CU Boulder preprint says SMR operating costs, not capex, drive poor returns

Small modular reactors at their manufacturers' advertised costs are uneconomical in US electricity markets mainly because of their operating and fuel costs, not their investment costs, according to a preprint posted to arXiv on September 8, 2026 by Pradyumna Rao, Daniel T. Kaffine, and Bri-Mathias Hodge of the University of Colorado Boulder. Dispatching SMR designs hour by hour against historical and projected wholesale prices, the authors find that almost all concepts have discounted payback periods longer than their technical lifetimes on energy-market revenue alone, that 18 of the 22 concepts considered would not reach a 20-year payback at current production tax credit levels, and that at advertised costs SMRs are about as profitable as advanced estimates of Westinghouse's AP1000. The paper is a preprint; arXiv lists no journal publication for it.

The model is a price taker: when the hourly price is at or above a reactor's marginal cost, fuel plus variable operating cost, it runs at full power, and otherwise it ramps down to 40 percent of maximum. Prices come from ten hourly scenarios in the National Renewable Energy Laboratory's Cambium datasets spanning 2025 to 2050 and from 2011 to 2023 hourly prices in six wholesale markets; capacity payments are fixed monthly payments. Cost inputs come from the dataset assembled by Steigerwald and colleagues in 2023 in the journal Energy, extended with IAEA operational data and manufacturer information, in 2020 dollars. The paper focuses on six concepts, GE Hitachi's BWRX-300, Rolls-Royce's UK-SMR, Holtec's SMR-160, NuScale's VOYGR, Oklo's Aurora-15, and X-energy's Xe-100, which it says "performed the best in our analysis." Advertised investment costs in its Table 1 run from $2,250 per kWe for the BWRX-300 to $7,500 for the Xe-100; the authors say they do not assess the validity of those figures in detail.

The mechanism is the merit order. Because ramping and startup constraints force the reactors to keep running, they "absorb significant intra-hour losses" in hours when the price falls below their marginal cost. The authors' example is the UK-SMR: with fixed operating costs of $68 per MWh, which the paper takes from a 2022 Power Technology article, and marginal costs of $14.5 per MWh, its discounted payback period exceeds its technical lifetime "even if it were to incur no investment costs." Removing the ramping constraints would not fix it, the paper adds, because high marginal costs also compress margins in the hours the reactor does earn.

Fuel is the largest single reason. Across the compiled dataset of manufacturer-provided costs, average SMR fuel cost is $23.69 per MWh against $9 to $11 for traditional baseload reactors, and with fixed costs averaging $21.81 per MWh the reactors' operating and fuel costs total $45.5 per MWh, above the mean price of every historical and non-decarbonization scenario in the study except the full-decarbonization Mid Case 100, whose mean is $55.24. The authors attribute the premium to high-assay low-enriched uranium and TRISO fuels without established supply chains. Modeled to minimize losses, the reactors stop being baseload: lifetime capacity factors fall as low as 47.4 percent, against the 92 percent the authors say is expected of SMRs.

Capacity markets and tax credits

Capacity revenue helps less than recent auction prices suggest. Payback periods are most sensitive to wholesale prices, and for capacity payments to matter, average lifetime prices would need to significantly exceed the historic high of PJM's 2027/28 auction, which the paper puts at $10.14 per kW-month. Twelve of the 22 concepts do not reach a 20-year payback even at ERCOT's 2023 average price of $65.13 per MWh combined with NYISO's 2023 capacity price of $19.46 per kW-month.

Investment tax credits do little. The maximum 50 percent credit is insufficient to bring most concepts' payback below their technical lifetimes in almost all scenarios, and for some designs, the UK-SMR and NuScale's VOYGR among them, a credit covering the entire cost of construction still would not bring payback below the reactor's lifetime. Production tax credits act on marginal cost, and there the BWRX-300 and the Aurora-15 reach a 20-year payback at or under the Inflation Reduction Act's levels while 18 of 22 do not. At a PTC of $95 per MWh, a level the authors tie to the social cost of carbon of the gas generation displaced, most SMRs get under 20 years with or without capacity revenue.

The trade-off runs the other way from the industry's pitch. Improving marginal cost from the current SMR average of $25 per MWh to a baseload $12 relaxes the allowable investment cost by about 30 percent on average, to a ceiling of $7,813 per kWe under high wholesale prices, high capacity prices, and a high PTC. If instead SMR investment costs escalate the way nuclear projects historically have, which the paper puts at 117 percent on 97 percent of projects citing Sovacool and colleagues, most concepts end up with a performance profile similar to moderate and conservative estimates of Vogtle 3 and 4. "This sharply brings into question the benefits of the transition to advanced nuclear plants," the authors write.

The projects being financed are not merchant

The paper prices a merchant investor selling into ISO markets, and it notes, citing Weibezahn and Steigerwald, that most nuclear projects only become bankable for private investors after de-risking through government involvement. The nuclear announcements The Duck Curve has covered in September 2026 fit that description rather than the model. NextEra's Duane Arnold restart, a 615 MW boiling water reactor rather than an SMR, closed a DOE loan of up to $1.9 billion on September 8, with Google holding a 25-year agreement to buy carbon-free nuclear energy from the plant.

Studsvik, whose four-unit BWRX-300 project in Sweden uses one of the two designs the paper's numbers favor, applied for state financing support on June 12, 2026. Rolls-Royce SMR's first three UK units sit inside a programme for which the UK government pledged over £2.5 billion in the Spending Review period. The preprint's own introduction lists Microsoft's Three Mile Island restart and Amazon's and Google's power purchase agreements as material commitments to new nuclear; those are bilateral contracts, and the model does not price them.

The paper's stated contribution is causal. Steigerwald's 2023 study, as the preprint characterizes it, found no SMR concept would be competitive or profitable; the new work argues the reason is marginal cost, not the investment cost that most of the SMR debate is about, and says future work should examine other market structures, including novel ones proposed in the United States.

Outlook

If the paper's mechanism holds, the SMRs that get built are the ones with contracted revenue or state backing, which is what all three September deals above look like; the merchant SMR the model prices is absent from those announcements. The BWRX-300, one of the two designs the analysis treats most kindly, is the one Studsvik chose, and Sweden's decision on its state-aid application is the next test of whether a low advertised investment cost attracts the public money the paper says the economics require. The fuel finding cuts the other way for the HALEU and TRISO designs: on these numbers, a cheaper supply chain would do more for them than a cheaper factory.