and powering the data centers that support them require electricity on a scale that few industries have ever demanded so suddenly. Hyperscalers and AI labs are racing to secure reliable, carbon-free power measured in gigawatts rather than megawatts. In this high-stakes environment, an unexpected solution has moved from the margins of energy policy into serious boardroom discussions: restarting retired nuclear power plants specifically to feed AI workloads.

Nuclear energy already supplies roughly 20% of U.S. electricity and about 10% of global electricity, delivering firm, around-the-clock power with near-zero operational carbon emissions. Many of the plants that were shut down in the last two decades still possess intact reactor vessels, containment buildings, cooling systems, and transmission connections. Restarting them is not the same as building new reactors from scratch. In favorable cases it can be faster, less capital-intensive, and more politically tractable than greenfield nuclear construction. For AI companies that need power in three to five years rather than ten to fifteen, that difference is decisive.

Why AI Needs Nuclear-Scale Power

Modern AI training clusters can draw hundreds of megawatts continuously. A single large training run for a frontier model can consume as much electricity as a small city for weeks. Inference demand is growing even faster as models move into everyday products. Data-center operators project that AI-related electricity consumption in the United States alone could rise by tens of gigawatts this decade. Renewable energy and battery storage will supply a large share of that growth, yet they cannot alone guarantee the 24/7 firm power that high-utilization GPU clusters require without massive overbuild and curtailment.

Nuclear plants excel at providing exactly that firm baseload. Once online, a reactor operates at high capacity factors—often above 90%—for 18- to 24-month cycles. This matches the continuous, high-load-factor profile of AI training and inference far better than intermittent sources. In addition, nuclear’s low marginal operating cost and long-term fuel security make multi-decade power-purchase agreements attractive to companies that plan to operate AI infrastructure for decades.

The Case for Restarting Rather Than Building New

Building a new large light-water reactor in the West has proven slow and expensive. First-of-a-kind projects have faced multi-year delays and cost overruns measured in billions. Restarting an existing plant sidesteps many of those risks. The site is already licensed, the transmission infrastructure exists, the local workforce and supply chain relationships remain in many cases, and the environmental footprint of the existing facility is already established. Regulatory pathways for restart, while still rigorous, are generally shorter than those for new construction.

Several concrete examples illustrate the opportunity. In the United States, the Three Mile Island Unit 1 plant in Pennsylvania was shut down in 2019. Microsoft has signed an agreement with Constellation Energy to restart the unit and dedicate its output to Microsoft’s data-center and AI needs, targeting a return to service around 2028. Similar conversations have occurred around other recently closed units. Abroad, countries with mothballed or early-retired reactors are also examining restart options when AI and data-center demand makes the economics more favorable.

Not every retired plant is a candidate. Reactors that were permanently defueled years ago, suffered severe degradation, or sit in locations with no remaining transmission capacity may never restart economically. Plants that closed primarily for economic reasons during periods of low natural-gas prices, however, often retained significant residual value. When power prices rise and a creditworthy offtaker such as a major technology company appears, the restart calculus changes dramatically.

Technical and Regulatory Realities

Restarting a nuclear plant is a substantial engineering and regulatory project, not a simple switch-flipping exercise. Key steps typically include:

Comprehensive condition assessment of the reactor vessel, steam generators, turbines, generators, and safety systems.
Replacement or refurbishment of components that have reached end of life or no longer meet current standards.
Updated probabilistic risk assessments and safety analyses reflecting current regulatory expectations.
Fresh fuel procurement and loading.
Retraining or recruiting licensed operators and support staff.
Negotiation of new power-purchase agreements and, in many jurisdictions, resolution of any remaining decommissioning-fund or liability issues.

Regulators treat restarts with appropriate caution. In the United States the Nuclear Regulatory Commission has processes for license renewal and restart that can take several years. Public engagement remains essential; communities that lived with a plant for decades may support restart for jobs and tax base, yet others retain concerns about safety or waste. Transparent communication and demonstrated safety performance are therefore non-negotiable.

From a technical standpoint, pairing restarted nuclear capacity with AI loads offers additional synergies. Data centers can often locate near existing transmission corridors, reducing the need for new high-voltage lines. Some operators are exploring direct or nearly direct supply arrangements that minimize grid congestion and transmission losses. In principle, excess heat from the nuclear plant could even support district heating or industrial processes colocated with the data center, though such combined heat-and-power arrangements remain early-stage for most AI campuses.

Economic and Strategic Advantages

The economics of nuclear restarts for AI hinge on three factors: capital cost of the restart, the price the AI offtaker is willing to pay for firm clean power, and the avoided cost of alternative firm capacity. When natural-gas prices are moderate and renewable-plus-storage portfolios can be built quickly, pure merchant restarts struggle. When a hyperscaler is prepared to sign a 20-year contract at a premium for carbon-free, always-available power, the numbers often work.

Strategically, restarted nuclear capacity strengthens energy security and technological sovereignty. Nations that can supply their AI industry with domestic, low-carbon firm power reduce exposure to volatile fuel markets and to the geopolitical risks associated with concentrated renewable-supply chains. For the United States and Europe, keeping skilled nuclear workforces employed and supply chains warm also preserves the industrial base needed for eventual new-build programs and for advanced reactors.

Climate policy reinforces the case. Every megawatt-hour generated by a restarted nuclear plant displaces generation that would otherwise come from fossil sources or from less reliable combinations of renewables and gas peakers. Lifecycle emissions of nuclear power remain among the lowest of any firm source. For technology companies that have committed to 24/7 carbon-free energy matching, nuclear is one of the few technologies that can fill the residual firming gap at scale.

Challenges and Legitimate Concerns

Optimism must be tempered by realism. Restart projects can still encounter unexpected material degradation, supply-chain bottlenecks for specialized components, and labor shortages for licensed operators. Public acceptance is not automatic; some communities associate nuclear plants with past incidents or with long-term waste storage. Waste management remains an unresolved national policy issue in many countries, and restarts do not eliminate the need for permanent disposal solutions.

There is also a risk of over-concentration. If too many AI workloads become dependent on a handful of restarted plants, outages—whether planned or unplanned—could cascade into compute shortages. Diversification across multiple sites, technologies, and geographies remains prudent. Finally, focusing exclusively on restarts should not crowd out parallel investment in advanced reactors, small modular reactors, and fusion research, which offer longer-term pathways to even safer and more flexible nuclear power.

Policy Levers That Matter

Governments can accelerate or impede this opportunity. Clear, predictable regulatory timelines for restart applications reduce uncertainty and financing costs. Tax credits or loan guarantees targeted at nuclear restarts—similar to those already available for new clean energy—can improve project economics. Workforce development programs that retrain operators and technicians help close the skills gap. Streamlined environmental reviews that recognize the existing footprint of a plant can shave years off schedules without compromising safety.

International cooperation also has a role. Sharing best practices on restart licensing, component requalification, and operator training can lower costs for countries with smaller nuclear fleets. Export credit agencies and development banks can support restarts in emerging markets where AI infrastructure is expanding rapidly and clean firm power is scarce.

Looking Ahead

The convergence of AI’s voracious appetite for electricity and nuclear power’s unique attributes is not a temporary curiosity. It is a structural feature of the energy transition in the age of large-scale machine learning. Restarting suitable retired plants will not solve every power constraint, nor will it replace the need for massive renewable deployment, grid modernization, and eventual new nuclear construction. It can, however, deliver several gigawatts of firm, low-carbon capacity on a timescale that matches AI’s growth curve.

For policymakers, the choice is whether to treat these plants as stranded assets or as strategic reserves of clean power. For technology companies, the choice is whether to compete for scarce firm capacity on the open market or to secure dedicated nuclear resources that de-risk their long-term compute roadmaps. For communities that once hosted these plants, the choice is whether the economic and environmental benefits of restart outweigh residual concerns.

The early deals already signed suggest that many actors are answering in favor of restart. If those projects succeed on schedule and within budget, they will demonstrate a practical model that other regions can adapt. In doing so they will help ensure that the intelligence revolution does not stall for lack of electrons, and that the electrons themselves arrive with a carbon intensity low enough to keep climate goals within reach.

Nuclear power was once dismissed by parts of the technology sector as yesterday’s energy. The demands of artificial intelligence are forcing a reevaluation. Restarting the plants we already built may prove one of the fastest ways to keep the future powered.

Contact

Head Office

Green Life Enterprises LLC
8 Biltmore Est
Suite 310,
Phoenix, AZ 85016

greenlifedatacenters@gmail.com

+1-813-220-0001

Canadian Office

Green Life Enterprises LLC
3142 Nicholson Ave
Suite 10
New Waterford, Nova Scotia B1H 1N8