|
|
| Fig. 1: Projected electricity consumption from data centers (left axis), and projected SMR deployment to meet this demand (right axis). I assume that the earliest possible deployments of VOYGR-12 SMRs would begin in 2030, however the rapid increase in the number of SMRs deployed will not be possible at that rate, and SMR deployment will likely grow exponentially, using eqs. (1) and (2) to make this projection. [24] This figure solely serves to illustrate the number of SMRs that would need to be deployed to entirely satisfy this future power demand from data C enters, not their deployment rate. (Images source: L. Garcia, following Shehabi et al.. [15]) |
Small Modular Reactors (SMRs) is a catch-all term for reactors that generate less than a gigawatt of power, and are generally capable of powering anything from a car to a small town. [1] SMRs have been in use by the U.S. Military since the 1960s powering aircraft carriers, submarines, and military bases. [2] SMRs have several advantages over traditional nuclear facilities:
Lower Upfront Cost: SMRs are smaller, requiring less up-front capital than traditional nuclear plants which often face dramatic cost overruns and delays resulting from their scale. [3,4]
Scalability: SMRs and the components to fabricate them can theoretically be standardized, produced at scale in factories, and quickly transported to the site of deployment. [5] They can be also be deployed at smaller, more remote sites with smaller grids that can't accommodate a larger reactor.
Co-Location with large energy users: SMRs are small enough to be co-located with large energy consumers like data centers, and the process heat from SMRs can be used by heavy industry and hydrogen producers. [6-8]
In spite of all of these clear advantages, the U.S. has yet to see an SMR deployed for commercial use. The most recent promising SMR deployment attempt, led by NuScale Power, failed in 2023 because it could not attract enough utility customers. [9]
Even before the rapid proliferation of AI data centers, the U.S. was not on track to meet its projected energy demand by 2040. Now, there is even more demand for reliable, baseload power that nuclear energy can provide and renewables like solar and wind cannot due to their intermittency issues (data centers need to run at night, and when the wind dies down) and lack of grid-scale storage that could solve it. The United States has exploited essentially all of its hydropower resources, which means that in order to meet exponentially increasing demand for power while maintaining their commitments to carbon neutrality, nuclear power presents the only realistic source of energy to meet it. Some AI hyperscalers like Microsoft have opted to bring traditional nuclear plants back online, and others like xAI have constructed data centers powered by temporary natural gas turbines triggering community blowback, lawsuits and an ultimatum from the EPA to remove these turbines or apply for permanent permits. [10-12]
Its useful to quantify how large this shortfall is, and how many SMRs would be needed in order to meet the projected growth of data centers. Since NuScale was the first commercially approved SMR design, and the VOYGR-12 reactors appear to be most likely to be first to deployment, I will assume that all of these new data centers would hypothetically be powered by this specific power station design. I assume it runs at 95% capacity 24 hours a day, yielding a capacity factor of 0.95, and I take Eplant from the NRC. [13,14]
|
(1) |
or 4.99 TWh per year. Then in order to get the high and low estimates, I take
|
(2) |
where E(t) is the projected annual electricity consumption of U.S. data centers in year t (in TWh/yr) from Shehabi et al., and ΔE(t) = E(t) − E(2023). [15]
I use the LBNL projection for data center driven growth in energy demand, projecting their figure 10 years past their projection assuming that current growth trends hold to create Fig. 1. This figure illustrates that there rapidly increasing demand for energy in one of the highest-value potential use cases for SMRs, which represents the largest viable market for this technology.
All of this begs the question: if there is a clear market for SMRs, why have none been deployed yet for commercial use?
The U.S. Department of Energy specifies that all nuclear waste must be retrievable under 10 CFR Part 63 at any time. [16] This presents a major roadblock to advocates for borehole disposal, which involves inserting High-Level Waste (HLW) several kilometers into Earths crust and then sealing it off. [17] Deep Borehole Disposal has been pitched as a method to solve the waste problem for all nuclear plants, and has been cited by SMR advocates as a promising method to dispose of this waste. However, a borehole is explicitly designed to be for all intensive purposes non-retrievable, which clashes with the U.S. regulatory philosophy regarding nuclear waste unless the Department of Energy and the Nuclear Regulatory Commission dramatically change their regulatory paradigm.
Cost overruns in traditional nuclear energy projects have been commonplace, which has remained true in recent years. The third and fourth reactors at Plant Vogtle in Georgia were $16 billion over their initial $14 billion projected cost, and forced their original contractor to declare bankruptcy. [18] Cost overruns for a future SMR plant would likely be an order of magnitude smaller than traditional nuclear facilities, but investors, utilities, and the government will need a concrete solution to manage risk when deploying these first of a kind facilities. The DOE has several suggestions to potential SMR/Advanced Reactor Deployers to mitigate financing issues in their Advanced Nuclear Liftoff Report. [19] They argue that cost overrun can be divided among the parties best able to mitigate and bear each risk, with the construction risk being reserved for the plant owner. This looks like consortia of utilities and large customers committing to taking on between 5-10 modular standardized reactors so cost overruns are shared across the consortium and a reliable supply chain can take form.
Any U.S. SMR boom will have to overcome an acute shortage of workers in many relevant industries to SMR buildout. A single large nuclear build already takes several thousand workers [20] we will need 375,000 additional workers to deploy and operate 200 GW of new nuclear capacity by 2050, with 275k in construction/manufacturing and 100k in nuclear plant operations. [19] Its much easier to raise large sums of money for a nuclear plant than it is to develop a pipeline for manufacturing nuclear-qualified journeymen at scale, and this labor scarcity shows up, and will continue to show up as schedule risk and cost overruns in essentially every nuclear project in the United States for the foreseeable future.
Assuming the shortage of skilled labor to fabricate parts for nuclear facilities is eventually reversed given relatively high salaries in this field, the parts they produce will have to go through multiple layers of approval, potentially halting SMR deployment at scale. In the U.S., nuclear safety work is monitored by 10 CFR Part 50, Appendix B quality assurance criteria (procurement & documentation control, inspections, record retention, among other things are described and strictly enforced). [21] The Nuclear Industry works with The American Society of Mechanical Engineers (ASME) to develop a unified set of quality assurance requirements such as ASME NQA-1 and also follow NRC guidance like Regulatory Guide 1.28 Rev. 6, which formalizes the NQA-1 requirements. [22,23] These quality assurance guidelines create hard (but necessary) constraints that slow the fabrication process and increase overhead like nuclear code stamps on each part, traceability, audit-ready paperwork for all parts at all times. One example of how the current quality assurance structures will create bottlenecks is the ASME N-stamp certification process, which can take over a year and can cost hundreds of thousands of dollars, and the DOE projects that with todays N-stamp-certified component base the U.S. supply chain is strained to add about 3 GW/year of new nuclear buildout, which is far below even the lowest growth projection for electricity consumed by datacenters in Fig. 1. [19] The constraint isnt fabricating parts, it's producing nuclear-grade parts with documentation that survives an NRC audit at scale.
 |
| Fig. 2: Global breakdown of uranium refining and conversion capacity. [32,33] (Image Source: L. Garcia.) |
The United States already lacks the capacity to process raw uranium and enrich it to the levels necessary to power existing nuclear plants. There are two steps to this process, conversion (to UF_6) and enrichment (refining UF_6 to increase U-235 content), and the US heavily depends on Russia for this supply chain. The DOE has explicitly documented how low the U.S.s enrichment capacity as of its 2017 Secretarial Determination, the U.S. only has one operating commercial enrichment facility (Urenco USA in New Mexico). [24] This is far below the current demand for nuclear fuel as U.S. enrichment capability is about 4.4 million SWU/yr (SWU measures energy and time to convert U-238 to U-235) compared to about 15 million SWU/yr demand, and if we deploy SMRs en-masse following a similar path to Fig. 1, about 45-55 million SWU/yr would be a reasonable estimate for this increase. The U.S. only has one functioning commercial UF_6 conversion facility (Metropolis Works), which was shut down in 2017 and subsequently restarted in July 2023. Conversion and refining capability are a single point failure that makes SMR deployment at scale impossible in the face of strong geopolitical headwinds. [19] Fig. 2 clearly illustrates that the United States only controls 11% of UF6 conversion capacity and 8% of enrichment. Compare this to Russia and China, who currently control 44% of global conversion capacity, and 57% of enrichment. This means that the U.S. has to rely on Russia and China in order to meet even the current demand for nuclear fuel for its existing reactor fleet. So, even though the first SMR deployment in the U.S. would most likely be an American design, the bottleneck for deployment at scale is whether the U.S. can convert and enrich enough uranium without leaning on its geopolitical competitors.
Licensing has proved to be a bottleneck for traditional nuclear deployment, and unless the NRC makes sweeping changes to their regulatory paradigm it will be difficult to power the AI data center boom with SMRs. The NRC is currently building out a new licensing pathway (10 CFR Part 53) which is billed as being more risk-informed, technology-inclusive, and with a clearer schedule for approving licensing and permits. This proposed rule was published Oct. 31, 2024, and the comment period was subsequently extended to close Feb. 28, 2025, [25,26] For deployment this means that the licensing factory is being completely rebuilt while orders are still arriving. NRC Staff and applicants are learning a completely new regulatory framework while still running major work through the existing Parts 50/52, so even though this Part 53 is a step in the right direction, its unclear when it will start to practically take effect. Alongside Part 53, the NRC has also been operating under an explicit federal push to reform its operations, exemplified by a May 23, 2025 executive order that directed NRC reform. [27] The NRC announced a major reorganization on Feb. 4, 2026 that aimed at more efficient and timely decision making and accelerating the safe deployment of nuclear technologies. [28] This complete reorganization could prove fruitful in the long run, but it creates short term disruption to permitting which could delay mass SMR deployment by several years.
The United States has pioneered much of the technology required for a mass deployment of SMRs, but faces deployment bottlenecks for a whole host of reasons. This is why its plausible that the first commercial SMR deployments using U.S. designs will appear in Eastern Europe before the U.S. does. Romanias Doicești project, which is built around NuScale technology, is the closest commercial SMR plant to deployment for a variety of reasons. This plant has explicit U.S.- backed financing support on the order of $4 billion and is far enough along in the deployment process that Romanian officials are now discussing how the state nuclear energy utility will structure its investment in the facility (although this could take some time). [29] Concurrently, Ukraine has uniquely strong incentives to pursue additional nuclear capacity after the loss of the Zaporizhzhia nuclear plant (Europe's largest nuclear plant) which supplied over 20% of Ukraine's electricity before the invasion and is has been occupied by Russian forces seized it in March of 2022. [30] In order to make up for this shortfall Ukraine's state nuclear utility Energoatom and Holtec have signed agreements explicitly aimed at deploying Holtecs SMR-300 in Ukraine. [31] If Romania or Ukraine brings a U.S.-designed SMR online before we do, it should serve as a clear wake-up call: clear the bottlenecks at home, or accept that were sleepwalking into the largest energy shortfall since the discovery of electricity.
© Léon Garcia. The author warrants that the work is the author's own, and that Stanford University provided no input other than typesetting and referencing guidelines. The author grants permission to copy, distribute and display this work in unaltered form, with attribution to the author, for noncommercial purposes only. All other rights, including commercial rights, are reserved to the author.
[1] C. Yu, "Small Nuclear Reactors," Physics 241, Stanford University, Winter 2011.
[2] C. Maya, "Momentum is Building to Meet Electricity Demand in Texas With Small Nuclear Reactors," Associated Press, 17 Feb 26.
[3] "Small Modular Reactors: Nuclear Energy Market Potential for Near-term Deployment," Nuclear Energy Agency, No. 7213, 2016.
[4] "Nuclear Power: Nuclear Regulatory Commission Relies on Information From its Reactor Oversight Process to Ensure Safety," U.S. Government Accountability Office, GAO-25-107807, September 2025.
[5] M. Levy, "New Wave of Smaller, Cheaper Nuclear Reactors Sends US States Racing to Attract the Industry ," Associated Press, 28 Mar 25
[6] A. Kwok, "Nuclear Energy for Data Center Decarbonization," Physics 241, Stanford University, Winter 2026
[7] K. Kline, "Energy and Industrial Use Cases for Advanced Nuclear Reactors," National Association of State Energy Officials, October 2024
[8] T. Moran and D. Vine, "Advanced Nuclear Process Heat for Industrial Decarbonization," Center for Climate and Energy Solutions, July 2024.
[9] I. Penn and B. Plumer, "Nuclear Energy Project in Idaho Is Canceled," New York Times, 8 Nov 23.
[10] C. Mandler, "Three Mile Island Nuclear Plant Will Reopen to Power Microsoft Data Denters," National Public Radio, 20 Sep 24.
[11] A. Sainz, "NAACP, Environmental Group Notify Elon Musk's xAI Company of Intent to Sue Over Facility Pollution," Associated Press, 17 Jun 25.
[12] "New Source Performance Standards Review for Stationary Combustion Turbines and Stationary Gas Turbines," Federal Register, 91 F.R. 1910, 15 Jan 25.
[13] "Standard Design Approval for the Nuscale Power Plan Based on the Nuscale Standard Plant Design Certification Application," U.S. Nuclear Regulatory Commission, September 2020.
[14] L. Garcia, "How Ghana Can Achieve Its Nuclear Energy Goals With SMRs," Physics 240, Stanford University, Fall 2024.
[16] "Disposal of High-Level Radioactive Wastes in a Geologic Repository at Yucca Mountain, Nevada," U.S. Code of Federal Regulations, 10 C.F.R. pt. 63 (2025).
[17] J. Winterle, R. Pauline, and G. Ofoegbu, "Regulatory Perspectives in Deep Borehole Disposal Concepts," Center for Nuclear Waste Regulatory Analysis, May 2011.
[18] J. Amy, "Georgia Nuclear Plant Begins Splitting Atoms For First Time," Associated Press, 6 Mar 23.
[19] "Pathways to Commercial Liftoff: Advanced Nuclear," U.S. Department of Energy, September 2024.
[20] A. Finan et al., "Nuclear Energy: Supply Chain Deep Dive Assessment," U.S. Department of Energy, February 2022.
[21] "Quality Assurance Criteria for Nuclear Power Plants and Fuel Reprocessing Plants," U.S. Code of Regulations, 10 C.F.R. pt. 50, App. B, (2025).
[22] "Quality Assurance Requirements for Nuclear Facility Applications," American Society of Mechanicsl Engineeres, ASME NQA-1-2019, December 2019.
[23] A. Keim, "Quality Assurance Program Criteria (Design and Construction)," U.S. Nuclear Regulatory Commission, September 2023.
[24] "Analysis of Potential Impacts of Uranium Transfers on the Domestic Uranium Mining, Conversion, and Enrichment Industries," U.S. Department of Energy, 26 Apr 17.
[25] "Risk-Informed, Technolgy-Inclusive Regulatory Framework for Advanced Reactors," U.S. Federal Register 89 F.R. 86918, 31 Oct 24.
[26] "Risk-Informed, Tecnology-Inclusive Regulatory Framework For Advvanced Reactors," U.S. Federal Register, 89 F.R. 92609, September 2024.
[27] "Ordering Reform of the Nuclear Regulatory Commission," U.S. Federal Register, 90 F.R. 22587, 29 May 25.
[28] "NRC Major Reorganization Supports Efficiency, Innovation," Nuclear Regulatory Commission, NRC News No. 26-017, 4 Feb 26.
[29] "Planned SMR Nuclear Power Plant to Cost $6-7 Billion, Romanian PM Says," Reuters, 13 Feb 26.
[30] "Russian-Held Nuclear Plant Resumes Power Supply to Ukraine, Two Reactors Connected," Reuters, 26 Aug 22.
[31] "Ukraine's Energoatom and Holtec International Sign Master Agreement for Establishment of an Advanced Manufacturing Infrastructure for Building Capital Nuclear Equipment and Components in Ukraine," Holtec International, 16 Apr 24.
[15] A. Shehabi et al., "United States Data Center Energy Usage Report," Lawrence Berkeley National Laboratory, LBNL-2001637, December 2024.
[32] D. Dolzikova, "Power Plays, Developments in Russian Enriched Uranium Trade," U.K. Royal United Services Institute, March 2024
[33] M. Bowen and P. Dabbar, "Reducing Russian Involvement in Western Nuclear Power Markets," Center on Global Energy Policy, Columbia University, May 2022.