Nuclear Energy for Data Center Decarbonization

Introduction

Fig. 1: Global map of large data centre clusters, 2024. Image source: A. Kwok, after the IAEA. [2]

At the dawn of cloud computing, artificial intelligence workloads, and economy-wide digitalization, data center operations have scaled at an unprecedented pace. [1] As shown in Fig. 1, over the past decade, global electricity demand from data centers has grown rapidly, transforming what was once a marginal load into a structurally significant component of global power consumption. Such acceleration consequently raises critical challenges across technical, environmental, and social dimensions of energy systems.

Hyperscale data center operators today primarily rely on grid connectivity as foundation of their energy supply strategy, supplementing grid energy with renewable electricity procured through, for instance, power purchase agreements (PPAs) to meet decarbonization targets. However, if incremental demand is met with current state of energy grid power generation, the emissions consequences are substantial. Under a scenario in which 30% of the projected 946 TWh data center electricity demand is met by coal (emission factor of 94.6 tCO₂ per PJ combusted, equivalent to 0.340 kg CO₂ per kWh) and 26% by natural gas (emission factor of 56.1 tCO₂ per PJ combusted, equivalent to 0.202 kg CO₂ per KWh), the expected carbon dioxide emissions would revolve around 146 million tons globally, equivalent to 0.4% of the worlds current carbon dioxide emissions. [2,3] This increase alone would rival the annual emissions of mid-sized industrialized economies, such as the Netherlands, highlighting the global climate impact of unchecked data center growth. [4]

While renewables have played a growing role in data center decarbonization strategies, their intermittence and temporal mismatch with constant computing loads limit their ability to alone provide 24/7 supply of energy. [5] Thus, against this backdrop, nuclear energy has re-emerged in the conversation, driven by its low operational emissions and high conversion factor able to match data center reliability requirements. [6] While hyperscalers interest in nuclear plant revivals (e.g. Three Mile Island) and new- build pathways (e.g. Small Modular Reactor (SMR) projects) have grown significantly in the past years, the central question revolves around feasibility, whether these future nuclear commitments actually translate into timely, additional generation at scale or whether they primarily function as accounting solutions that leave the grids marginal fossil generation to meet the gap.

Data Center Incremental Energy Demand Projections and Nuclear Reactor Implications

The first question revolves around scale: how many nuclear reactors would be required to meet a proportion of projected incremental energy demand from data centers?

In terms of demand, as major cloud providers race to expand their data center capacity, so does their demand for electricity and power. According to the International Energy Agency (IEA), in 2024, global data center electricity use was estimated at 416 TWh (~1.5% of global demand) and by 2030, data center energy consumption is expected to reach 946 TWh, equivalent to a compound annual growth rate (CAGR) of 15%. This implies an incremental energy demand of 530 TWh between 2024 and 2030. To maintain the proportion of nuclear mix powering data centers (~10% according to the IEA), 53 TWh of the incremental 530 TWh would need to be supplied by nuclear generation between 2024 to 2030. [2] We will be utilizing these projections and proportions to estimate future incremental electricity demand, using 2024 as base year and assessing implications of different generation pathways through 2030.

In terms of supply, according to the International Atomic Energy Agency (IAEA), as of December 2024, the global operational nuclear power capacity was 377 GW provided by 417 reactors across 31 Member States, equivalent to an average reactor nameplate electric capacity of roughly 1 GW. [7] Assuming a capacity factor of 81%, a conservative average based on historical global performance, we can estimate the annual energy production per reactor from these nuclear plants: [8]

Table 1 provides visibility on how the number of nuclear reactors required to meet demand scales with across scenarios of varying nuclear energy share under the base scenario, where total incremental demand equals 530 TWh.

A scenario in which the current energy mix extends proportionately with increasing data center demand (~10% of energy powering data centers coming from nuclear power) would imply that 53 TWh of 530 TWh projected incremental data center electricity demand between 2024 and 2030 is to be met by nuclear energy. At an average output of 7.1 TWh per gigawatt-scale reactor annually, this would require deployment of approximately 8 gigawatt-scale reactors globally within six years to merely preserve the 10% nuclear share of incremental demand, not even accounting for reactor retirements, delays in commissioning, or competing electricity demand from growth in other industries. Scaling nuclear energy's contribution beyond its current share further magnifies the challenge: meeting 30% of incremental data center demand would require nearly 23 new gigawatt-scale large reactors, while supplying the entirety of incremental demand would imply approximately 75 new gigawatt-scale reactors coming online by 2030. And if future projects skew towards smaller reactor sizes, such as Small Modular Reactors (SMRs), the required number of units increases substantially (from 88 in the minimal contribution case to 177 in the status quo case up to 1767 in the full substitution case), amplifying the supply chain, siting, and scale complexity.

Such deployment rate is substantial but structurally constrained. The United States, being the largest consumer of data center electricity, has completed only a limited number of new large nuclear reactors projects in the recent decades, the most recent one being Vogtle Unit 3 after over three decades of development. [9] In the case of gigawatt-scale reactors, while announcements of plant restarts signal renewed interest, most appear to remain in the early-stage licensing phases with unclear information on how many are actually being revived or built currently. In the case of Small Modular Reactor (SMR) projects, by mid-2025, while over 80 SMR concepts were under development globally, only 2 were operating commercially (the Akademik Lomonosov FNPP in Russia and the HTR-PM in China) with many remaining years away from proven commercial designs. [9,10] And even with these new plans, the global nuclear share has decreased its peak of around 18% in the late 1990s to 9-10% as of today. [11] These historical trends thus infer uncertainty on the feasibility of delivering substantial additional nuclear capacity within such short timeframes and consequently on whether nuclear energy will constitute a meaningful share of the electricity mix supplying data centers.

Sensitivity Analysis: Demand Growth Scenarios

To capture uncertainty in future data center growth, we further break down the analysis to consider three demand scenarios. The Lift-Off Case assumes stronger AI adoption and rapid deployment of complex, power-heavy workloads; the Base Case reflects the trajectory of electricity consumption in data centers assuming current regulatory conditions and industry projections; and the Headwinds Case captures the impact of a downside in data center deployment outlook, with slower than expected AI adoption due to tighter supply chains causing delays in capacity expansion. Resulting energy projections are summarized in Table 2.

Building on these demand projections, reactor requirements can be scaled across different nuclear energy proportions in the energy mix. Fig. 2 illustrates the number of gigawatt-scale reactors required as a function of nuclear energy's share of incremental demand under each growth scenario.

Fig. 2: Nuclear Reactor Requirements Across Demand Growth Scenarios (2024-2030). (Image source: A. Kwok, after the IAEA. [2])

The scaling problem thus becomes evidently significant. Across all three scenarios, reactor requirements increase but diverge in magnitude depending on the demand growth. Under the headwinds case, supplying 100% of incremental demand would require ~29 gigawatt-scale reactors by 2030; under the Base Case, this increases to ~75 reactors; while under the Lift-Off Case ~120 reactors would need to come online within six years.

Even more moderate nuclear contributions are significant, where maintaining only a 10% nuclear share would require around 12 new gigawatt-scale reactors in the lift-off case, 8 in the base case, and 3 in the headwinds case. While faster demand growth clearly increases scale of deployment required, even conservative scenarios still necessitate several new reactors coming online within a relatively short period. Against recent nuclear construction trends, this represents significant acceleration in pace, where over the past decade many large reactor projects have faced lengthy permitting processes, construction delays, and cost overruns. [12] These thus further the practical questions about how quickly nuclear capacity can realistically expand to keep pace with projected data center demand and become a non-marginal share of the decarbonization strategy.

The Limitations of Electricity Prices, Clean Reallocation, and Waste Management

The second question revolves around societal and environmental implications: what are limitations of nuclear energy in the data center landscape?

From a social perspective, in the scenario where nuclear or other forms of energy are unable to grow with the accelerating incremental pace of demand, rapid data center expansion would place upward pressure on electricity prices. We have already seen upward pressure in electricity prices, for instance, in California, where inflation-adjusted average retail electricity prices have increased by 6.2 cents/KWh. [13] With the additional load from large data centers over short time periods, grid constrained regions as large datacenters could tightening reserve margins and increasing reliance on higher-cost generation. In a market where marginal pricing determines wholesale electricity rates, this incremental demand from energy-intensive workloads driving strains on the grid would cause increase prices for all consumers, raising concerns about equity and distributional impacts as households and smaller businesses that share the same grid infrastructure would have to face the increased electricity cost without contributing meaningfully to the supply shock.

From an environmental perspective, even assuming nuclear energy procurement expand, the emissions outcome depend on whether new generation is truly additional. Currently, contractual instruments used to certify green energy is debatable as while environmental benefits are only guaranteed through certain criteria, such as additionality, the GHG protocol does not require fulfilling such criteria. [14] Thus, if data centers are unable to expand energy production as fast as demand, firm may claim carbon- free energy procurement but would contract electricity from existing nuclear plants or secure output from renewable energy facilities who would otherwise have served the grid, causing the physical grid to continue relying on fossil generators to balance displaced demand already existing in the grid. This creates the problem of clean reallocation, where without sufficient new low-carbon capacity entering operation, data center growth risks reinforcing dependence on brown energy at the margin, undermining decarbonization goals.

Finally, from a waste management perspective, assuming that nuclear generation is able to grow substantially, there comes the question of long-term waste management. High-level nuclear waste stored is extremely dangerous and takes thousands of years to decay and expanding nuclear capacity to meet incremental data center electricity demand would proportionally increase the quantity of spent fuel requiring storage and disposal. Although technical solutions, such as deep geological repositories, have been proposed, risks associated with storage, national security, and nuclear proliferation, have historically complicated project deployments. [15] As such, limitations are not solely constrained to engineering feasibility and scale, but also the governance and societal implications of managing radioactive waste. These all create further constraints surrounding the feasibility of nuclear energy as a structural source of power for data centers.

Conclusion

The acceleration of data center electricity demand represents a significant shift in how energy systems must evolve to support the digital economy. While nuclear energy offers clear advantages of low operational emissions and reliable generation, this analysis shows that feasibility hinges highly on limitations involving scale and timing, where even maintaining nuclear energy's current proportional role requires rapid reactor deployment acceleration relative to recent construction trends. These consequently raise practical questions about whether nuclear capacity can expand quickly enough to meet incremental data center energy demand in a timely manner.

Beyond scale, additional system-level challenges further complicate nuclear expansion. Potential upward pressure on electricity prices in constrained markets, long-term radioactive waste challenges, and risks of clean reallocation raise societal and environmental concerns about rapid nuclear expansion to match data center demand. Thus, while nuclear energy might offer an attractive solution to the data center energy demand problem on paper, without true additionality and feasible deployment pathways, such efforts risk remaining more accounting-based than physically transformative.

© Amber Kwok. 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.

References

[1] D. R. E. Ewim et al., "Impact of Data Centers on Climate Change: A Review of Energy Efficient Strategies," J. Eng. Exact Sci. 9, 16397-01e (2023).

[2] "Energy and AI," International Energy Agency, April 2025.

[3] "Global Energy Review 2025," International Energy Agency, March 2025.

[4] "The Netherlands 2024," International Energy Agency, January 2025.

[5] W. E. Gnibga, A. Blavette, and A.-C. Orgerie, "Renewable Energy in Data Centers: The Dilemma of Electrical Grid Dependency and Autonomy Costs," IEEE 10227588. IEEE Trans. Sustain. Comput. 9, 315 (2023).

[6] K. Grandin, P. Jagers, and S. Kullander, "Nuclear Energy," Ambio 39, 26 (2010).

[7] "Nuclear Power Reactors in the World," International Atomic Energy Agency, IAEA-RDS-2/45, 2025.

[8] N. Bolson, P. Prieto, and T. Patzek, "Capacity Factors For Electrical Power Generation From Renewable and Nonrenewable Sources," Proc. Natl. Acad. Sci. (USA) 119, e2205429119 (2022).

[9] The Path to a New Era for Nuclear Energy, International Energy Agency, January 2025.

[10] Z. Zhou et asl., "Effect of KLT-40S Fuel Assembly Design on Burnup Characteristics," Energies 16, 3364 (2023).

[11] "Energy, Electricity and Nuclear Power Estimates for the Period up to 2050," International Atomic Energy Agency, IAEA-RDS-1/45, December 2023.

[12] S. Yüksel et al., "Evaluating the Advantages and Disadvantages of Thorium-Based Nuclear Power Plants," in Renewable Energy Investments for Sustainable Business Projects, ed. by H. Dinçer and S. Yüksel (Emerald Group Publishing Ltd., 2023).

[13] R. Wiser et al., "Factors Influencing Recent Trends in Retail Electricity Prices in the United States," Electr. J. 38, 107516 (2025).

[14] T. Aslan et al., "Toward Climate Neutral Data Centers: Greenhouse Gas Inventory, Scenarios, and Strategies," iScience 28, 111637 (2025).

[15] O. P. Unokiwedi et al., "Deep Geological Repositories - A Review of Design Concepts, Near-Field Evolution, and Their Implications For Nuclear Waste Containment," J. Environ. Radioact. 289, 107750 (2025).