|
|
|
| Fig. 1: Simplified pressurized water reactor (PWR) schematic showing reactor, steam generator, turbine, and grid interface; used here to illustrate SMR coupling to a campus microgrid. (Source: Wikimedia Commons) |
Data center campuses hosting artificial intelligence (AI) models represent some of the fastest- growing loads on the U.S. grid. Their demand is concentrated, continuous, and highly sensitive to cost, reliability, and interconnection constraints. Here we ask a focused question: if on-site small modular reactors (SMRs) are deployed to power AI campuses, how do their unit costs compare to grid retail electricity, and what magnitude of annual savings could result? To make this comparison concrete, Fig. 1 illustrates a simplified SMR-based campus microgrid configuration used in the analysis below.
The North American Electric Reliability Corporation notes elevated risk of resource shortfalls in multiple regions as electrification accelerates and extreme weather strains legacy infrastructure. [1] NERC technical assessments also highlight that large new loads stress interconnection and voltage stability, and incident reviews document simultaneous tripping of voltage-sensitive loads during disturbances. [2,4]
On the demand side, large loads are scaling quickly. A report from the Lawrence Berkeley National Laboratory estimates 176 TWh of data-center consumption in 2023, with scenarios reaching 325- 580 TWh by 2028. [5] Individual AI campuses are commonly modeled at 100 - 500 MW, equivalent to the load of a mid-sized city. Their flat, around-the-clock demand profile makes them strong candidates for on-site baseload generation such as SMRs.
A civilian small modular reactor has been deployed. The floating nuclear power plant Akademik Lomonosov (two KLT-40S units) entered operation to supply grid power in Pevek, Chukotka, Russia, providing a non-military example of SMR-class technology in service. [6]
SMRs can be deployed in a campus microgrid alongside bulk transmission. In this architecture, the reactor supplies continuous baseload and can island the site during grid disturbances. Unlike diesel backup, which is idle until emergencies, an SMR operates continuously at high capacity (capacity factors 90%-95% per EIA assumptions), effectively offering much higher availability by design. [7,8]
The design provides two advantages: (1) cost hedging, since the LCOE of nuclear power can be compared directly to grid retail rates; and (2) interconnection relief, since local generation reduces transmission draw, easing transmission constraints. [3]
The unit-cost comparison used here is a back-of-the-envelope calculation:
| ΔC = Priceretail - LCOESMR |
| Savingsyr = ΔC × E |
| E = P × 8,760 |
where ΔC is savings per MWh; Priceretail is the retail grid electricity price ($/MWh); LCOESMR is the advanced-nuclear levelized cost of electricity ($/MWh); P is campus load (MW); and E is annual energy demand (MWh). According to EIA’s non-volatile publications, advanced nuclear (used here as a proxy for SMRs) has an LCOE of $67.09/MWh for 2030 service (2024 dollars). [7]
U.S. average commercial retail prices are approximately $141.5/MWh, with industrial prices varying by region (for example, from $66.9/MWh in the West South Central region to $253.1/MWh in California). [8] All LCOE values here are in 2024 dollars, whereas retail prices are nominal (current dollars). No inflation normalization is applied; results are therefore conservative if inflation is positive. [7,8]
|
||||||||||||||||||||||||||||||||||||||||||
| Table 1: Illustrative SMR vs. Grid Cost Scenarios. [7,8] |
Thus, for example, at U.S. commercial retail rates (~$141.5/MWh), an AI campus buying 300 MW continuously would pay about $372 M per year. With EIA’s advanced nuclear LCOE of ~$67/MWh, the same load would cost about $176 M per year, a difference of roughly $196 M per year. As shown in Fig. 2, these unit-cost deltas persist across regions: in high-price states such as California ...the difference approaches nearly half a billion dollars annually, whereas in the lowest-price industrial regions savings may vanish. [7,8]
 |
| Fig. 2: Retail vs. SMR unit electricity costs across representative cases. [7,8].) (Image source: L. Skarada.) |
The comparison here uses LCOE as a proxy for self-supply and retail rates as a proxy for purchase. EIA notes that LCOE omits system integration and interconnection costs, while retail prices vary with tariff riders and location. [7,8] In practice, actual site economics would also depend on financing, licensing, and backup requirements. Nuclear policy constraints further affect siting and schedules: in the U.S., spent fuel remains on plant sites pending a federal repository, with associated safety, proliferation, and public-acceptance implications. These policy frictions introduce schedule and financing risk not captured by LCOE; delays increase carrying costs and can erode the apparent unit-cost advantage.
As AI campuses grow toward hundreds of megawatts, the relative magnitude of retail versus nuclear self-supply costs will scale proportionally. Demonstrations of campus-dedicated SMRs could validate whether the unit savings identified here translate into net project savings once full capital, regulatory, and integration factors are incorporated.
An on-site SMR for a 300 MW AI campus could plausibly reduce purchased-power costs by $196 M/year at average U.S. commercial prices, with bounds ranging from negligible savings in the cheapest regions to nearly $500 M/year in high-price states.
© Lance Skarada. 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] "2024 Long-Term Reliability Assessment," North American Electric Reliability Corporation, December 2024.
[2] "2024 State of Reliability," North American Electric Reliability Corporation, June 2024.
[3] "Characteristics and Risks of Emerging Large Loads," North American Electric Reliability Corporation, July 2025.
[4] "Incident Review: Considering Simultaneous Voltage-Sensitive Load Reductions," North American Electric Reliability Corporation,, January 2025.
[5] A. Shehabi et al., "2024 United States Data Center Energy Usage Report," Lawrence Berkeley National Laboratory, LBNL-2001637, December 2024.
[6] Small Modular Reactors: Advances in SMR Developments 2024," International Atomic Energy Agency, 2024.
[7] "Capital Cost and Performance Characteristics for Utility-Scale Electric Power Generating Technologies," U.S. Energy Information Administration, January 2024.
[8] "Electric Power Monthly July 2025," U.S. Energy Information Administration, July 2025, Table 5.3.
