November 18, 2023: Nuclear Power
Good afternoon. I have been meaning for a long time to write about nuclear power. A few weeks ago, when I solicited topic ideas, a reader suggested commenting on Jack Devanney’s ideas. Thanks for the suggestion. In particular, I will look at Devanney’s book, Why Nuclear Power has been a Flop. The book is from Devanney’s organization, The Gordian Knot Group, which posits nuclear power as a solution to the twin challenges of energy poverty and global warming. The book can be purchased on Amazon or downloaded for free with registration. There are also 54 papers on various topics that are explored in the book. The material can also be viewed on Devanney’s Substack.
The story is familiar. After the development of the atomic bombs, it was realized that fission could be used for power plants as well. Hopes in the 1960s were that nuclear power would become “too cheap to meter”. But in the 1970s, construction costs escalated. Then came Three Mile Island in 1979, followed by Chernobyl in 1986 and Fukushima in 2011. Regulations compounded, public opinion soured, and the industry atrophied. The growth in the amount of nuclear electricity slowed in the 1980s and stopped around 2000. What went wrong, and is the problem fixable?
Let’s start with cost. The most common cost metric is the levelized cost of electricity (LCOE), which is the rate at which a power plant must sell electricity over its lifetime to break even financially. LCOE takes into account capital costs of the plant, fuel costs, maintenance and other operating costs, and eventual decommissioning. LCOE also takes into account a discount rate, which is a percentage financial devaluation over time. Since a power plant has the bulk of its expenses up front and generates power later, the LCOE is typically higher with a higher discount rate, because the future electricity is not as valuable. With high construction costs and long operating lifetimes, nuclear plants are particularly sensitive to the discount rate.
At a 7% discount rate, the LCOE of nuclear power for new plants ranges from 4.2 to 10.2 cents/kWh across major nuclear producers, with 7.1 ¢/kWh in the United States. The aforementioned paper also documents how nuclear costs vary greatly across countries, in addition to over time. People who argue that nuclear power is inherently expensive are at a loss to explain this wide variation, nor do they seem to be terribly curious. Devanney argues that, if nuclear power was properly regulated and if there was a well-functioning industry, then the cost should be no more than 3 ¢/kWh.
Devanney gives several reasons for why the cost is well over twice what it “should” be. One reason is ALARA (As Low As Reasonably Achievable), which is a regulation imposed by the Nuclear Regulatory Commission that nuclear plant operators must reduce radiation to the lowest level that they can afford. This, according to Devanney, guarantees that nuclear power cannot be cheaper than alternatives, because if it was, radiation restrictions would be tightened to the point where nuclear power is no longer cheaper. This creates a major disincentive for U.S. plant operators to innovate to reduce costs, because any cost-saving measure they implement would be eaten up by the regulatory ratchet.
ALARA is based on the Linear No-Threshold (LNT) model of radiation exposure, which holds that the health damage from radiation is proportional to the cumulative cost over a lifetime. Devanney marshals extensive evidence that LNT is a bad model: in fact, an acute dose received in a short time is much more hazardous than the same amount of radiation spread over a long time. The reason is that cells are equipped with repair mechanisms that don’t simply allow low-level damage to accumulate over time. It is similar to the idea that is one drinks one glass of wine every night for a year, the result will be little or no health damage, but if one consumes the alcohol equivalent of 365 glasses of wine in one evening, the result will probably be death from alcohol poisoning.
LNT, compared to other radiation damage models, greatly elevates the risk of low, chronic doses of radiation and thus provides a justification for ALARA. Devanney instead posts a sigmoid no-threshold (SNT) model, in which the damage from radiation is cumulative and follows a sigmoid (S-curve) function for acute doses. There are other potential models as well. LNT and ALARA are used outside of nuclear power regulation, such as in radiology. Medical imaging procedures, such as an MRI or a CT scan, involves injecting a radioactive material in the patient, the decay of which can be observed externally.
Another criticism of LNT come from Siegel et al.’s literature review. Another literature review from Weber and Zanzonico offers a somewhat more balanced but still critical review of LNT. This review cites a few papers that support LNT, but the bulk of the evidence is against its validity.
Nor can LNT be protected behind the precautionary principle, which is the notion that, in the face of uncertainty, it is better to assume the highest plausible levels of risk for regulatory purposes. Whether LNT is even plausible is another matter, but the problem with this use of the precautionary principle is that there is no “safe” alternative when it comes to radiation exposure. In the case of medical imaging, if LNT is not valid but nevertheless used in medical decisions, then there is a risk of foregoing discovery of vital information, endangering patients’ lives. In the case of nuclear power, if LNT is not valid but nevertheless is used as a basis of regulation that pushes nuclear power out of the market, then the result will be more fossil fuel combustion, which is known to contribute to global warming, and in the case of coal—nuclear’s most immediate competitor—severe health problems.
To go back to the book’s titular question, Devanney does not put the bulk of the blame on anti-nuclear activists, as one might expect, but on the industry itself. “Industry” might be too generous a word. Devanney defines the nuclear establishment as comprising the following entities:
vendors of nuclear plants,
utilities,
the Department of Energy and national labs,
university nuclear engineering departments,
and the Nuclear Regulatory Commission.
These entities are comfortable with the status quo. Vendors in particular like subsidies and the regulatory moat that exists today, which prevents new players from gaining traction and posing genuine competition. Vendors like ALARA in particular for this reason. The industry has also engaged in two lies which further underpin regulation and keeps the regulatory moat full.
A radiation release can’t happen. Obviously it can; three prominent releases, as noted above, occurred at Three Mile Island in 1979, Chernobyl in 1986, and Fukushima in 2011, and there have been other, less prominent releases. The industry promises perfect safety—past releases are flukes caused by extremely unusual circumstances that won’t happen again. When the next release comes, and it surely will, public trust will be further damaged, and there will be demands for yet more regulation which will again help the incumbents.
A radiation release is a catastrophic event. LNT greatly exaggerates the expected death toll from a release. The direct death toll (excluding indirect deaths, such as the misguided evacuation of elderly residents from around Fukushima) from all radiation releases is likely in the hundreds. Commercial aviation is a bit older than commercial nuclear power. Devanney estimates the death toll from plane crashes at 45,000, excluding terrorism. Yet aviation is considered to be safe by the general public, while nuclear power is considered to be dangerous.
In a rational world, we would accept some risk from nuclear power, as we do from flying, from driving, and from any number of activities.
Aside from the specific ALARA rule, Devanney finds the regulatory regime around nuclear power to be overly prescriptive, perhaps reflecting the industry’s origins as a direct government activity. Regulatory bodies use fault tree analysis, which seeks to enumerate all possible ways a reactor could fail, assign probabilities to them, and regulate to ensure those probabilities are not over a threshold. Fault tree analysis, in the context of nuclear regulation, has several problems. First, there are so many variables in the tree whose probability can’t be reliably estimated that the result is nothing more than guesswork. Second, actual failures occur from factors that are not in the tree.
In aviation, we assume that plane crashes will happen, and that they will happen due to reasons that are not foreseen in advance. Aircraft manufacturers, at significant expense, install black boxes in airplanes so that when the next crash happens, they can deduce and correct the problems that caused it. The public understands and favors this practice. It is understood that, for airline safety, computer models are not a substitute for real-life operating experience. The same principle should be recognized for nuclear power.
Not only is the overly prescriptive regulatory framework unsound, it discourages innovation. Plant designers cannot make even minor changes to reactor design without a lengthy and expensive review from the NRC, and so it is better to not innovate and stick with the prescribed design, even if it is suboptimal or even poses a safety risk.
Devanney considers the issue of waste disposal and regards it as a non-issue. Today, used fuel is stored on-site in dry casks, and this is a perfectly safe solution and one that costs about 1% of the total cost of nuclear electricity. Yet the nuclear establishment has embraced the idea that waste storage is a serious problem and has collected billions of dollars in an unsuccessful attempt to solve it. Efforts in the United States have revolved around a repository at Yucca Mountain, a project that has been put in limbo by frivolous NIMBY concerns. Many jurisdictions, such as the State of Oregon, prohibit new nuclear power until the disposal issue is “solved”. The Shirky Principle strikes again: anti-nuclear activists don’t want to solve the waste problem when they can use it as a pretext against nuclear power. In short, waste is a problem only insofar as several politically powerful entities have an interest in keeping it a problem.
Another non-issue that Devanney discusses is proliferation. Despite anxieties, states with civilian nuclear problems have not been historically more likely to develop nuclear weapons than those without, whereas there is plenty of precedent—with the United States being a prime example—that states can develop nuclear weapons without nuclear electricity if they are determined to do so. A fundamental issue is that the level of enrichment required for a nuclear bomb is far greater than that required for a power plant, and so repurposing civilian infrastructure for military use is, while theoretically possible, a terribly inefficient way to develop nuclear weapons. As with other issues, the industry likes the proliferation issue because is further creates a regulatory moat, and national security is the perfect pretext for protectionism.
Devennay is not impressed by small modular reactors, which are only going to suffer from economies of scale. The problem is especially acute with nuclear power, since the cost of regulatory compliance scales sublinearly in the size of a reactor. The recent failure of what was supposed to be NuScale’s first SMR plant would seem to vindicate these concerns. More broadly, unless these major regulatory problems are solved, no small modular reactor, “advanced” reactor, or any other proposed design will get beyond the liminal space between success and failure in which the industry resides today. To really solve the problem and let nuclear power flourish as it should, Devanney argues that the industry needs to renounce the two lies above, move away from prescriptive and toward a standards-based regulatory regime, and subject the industry to genuine competition.
Devanney is not optimistic that these things can happen in any wealthy country. But there might be a chance is a rapidly developing country, which will both appreciate the importance of electricity abundance and not have an entrenched nuclear establishment like the United States has. That is why Devanney’s company, ThorCon, is developing a project in Indonesia. ThorCon’s goal is a shipyard-assembled, molten salt reactor that will achieve a price point of no more than 3 cents/kWh. Devanney is not certain that nuclear power can flourish from a developing country, but he is certain that it can’t from a wealthy country. Here is a write-up of ThorCon. The company is aiming for a 500 megawatt (about a tenth the output of a full-scale nuclear power plant) demonstration in Indonesia by 2029.
In my lifetime, I have been nothing but stagnation in nuclear power, and I have seen quite a few initiatives come and go without remaking the industry as hoped. Thus I find it difficult to be optimistic about nuclear power. But Devanney makes a strong argument, and his approach may be the most promising available.
Unfortunately, there are many parallels between the historical trajectory of nuclear power and current events with artificial intelligence. In testimony to Congress earlier this year, then-OpenAI CEO Sam Altman stated that the Nuclear Regulatory Commission is a good model for AI regulation. Both the industry and anti-AI activists argue that there is an unacceptable risk of grievous harm from AI. Theoretical musings about existential risk from AI are as disconnected from reality as are fault tree analyses for nuclear plant failures. Altman did not explicitly state that his goal in talking up AI risk is to build a regulatory moat, but the conclusion is hard to avoid. The worst part is that the notion that AI is somehow dangerous is seeping into the public consciousness, and as with the perceived risk of nuclear power, once the fear is learned, it is next to impossible to unlearn.
Quick Hits
Nuclear power isn’t the only energy sector that is struggling. The news for offshore wind in 2023 has been grim. The industry chalks this up to temporary factors—inflation and supply chain issues—and calls for more subsidies. The price of offshore wind was about 11 cents/kWh in 2021, which is infeasibly expensive. The hope is that, with deployment, improved technology and learning curve effects will bring the price down to a more reasonable 5-6 ¢/kWh. That has been the hope for a long time, and the story of offshore wind is a cautionary tale about relying too heavily on learning curve effects.
Myanmar has been in a state of civil war for a long time, but the last few weeks have brought about dramatic advancements for the opposition to that country’s military junta, so much that it looks like the junta may fall. Earlier this year, Brookings published a background piece on the civil war, which was continued at varying levels of intensity since 1948.
Allison Schrager considers the implications of higher interest rates, and among other things considers how this will impact the tech industry. Many people I talk to still imagine that things will go back to normal. Conditions are more normal now; it was an extended period of zero interest rates and dubious crypto ventures that were the aberration. One sign I noticed that the salad days of the industry were over slightly predates the rise of interest rates. Portland had a thriving community of events and seminars prior to March 2020. Most of those events never returned after COVID-19-era restrictions eased, and those that returned are shadows of their formers selves.