Powering the Digital Future: Navigating the Nuclear Option for Data Centers
Feb 26 2025
This is the fourth article in Womble Bond Dickinson’s Energy & Natural Resources thought leadership series titled “Going Nuclear: A Sustainable Solution to Rising Energy Demand.” The series will explore the changing landscape of nuclear power, both domestically and internationally, and provide stakeholders with the strategic insights they need to navigate this expanding field.
Modern data centers are the foundation of our information society and now use artificial intelligence to generate new forms of machine intelligence and learning – though at the cost of considerable energy consumption. Their energy demand outstrips the ability of existing generation and transmission systems to meet that demand, making nuclear energy particularly well-suited to supply the shortfall given its base load (round-the-clock) generation profile, low fuel cost and insulation from carbon emissions concerns. Powering data centers with purpose-built, reactivated, or newly-completed nuclear generation is an attractive way to accelerate power supply to meet the needs of the AI economy.
The path for powering data centers with nuclear energy depends on multiple factors including whether the power will come from:
More than 65 individual balancing authorities manage the United States electric grid, and most are either regional transmission organizations or large investor-owned or publicly-owned power utilities. Many smaller utilities, electric cooperatives, and municipal power systems operate behind those larger balancing authorities.
The commercial and operational terms by which new nuclear capacity will be integrated onto the grid will depend on the laws, regulations and tariffs that apply to each of these entities. This is true even for on-site or ‘behind the meter’ nuclear generation, since connecting data centers to the grid is typically required for emergency or standby generation, moment to moment load balancing, and the export of excess power when power consumption at the data center drops. The costs of operating without a robust grid connection can be quite expensive considering the cost of building and maintaining high-capacity battery or gas-fired generation back up.
For a more detailed description of the US grid issues that apply to nuclear generation generally, please see the earlier installment of our “Going Nuclear” series, Powering Progress: Navigating the Intricacies of On-Site Nuclear Generation.
Data center developers are increasingly viewing SMRs as an attractive alternative to traditional large-scale nuclear reactors for powering their facilities considering their automatic or ‘walk-away’ safety features, their scalability (300 MW or less per unit vs. approximately 1,000 MW per full-scale unit), and the ability of SMR reactor units to be manufactured in factories and transported fully assembled to their final location for installation.
But substantively, the costs and benefits of the two technologies are closely balanced since, like SMRs, today’s full-scale reactors have comparable walk-away safety features and key components can be built in a series of modules on factory floors. Both SMRs and full-sized units require significant on-site ‘stick built’ construction for balance-of-plant equipment (including steam turbines, condensers, water cooling systems, switchyards, and control, maintenance and administration facilities) as well as site-specific NRC licensing and environmental permitting. Neither represents a true plug-and-play solution. In addition, full-scale units have significant economies of scale in the form of lower per-unit staffing and operating cost and produce less high-level waste for future disposal. Outside of the United States, there have been a number of successful recent projects to build new, full-scale reactors.
In short, SMRs represent nuclear capacity that data center developers can install in smaller increments reducing financial risk (capital costs) and time to start up, while creating the redundancy inherent in multiple units that can produce energy independently of each other. Full-scale reactors have significant operating economies, but in a single generating package.
SMRs represent nuclear capacity that data center developers can install in smaller increments reducing financial risk (capital costs) and time to start up, while creating the redundancy inherent in multiple units that can produce energy independently of each other. Full-scale reactors have significant operating economies, but in a single generating package.
Locating nuclear capacity on data center sites can create significant cost and time saving if major transmission upgrades can be avoided. But the savings may evaporate if for operational requirements grid connectivity must still match the nuclear facility's total output or the data center's peak power needs. These requirements include maintaining reliability standards, managing excess power, balancing loads, or meeting the data center's full power demands. Where robust standby transmission access is required, the same transmission-related regulatory and construction issues will arise with on-site generation as with generation located elsewhere on the grid. In those cases, the primary operating advantage of on-site nuclear generation, or other onsite generation, may be accelerating availability of capacity or insulation from the effects of curtailments of service or loss-of-load (“LOA”) events on the grid.
As we will discuss with more depth in an upcoming installment of our “Going Nuclear” series, completing or restarting existing but non-operating units is currently being considered for multiple plants including the Palisades Unit, the Bellefonte Units, Three Mile Island Unit 2, and V. C. Summer Units 2 and 3. Completions and restarts leverage investments already made) and transmission interconnectivity already in place. Assuming land use patterns and the constraints of nuclear exclusion zones would support doing so, building new data centers alongside such completions and restarts can be a powerful strategy for delivering new capacity quickly.
State statutes and regulations generally allow customers to build and operate their own generation. But where that generation is connected to the grid (i.e., behind a utility meter) it may fall under the provisions of state distribution energy resources (DER) legislation that typically were drafted for smaller solar and renewable projects and prohibit large behind the meter installation. These caps, however, are not the last word, and can be removed or waived especially if the incumbent utility agrees.
Some jurisdictions may prohibit interconnection behind the meter facilities that do not sell their capacity and energy to the grid, and where the facility will be located inside an RTO or ISO, FERC may have jurisdiction over interconnection. FERC recently rejected an agreement to power an Amazon data center in Pennsylvania through a direct connection with an adjacent operating nuclear station based on the potential impact taking existing nuclear generation out of PJM’s constrained capacity markets.
FERC is also considering a request by Constellation Energy to require PJM to adopt tariff provisions to support co-located or directly connected nuclear and other generation while addressing concerns about effects on reliability and rate payer costs.
If a data center plans to purchase power from an offsite nuclear unit, a power purchase agreement (PPA) with the owner and operator of the unit will determine the terms of sale. If the unit will operate in a competitive retail market, then the PPA delivery will take place under the open access transmission tariff (OATT) of the resident transmission operator, and retail power delivery tariffs of the local distribution entity. However, the structure of most deregulated markets involves all generation being sold into a single market, with contracts for differences giving end-users the economic benefit of their PPA transactions. A data center customer will want some assurance that service will not be curtailed so long as the nuclear capacity it is providing is online and supplying power to the grid. The terms of existing tariffs should not be considered to be the last word on what is possible. It may be possible to negotiate and obtain regulatory approval for contractual terms or special tariff provisions tailored to the specific transaction.
A data center customer will want some assurance that service will not be curtailed so long as the nuclear capacity it is providing is online and supplying power to the grid.
Most states require electric generation developers to obtain some form of CPCN to construct systems sized at 75-85 MW or more, and nuclear construction would almost certainly require certification under those statutes. These requirements often apply whether or not the new unit is considered self-generation, i.e., it is owned by and serves only the data center owner. These statutes were not typically drafted with single customer, large-scale generation in mind, and so adapting a new nuclear project under their terms may require some creativity.
Vertically Integrated Markets: If the state follows a vertically integrated utility service model (i.e., non-deregulated), then the local utility will likely have territorial service rights which extend to generation construction. This may allow the utility to block the construction of new generation to serve a customer within its service territory, especially if it is to be owned by an entity other than the data center and its customers. However, there can be exceptions. Some states have statutes or tariffs that allow industrial choice, distribution energy resources (DER), or voluntary renewable energy projects (“VREPs”). Otherwise, regulatory support from the incumbent utility and a one-off agreement may be needed with to site new nuclear generation. Further , the incumbent utility’s public service commission will need to approve, and such agreement would be a contractual exception to the utility’s generally applicable tariffs.
A data center that is connected to the grid for backup power purposes will be a retail customer of the incumbent electric utility. The upside of being a retail tariff customer is that the data center can use its grid connection to buy standby power to deal with fluctuations in its energy demand (and to sell excess power onto the grid when necessary). But the presence of a retail meter will make the data center subject to the costs built into its retail tariff. The tariff may be out of alignment with the standby nature of the service being purchased and may include services that do not benefit the data center owner (e.g., cost for renewable portfolio standards, demand side management (DSM) programs, and other social or environmental costs). Depending on the tariffs, the data center may be subject to curtailment in times of system emergency even if the nuclear plant is producing sufficient power to meet its demands.
Nuclear power presents a compelling solution for meeting the exponentially growing energy demands of modern data centers, particularly those supporting AI operations. However, successful implementation requires careful navigation of multiple regulatory and licensing complexities.
Whether choosing SMRs or full-scale reactors, data center operators must carefully evaluate their specific needs against various factors: initial capital costs, operational economies, regulatory requirements and uncertainties, and grid integration challenges. The decision between on-site and off-site generation, or whether to participate in recommissioning existing facilities, requires thorough analysis of federal, regional, and local regulations, transmission infrastructure, and operational requirements.