The Atomic Compact

The problem

The United States operates 94 commercial reactors. They produce about 19 percent of American electricity and roughly half of its carbon-free electricity, at a capacity factor near 93 percent, which is the highest of any generating technology by a wide margin. A reactor runs when it is dark and when the wind stops.

The country that built all of them has nearly forgotten how.

Vogtle units 3 and 4 in Georgia, the first new American reactors in three decades, were estimated at about $14 billion for the pair and came in near $35 billion, roughly seven years late. Their lead contractor went bankrupt during construction. On a per-kilowatt basis this is something like $15,000, against $2,500 to $4,000 for recent Korean-built units of comparable output in Korea and the UAE.

The usual conclusion drawn from Vogtle is that nuclear power is inherently unaffordable in the West. That conclusion does not survive contact with the history.

After the 1973 oil shock, France resolved to build out nuclear generation and did: 56 reactors in about 15 years, reaching roughly 70 percent of national electricity. It managed this by doing the single most boring thing available. It picked a small number of standard designs and built them repeatedly, with the same firms, the same crews and the same regulator, moving from site to site.

Vogtle was a first-of-a-kind unit, built by a workforce that had not built a reactor in a generation, under a design still being revised during construction, with a supply chain that had atrophied to nothing. That is not the cost of nuclear power. That is the cost of restarting an industry, paid twice, and then abandoned before the learning could be banked.

Then there is the waste, which is not primarily a technical problem and has not been for a long time. Roughly 90,000 metric tons of spent fuel sits in dry casks at more than 75 sites, most of them operating or former power plants, because the federal government legally undertook to collect it starting in 1998 and never did. Ratepayers have paid into a Nuclear Waste Fund holding on the order of $45 billion. Yucca Mountain has been stalled since 2010. Meanwhile the government pays utilities roughly $1 billion a year in damages for its failure to perform, an amount now exceeding $10 billion in total, and that money buys precisely nothing.

Finland, in the meantime, built one. Onkalo, the world's first deep geological repository, sited by asking communities whether they wanted it and compensating the one that said yes.

The proposal

Order a standardised fleet instead of one-off projects, put the licensing on a published clock, and site the waste by consent with a permanent endowment for the community that accepts it.

How it would work

1. A fleet order, not a project. The federal government contracts for at least 10 identical units of one or two already-certified designs, awarded to a single delivery consortium, with a fixed schedule of deliveries. The entire economic case rests on repetition: nth-of-a-kind cost declines are the only mechanism that has ever made reactors cheap anywhere, and they are unavailable to a country that builds one reactor per decade.

1. Design freeze, enforced. No design changes after the first concrete pour except for genuine safety findings. Vogtle's overruns trace substantially to a design still in revision during construction, which is the most expensive possible sequence.

1. A licensing shot clock. Statutory deadlines for NRC decisions on already-certified designs at pre-approved sites, consistent with № 045, with fees funded by appropriation rather than by the applicant hour. This is a guarantee of an answer within a defined period, not a guarantee of approval, and the NRC's safety authority is untouched.

1. Federal offtake, not a subsidy cheque. The government commits to purchase the output of the fleet at a fixed price for 20 years, for federal facilities and for the strategic loads in № 030, and resells the surplus. This converts the financing problem, which is that a 10-year construction with an uncertain revenue tail cannot be financed privately at reasonable cost, into a contract. If units come in below the strike price, the government profits and the receipts go to the Fund under № 023.

1. Consent-based waste siting, paid properly. Communities volunteer to host interim consolidated storage and, eventually, a repository. The host receives a permanent endowment funded from the Nuclear Waste Fund under the rules of № 009 and № 036, alongside the direct payments. Not a one-off construction boom. An income in 2120.

1. Fuel independence. Domestic high-assay low-enriched uranium capacity, because a fleet dependent on a single foreign enrichment supplier is a supply chain of exactly the kind № 030 exists to prevent.

The numbers

Ten units at 1,000 megawatts each, at Korean-comparable costs of $4,000 per kilowatt, is $40 billion. At a more pessimistic $7,000 per kilowatt, reflecting higher American labour costs and an industry still relearning, it is $70 billion.

That buys 10 gigawatts of firm, always-on, carbon-free capacity running at 93 percent, which is about 81 terawatt-hours a year, roughly 2 percent of US electricity, from a footprint measured in a few square miles.

The learning curve is the whole wager. International experience with serial construction suggests unit costs falling 20 to 30 percent between the first unit and the fifth. A fleet that starts at $7,000 per kilowatt and ends near $4,500 has produced something more valuable than the reactors: a supply chain, a licensed design, and several thousand people who know how to do it. The second fleet is then a commercial proposition rather than a federal one.

Against this: the government is currently paying $1 billion a year in damages for not having a waste repository, indefinitely, forever, in exchange for nothing at all. Item 5 costs less than the damages.

The honest objections

"Solar plus storage is already cheaper per megawatt-hour and getting cheaper. Nuclear is a solved question and the answer is no." The strongest objection and it is winning on current numbers. Utility-scale solar is remarkably cheap, and battery costs continue to fall. The counterargument is about what is being bought: nuclear supplies firm capacity through a week of low wind and low sun, which storage at that duration does not yet do affordably, and it does so on a small footprint near existing transmission, which matters enormously given the constraints in № 024. The honest position is that a grid wants both, that the marginal value of firm power rises as intermittent share rises, and that a country which forecloses the option now cannot rebuild it in 2045 when it discovers it wanted it.

"You promised learning curves at Vogtle too. American nuclear construction costs have gone up with experience, not down." Historically accurate and genuinely damning. US reactor costs rose through the 1970s and 1980s, the opposite of a learning curve, driven by mid-construction regulatory change and one-off designs. Items 1 through 3 are aimed squarely at those two causes, and if a standardised fleet under a frozen design and a fixed clock still shows rising unit costs by the fifth unit, the programme should be stopped and this proposal is wrong. That is a falsifiable commitment and it should be written into the authorisation.

"NuScale's flagship small modular project was cancelled because the power got too expensive." It was, and the price moved from roughly $58 to $89 per megawatt-hour before the subscribing utilities walked. That is the best available evidence that SMR economics remain unproven, which is why item 1 specifies already-certified designs and does not stake the fleet on small modular technology. If SMRs work, they will do so on their own merits and later.

"There is still no repository, and consent-based siting has been tried and failed here before." True on both counts, and the Yucca Mountain history, in which a site was selected by statute over a state's objection, is the reason. Finland and Sweden both succeeded with consent-based approaches, and the distinguishing feature was that the compensation was real, permanent and negotiated rather than announced. Item 5 is that lesson. It may still fail, in which case the country continues doing what it does now, which is storing fuel in casks at 75 sites indefinitely, an outcome that is safe in the near term and indefensible as a plan.

"Proliferation and accident risk." Enrichment and reprocessing are the proliferation-relevant steps, and this proposal builds domestic enrichment under existing safeguards rather than reprocessing. On accidents: the Western commercial fleet's safety record per terawatt-hour compares favourably with every other major generating source including hydro, once one counts the full toll of fossil combustion. That is not a reason for complacency about the regulator, whose independence item 3 deliberately leaves intact.

Sources

• US operating reactor fleet, share of generation and capacity factor; Energy Information Administration (eia.gov)
• Vogtle units 3 and 4 cost and schedule history; Georgia Public Service Commission filings; Westinghouse Chapter 11 (2017)
• French nuclear buildout under the Messmer Plan, 1974–1990; International Energy Agency and IAEA PRIS data (iaea.org)
• Comparative overnight construction costs for Korean APR-1400 units domestically and at Barakah, UAE
• Spent nuclear fuel inventory (~90,000 metric tons at 75+ sites); Nuclear Waste Fund balance; federal liability payments for failure to accept spent fuel; GAO and DOE reporting (gao.gov)
• Onkalo deep geological repository, Finland, and Sweden's Forsmark decision, as consent-based siting precedents
• NuScale/UAMPS Carbon Free Power Project termination (2023) and its published price trajectory
• ADVANCE Act of 2024, on NRC licensing efficiency and fee structure
• Proposals № 009 (The Permanent Fund Compact); № 023 (The Debt Covenant); № 024 (The Grid Dividend); № 030 (The Compute Reserve); № 036 (The Rural Compact); № 045 (The Permitting Clock)