Catalyst N° 007 of 125 16 Nov 2023
The cost of nuclear
with Jessica Lovering, co-founder and executive director, Good Energy Collective
In this note
The question
Building a reactor in South Korea costs about a quarter what it costs in the United States. Why, and what would actually make American nuclear cheap?
The answer
Not regulation, which is the answer both sides of the usual argument reach for. The dominant factor is volume: nuclear gets cheap where a country builds many reactors of the same design in a row. And the reason that works is not really about nuclear at all, because the reactor is only about an eighth of what a nuclear plant costs.
03The argument
Start with the spread, because it is the clue. A solar panel costs roughly the same anywhere on earth. Nuclear ranges from about $2,200 per kilowatt in South Korea to roughly $8,000 for Vogtle in the US and the recent Finnish project, a four-fold difference. That spread exists because a nuclear plant is not a manufactured product, it is an infrastructure project, more like a bridge or a dam than like an airplane. Each one is substantially unique, and uniqueness is expensive.
Now open up the cost. For an AP1000, the Westinghouse design built at Vogtle, roughly half the cost is the plant outside the reactor: the yard, cooling infrastructure, groundwork. The nuclear island itself, meaning the reactor and its containment structure, is only about 12%. Another 35% or so is engineering, procurement, construction management and owner’s costs such as interest accruing during construction. Which means that when people say nuclear is expensive, they are mostly making a claim about large construction projects and financing, not about atoms. That reframing carries most of the episode.
It also explains what makes nuclear cheap where it is cheap. Building the same design repeatedly produces what Lovering calls economies of multiples, where not only the technology improves but the industry, the workforce and the regulator all climb the same learning curve. South Korea started around $4,000 to $5,000 per kilowatt and came down. France shows the same learning, and tellingly the curve jumps back up every time France introduces a new reactor design. The US stopped building in the 1980s and never accumulated the repetitions.
Which sets up the regulation question directly. Lovering’s evidence for demoting it is clean: South Korea’s regulator is modeled on the US NRC, which is widely treated as the international gold standard. If American regulation were the dominant cost driver, Korea building under a similar framework should be expensive too, and it is the cheapest in the world. She grants that licensing could be modernized for passive-safety designs, and calls it a real factor, but not a silver bullet. Her alternative explanation is that the US never had demand-pull policy for nuclear. Renewables got three decades of tax credits and portfolio standards that manufactured demand and induced the innovation that took solar from roughly $100,000 per kilowatt in the 1970s to today’s prices. Nuclear got essentially nothing comparable until recently.
04What you need to know first
- Dollars per kilowatt
- The upfront capital cost to build capacity, which is the number that varies wildly. Distinct from dollars per megawatt-hour, the cost of the electricity produced.
- The nuclear island
- The reactor and containment structure specifically, as opposed to everything else on the site. Only about 12% of project cost, which is the single most counterintuitive fact in the episode.
- First-of-a-kind
- The first unit of any new design, always the most expensive. The whole cost-reduction argument depends on what happens after it.
- Passive safety
- Safety from physics rather than machinery. Lovering’s example: traditional reactors use robust, redundant pumps to move coolant, while many advanced designs use convection, hot fluid rising and cool fluid sinking the way a kettle works. Fewer over-engineered parts means cheaper to build, run and maintain.
05Details worth keeping
- Existing nuclear is a cost paradox. Enormous to build, then among the cheapest electricity available: second cheapest in the US after hydro, and France has the cheapest electricity in Europe at 80% nuclear. Nearly all of nuclear’s cost is capital, with fuel a minor component, so it behaves more like wind or solar economically than like a gas plant.
- Globally this is a growth industry even though it feels stagnant from the US. China had around 22 reactors under construction, more than anyone, and was about to pass France for the second-largest fleet. Over 30 countries are pursuing their first plants.
- China’s costs are genuinely opaque because the developers are state-owned, but construction duration is reported internationally and serves as a decent proxy, putting estimates near $2,500 per kilowatt. China deliberately indigenized, licensing French and American designs, then developing domestic ones, reaching roughly 80% domestically manufactured components on the way to 100%. That cost more in the short run and was judged worth it.
- Cost uncertainty may hurt more than cost level. A utility planner can build around an expensive number but not around 50% to 100% overruns. Two proposed answers: fixed-price contracts set partway down the learning curve, the way Boeing sells early aircraft at a loss against a committed order book, and build-own-operate models where the vendor keeps the asset and sells power under a PPA, which moves the risk off small utilities entirely.
- On the NuScale cancellation, Lovering makes a point worth keeping: the communities had an opt-out and used it when the price moved from roughly $65 to $90 per megawatt-hour. It looks like failure and it is also the mechanism working, since they avoided being saddled with an expensive project.
06Claims worth citing
As of November 2023 and now nearly three years old. Nuclear has moved considerably since; verify anything before quoting.
- South Korea roughly $2,200 per kilowatt, the cheapest in the world; Vogtle and the recent Finnish project roughly $8,000. Lovering
- Under $2,000 per kilowatt is the moonshot target for competing with natural gas, translating to roughly under $60 per megawatt-hour. Lovering
- AP1000 cost breakdown: about half outside the reactor, about 12% nuclear island, about 35% engineering/procurement/construction management and owner’s costs. Sourced to a 2012 report she describes as one of the few available breakdowns, so it was already dated when cited. Lovering
- South Korea began around $4,000-5,000 per kilowatt before descending.
- China around 22 reactors under construction, estimated near $2,500 per kilowatt, roughly 80% domestic components. Lovering
- UAE’s first nuclear project: four reactors, 5.6 gigawatts, about 20% of national electricity, roughly a decade to build. Lovering
- Six or seven advanced designs then working seriously with the regulator.
- Micro-reactors under 10 megawatts, some near 1 megawatt, small enough to fit in a shipping container or two. Lovering
- Cited supporting research on smaller technologies learning faster: Wilson et al. 2020 and Schwarz et al. 2020.
07Where it’s contested
- The central US argument is one she declines to join. Advocates blame regulation; opponents cite cost. She places herself between them, calling regulation a real but non-dominant factor and adding that the industry made genuine mistakes in construction management and supply chains, and historically scaled faster than its own understanding of safety.
- Whether advanced designs will actually be cheaper is unproven and she says so. She notes the promise has been made and broken before, invoking “too cheap to meter,” and that Vogtle and the European projects went far over budget. Her case for optimism rests on modularity, factory fabrication and passive safety, all of which require being built before they prove anything.
- Design diversity cuts both ways. Kann presses the sharpest question in the episode: if fifteen designs each reach first-of-a-kind, does everyone sit at the top of the cost curve forever? Lovering’s analogy is that wide-body aircraft support only a Boeing-Airbus duopoly, and only with state support. She expects several demonstrations, some failures, and eventual consolidation, while arguing different market segments genuinely want different sizes.
- Her timeline is a forecast, and now a testable one. She predicted ground broken on several projects within five years and a handful online before 2030.