Why Does Building a Nuclear Reactor Take So Long?

How long the recent builds took

Dates are taken from operator announcements and the World Nuclear Association reactor database. Hinkley Point C’s 2030 operation date is EDF’s expectation, not an outcome or a confirmed first-grid date. Barakah elapsed times are approximate unit-level grid-connection intervals.

The United States: a skill that was allowed to lapse

In the early 1970s, U.S. reactors cost about $1–2 per watt to build in today’s money. By the late 1980s, the same kind of plant cost $4–10 per watt. Regulation tightened after the Three Mile Island accident in 1979, but the MIT analysis of the period found the bigger story was on site: containment buildings cost twice as much because workers installed less per hour, and later plants were more expensive than earlier ones. Then construction stopped. When Georgia restarted with the AP1000 in the 2010s, its crews, suppliers and regulators were all learning the design for the first time.

Vogtle’s cost works out to about $15,700 per kilowatt including financing, roughly two and a half times its original budget. The reactors now run reliably and supply carbon-free power to millions of homes. The lesson is not that nuclear cannot be built in America; it is that the first unit of a new design, built by a workforce without practice, carries the entire learning cost.

France: from the fastest builder to the slowest

France built most of its fleet of around 56 reactors between the late 1970s and the 1990s, standardising a few designs and building them in series. That programme still supplies most of France’s electricity. A study of construction costs across seven countries found that France and South Korea avoided the steep cost escalation the United States experienced after 1979.

Then France paused for more than a decade and began its next reactor, Flamanville 3, with a new and more complex design. It took 17 years to connect to the grid and, by the audit office’s count, €23.7 billion in 2023 euros including financing; its initial €3.3 billion construction budget had a different price year and scope. France’s next step, six EPR2 reactors estimated at €72.8 billion, is designed explicitly to rebuild a series: one simplified design, three sites, a state loan covering at least half the cost and a 40-year contract for difference.

China: the learning curve kept

China connected its first reactor in 1991 and has built steadily ever since, approving around ten new reactors a year between 2022 and 2025. Its nuclear capacity nearly doubled in the decade to 2026. The Hualong One is now built in about five years, and China’s state planners approved eight more units in 2025 alone. The mechanism is not cheap labor alone. Researchers point to a domestic supply chain, the same few designs repeated, stable rules and low-cost state finance: once domestic content passed roughly three-quarters, Chinese and Korean costs stopped rising.

Chinese cost figures come from the state developer and are not independently audited. They are best read as evidence that a series build can hold costs steady, not as a precise international benchmark.

Take a plant that costs $10,000 per kilowatt to build, spread evenly over ten years at an 8% cost of capital. By the time it starts running, interest has added about $4,500 to every kilowatt (Derived). Build the same plant in five years at 4%, and the interest is about $800. The physical plant is identical; the electricity is not. This is why the countries that build cheaply also tend to finance cheaply: governments that lend at low rates and guarantee demand turn a risky project into something closer to public infrastructure.

Delay creates more delay

A construction delay is more than an extra month of wages. An unfinished drawing can hold up a component, which holds up installation, which makes an inspector’s planned visit useless. The site still pays for supervision and temporary facilities. Work is resequenced, access becomes harder, and completed work may need to be opened again. This is the feedback loop that turns a design problem into a productivity problem and then a financing problem. MIT’s historical analysis identifies indirect costs and declining on-site productivity as important drivers; it does not establish that regulation or reactor complexity is irrelevant. Eash-Gates et al., Joule, 2020

There are also several clocks. Site preparation, first nuclear concrete, first grid connection and commercial operation are different milestones. A reactor can finish its physical construction and still spend months commissioning equipment and demonstrating safe operation. Comparing China’s first concrete with another country’s investment decision adds years before comparing any construction skill. For financing, however, the clock begins when money is spent, even if the construction-time table starts later. EDF’s February 2026 project update

The feedback loop is editorial synthesis. International project costs differ in scope, price year, currency and financing; the chart is a comparison of reported figures, not a controlled experiment.

A repeat order matters more than a smaller reactor

France’s six-unit EPR2 plan is an attempt to buy continuity: three pairs at Penly, Gravelines and Bugey rather than isolated projects. EDF’s €72.8 billion estimate is in 2020 euros. The government’s March 2026 statement targets an investment decision before the end of 2026 and first service by 2038. That is a programme and a target, not evidence that the learning curve has already returned. The test is whether later units use fewer labour hours and have fewer defects under comparable accounting. EDF’s EPR2 programme; French nuclear policy council, March 2026

Darlington gives small modular reactors a concrete test. OPG budgets C$20.9 billion for four 300 MW units, including interest, escalation and contingency: about C$17,400 per kilowatt, before any currency conversion. It separates C$6.1 billion for the first reactor from C$1.6 billion of shared systems. Allocating all shared infrastructure to unit one would exaggerate its cost relative to its successors. SMRs could shorten schedules and make repetition easier, but factory investment needs orders, and smaller reactors lose some economies of scale. The decisive evidence will be delivered units and audited costs. OPG’s first-quarter 2025 budget

Darlington’s cost per kilowatt is Derived from the announced budget and capacity, in Canadian dollars. It is a forward budget, not a completed-project cost.

Who is building what

National programmes, reactor vendors and financing models that are trying to turn single projects back into series. Search the record, or filter by approach.

Announced capacity and signed agreements are not construction. A project counts as progress on this curve when nuclear concrete is poured, and as proof when the second unit of the same design is cheaper and faster than the first.

Four numbers to watch

The best evidence will not come from announcements but from construction sites. Watch, first, the number of reactors of the same design under construction in each country at once: learning needs a pipeline. Second, the months from first nuclear concrete to grid connection for each successive unit. Third, the cost per kilowatt of second and later units compared with the first, which is the only honest measure of a learning curve. Fourth, the financing structure: the share of each project funded by state loans or regulated returns, which can matter as much as engineering.

China’s 36 reactors under construction already provide the first of these. The United States and Europe will show theirs over the next five years.