Is Nuclear Power’s High Cost Justified for Clean Energy?

In a November 4 substack post, I discussed the economics of nuclear power in light of recent nuclear deals by Google, Amazon, and Microsoft. While optimists envision an important role for nuclear power in a transition to clean energy, historical and recent experience with nuclear power and its continued high cost suggests that a more skeptical view is warranted. 

Of particular importance is the fact that though efforts to reduce the regulatory burden on nuclear power are showing some signs of success, subsidies to nuclear remain high. I argued that both unnecessary regulations and subsidies to nuclear and other energy technologies should be removed.

Given growing concerns about climate change, the key question for nuclear is whether its societal benefit as a zero-carbon electricity source is worth its large construction costs. If the answer is yes, then the benefits may justify subsidies or other government interventions to support the building of new nuclear power plants. A definitive answer to this question, though, requires knowledge of the actual damages of carbon emissions, which continues to be debated. 

In our 2020 working paper on the economics of nuclear power, Peter Van Doren and I ask an alternative question: What level of carbon damages would justify construction of a nuclear power plant instead of an alternative fossil fuel power plant? With the recent flurry of nuclear activity and legal and regulatory changes, it is worthwhile to revisit our calculations.

To do so, I compare the levelized cost of electricity (LCOE) of a new nuclear power plant and a new natural gas combined cycle (NGCC) plant under different assumptions. The LCOE is an estimate of the lifetime costs of a power plant expressed in cents per unit of electricity produced (kilowatt-hours, kWh). The calculation accounts for the construction costs, construction time, financing, fixed and variable operations and maintenance, and fuel costs.¹

The assumptions for nuclear are based on a recent Department of Energy (DOE) report on nuclear power, though my assumptions differ in a few key ways.² In my last blog, I outlined why the DOE report is overly optimistic, which is reflected in its estimates of future nuclear LCOEs (see Figure 3 on page 4). Most importantly, it commits a common error made by many pro-nuclear projections. It describes the high cost of recent projects, outlines vague ways in which these costs will be reduced in the future, and then simply waves its hands and assumes substantially lower costs. 

The DOE estimates that the most recent US nuclear project (at Vogtle in Georgia) had a construction cost of $15,000 per kilowatt of capacity built ($/​kW) and took 11 years to build. It also provides LCOE estimates if construction cost and time are reduced by around 40 percent, to 8,300 $/​kW and 6 years. The most important factors for nuclear LCOE are construction costs and time (which affects the costs of financing), so such a large assumed reduction has a substantial effect.

The DOE also includes the effect of subsidies, including favorable federal loans and investment tax credits created by the Inflation Reduction Act. My calculations ignore these because: 1) the subsidies don’t change the underlying economics of nuclear power, they simply transfer some of the cost to taxpayers; and 2) as long as these subsidies are motivated entirely by climate change, their value is included as a portion of the carbon tax necessary to equate the cost of nuclear and alternative energy sources that I estimate below (i.e., a subsidy to clean energy is a negative carbon tax).

The LCOE of nuclear, at both the cost level seen at Vogtle and after assuming a 40 percent reduction in construction cost and time, are reported in Table 1.³ Also included are the estimated LCOEs of nuclear’s primary alternative, a natural gas combined cycle plant. While nuclear is dependent on construction costs, the most important variable for NGCC is fuel costs. The LCOE of NGCC is thus evaluated at varying fuel costs based on Energy Information Administration natural gas price projections.⁴

Table 1 also includes estimated LCOEs for NGCC with 95 percent Carbon Capture and Sequestration (CCS). This spring, the Environmental Protection Agency (EPA) enacted a new rule requiring that new natural gas plants operating more than 40 percent of the time must reduce their emissions by at least 90 percent, with the EPA determining that CCS is the best system for reducing emissions. The rule is likely shortsighted and is facing legal challenges. However, if the rule goes into effect as currently envisioned, NGCC without CCS will likely no longer be possible to build. Thus, I also estimate the LCOE of NGCC with 95 percent CCS based on cost estimates from the National Renewable Energy Lab (NREL).

Overall, the picture is clear. Nuclear at either assumed construction cost and time is expensive. NGCC without CCS is much cheaper, and NGCC with CCS is slightly more expensive but still less than nuclear at any assumed fuel cost. How does this picture change if climate damages are considered?

The social cost of carbon necessary to justify building a nuclear power plant instead of the much cheaper NGCC plants is calculated using NREL estimates of the carbon intensity of the modeled NGCC plants.⁵ The results, reported in Table 2, make it clear that, at the high-cost level recently seen in the United States, nuclear’s clean energy benefit does not outweigh its large costs. 

For example, for the construction of nuclear at the high assumptions (LCOE of 22.3 cents per kWh) instead of an NGCC with low assumed future natural gas prices (LCOE of 4.1 cents per kWh) to be justified by climate change the estimated damages of carbon emissions would need to be more than $400 per ton.

The estimates of the carbon tax needed to equate the LCOEs of NGCC without CCS and nuclear at the assumed low-cost level are closer and in line with recent federal estimates of the social cost of carbon ($190 per metric ton in 2020 dollars), though the assumptions behind those estimates are up for debate. Of course, whether those construction costs and times are even achievable is also doubtful.

When NGCC with CCS is considered, the required assumed costs of carbon damages are enormous because the modeled natural gas plants are emitting only 5 percent of the carbon emitted by the plant without CCS. Of course, CCS technology is still unproven, so the cost estimates are theoretical (and highly controversial in their own right) and exclude important factors like the transport and storage of the captured carbon emissions. Ironically, prospects for either CCS or nuclear may end up depending on how economically infeasible the other is.

Overall, this analysis highlights two points. First, nuclear will require substantial cost reductions before it is worth investing in, whether its clean energy benefit is considered or not. Second, our current web of subsidies and regulations makes it difficult to evaluate the benefits and costs of various energy sources. The complexity restricts our ability to optimize energy sources considering a range of factors including cost, reliability, and environmental damages. We should stop favoring certain sources through subsidies and punishing others through regulation, and, if necessary, find ways to directly price in unquantified costs and benefits.

Read the Cato-at-Liberty blog post here.

1. For a more detailed explanation of our LCOE methodology, see the appendix of our paper.

2. An additional key consideration is the discount rate used, which for the sake of simplicity I have omitted here. The DOE uses a lower discount rate, derived from assumed debt-to-equity ratios and interest rates. Following our paper, I instead envision the discount rate as the opportunity cost of investing in the project as opposed to the expected return from investing elsewhere. This results in a higher discount rate (7 percent) than used by the DOE. The assumed discount rate is especially important for nuclear with high upfront costs, and use of a lower discount rate creates lower LCOE estimates. See pp. 43–44 in our working paper for more detail.

3. The levelized cost is calculated according to my paper with Peter Van Doren with the following assumptions: All technology has a 7 percent cost of capital (discount rate). Otherwise, as in the DOE report, assumptions are based on the National Renewable Energy Lab Annual Technology Baseline. With the exception of the varying overnight costs and construction time described in text, both high and low nuclear estimates are modeled as a power plant with 1000 MW capacity, 93 percent capacity factor, 60-year life, heat rate of 10,500 MMBtu/​MWh, fixed O&M costs of 175 $/kw-yr, variable O&M costs of 2.8 $/​MWh, and fuel cost of 0.97 $/​MMBtu. The natural gas plants modeled are H‑frame 2‑on‑1 combined cycle plants with and without 95 percent CCS. Both are assumed to have capacity factors of 85 percent, 30-year lives, take 3 years to construct, and varying fuel costs as described in the text. The plant without CCS has a capacity of 992 MW, OCC of 1,280 $/​kW, heat rate of 6,200 MMBtu/​MWh, fixed O&M of 34 $/kw-yr, and variable O&M of 2.16 $/​MWh. The plant with CCS has a capacity of 877 MW, OCC of 2,520 $/​kW, heat rate of 7,010 MMBtu/​MWh, fixed O&M of 66 $/kw-yr, and variable O&M of 4.86 $/​MWh.

4. Natural gas price is calculated based on Annual Energy Outlook 2023 projections for natural gas price delivered to electric power sector (Table 13, “Natural Gas Supply, Disposition, and Prices”). The real price per Mcf is converted to 2022 $/​MMBtu using conversion of 1.034 MMBtu/​Mcf. The prices used are the average price for the reference, high gas supply, and low gas supply scenarios for the full 2022 to 2050 period.

5. As described in the working paper, we calculated a static average carbon tax over the lifetime of the power plant. If the carbon tax is assumed to escalate over time, the tax amount in 2024 will be smaller than this average. Table 2 here reports the approximate carbon tax in 2024 assuming a real escalation rate of 2 percent. See pp. 54 and 96–97 of our paper.