How long does it take to build a nuclear reactor?
The average time is 6 to 8 years. Some can be built in under 5 years, but others have long over-runs.
“Nuclear takes too long to build” is one of the most common arguments against nuclear power. But does it really? How long does it take to build a nuclear reactor? That’s the question I want to answer in this post.
To do this I dug into the data on the construction times of nuclear reactors across the world since the 1950s. I did this from the International Atomic Energy Agency’s (IAEA) PRIS database, supplemented with additional reactor data from Wikipedia (which has an impressively curated list of more than 600 reactors).
‘Construction time’ is measured as the time it takes from the actual building to begin – the first laying of concrete – to the date that commercial operation starts.
It would also be useful to look at the length of the planning stage before construction began, but I couldn’t find this data across a large sample of reactors. Maybe it’s a question I’ll come back to later.
Here’s a quick summary of what I found:
It takes around 6 to 8 years to build a nuclear reactor. That’s the average construction time globally.
Reactors can be built very quickly: some have been built in just 3 to 5 years.
Some have long over-runs, spanning multiple decades.
The average construction time varies by country: Japan, South Korea, and China are the fastest. But, most countries with a nuclear fleet can and have built quickly, even if the average time is much higher.
I find it unlikely that large nuclear reactors will make a comeback in many countries. The best shot of avoiding large over-runs are small, modular reactors.
It takes, on average, 6 to 8 years to build a nuclear reactor
Let’s start at the global level and look at the distribution of construction times through time. This includes all reactors that came into operation by March 2023 (including those that have been shut down). I’ve plotted this on the chart.
Construction time is on the x-axis and the percentage of reactors constructed within this time is shown on the y-axis.
The median time to build a nuclear reactor was 6.3 years. The mean was a little higher – 7.5 years.
Some reactors were built very quickly: one-in-five in less than five years. Some in less than three years. The US built some small ones very quickly in the 1950s and 1960s. Its Vallecitos reactor took just 21 months: construction began in January 1956 and it came online in October the following year.
Most reactors – 83% – took less than a decade to build.
What you might also notice is that there is a long tail. Most reactors are built within a reasonable timeframe, but there are a small number that have very large over-runs. Some take decades. The longest is the second unit of the Watt bar station in the United States. It was 43 years between construction starting and commercial operation. However, it’s important to note that the project was paused midway through (because the government decided it didn’t need the power anymore) so this wasn’t 43 years of continual construction.
Overall, this picture suggests that nuclear plants don’t take a really long time to build (with a few exceptions). Most are built in 8 to 10 years. Many are built much faster.
But, this doesn’t tell us how construction times are changing over time. It’s often argued that they were built quickly in the past, but modern reactors and regulatory environments make them much slower. Let’s see if that’s true.
Average construction times have not increased much at a global level
Is the world getting slower at building nuclear reactors? I’ll use a few different charts and slices of the data to answer this question.
At a global level, not really. In the bar chart, I’ve graphed the average time of construction by decade. Each reactor is included in the decade that construction began, even if the complete date falls into a later decade. A reactor that started being built in the 1970s is included in the 1970s even if it wasn’t finished until the 1990s.
The data suggests that the world is not getting slower. Times vary a bit from decade to decade but average times are not slower than in the 1970s or 1980s.
We can also look at the distribution – not just the average – of construction times, in the same way we did before. Except, this time we’ll look at the distribution for reactors that started construction in 1990 or later.
This is shown in the chart. Again, construction time is on the x-axis and the share of reactors built within this timeframe is shown on the y-axis. I’ve included the previous distribution – for all reactors since 1950 – in light grey.
The median time for reactors built post-1990 is actually lower than for the full dataset – just 5.7 years. The mean is 6.5 years.
Reactors built after 1990 were more likely to be built faster. 58% took less than six years. 89% took less than a decade. The extremes are also missing: no reactor was built in less than 4 years, but there were no long tails either: all reactors have been completed within 15 years.
This doesn’t include reactors that are still under construction. This could change the distribution a bit; a number of reactors under construction have already hit several-year delays. As we’ll see later, the Vogtle reactors in the United States will have taken about 10 years when they come online this year. The UK’s Hinkley Point C will also take about a decade if its units are delivered in 2027 and 2028 (which is a big ‘if).
The one outlier is France’s Flamanville. If it comes online in 2024, as planned, it will have taken around 16 years. That will make it the longest build of any country post-1990, so far.
The final way to visualise this data is to plot construction dates over time. Each dot is one reactor, and these are coloured by region.
Again, we see no clear increase in construction times. What’s noticeable is the regional shift. Europe and the US were building a lot in the 1950s, ‘60s, and ‘70s. But have not built much recently.
Countries in Asia dominate nuclear reactor builds today – see the dominance in red dots since the 1990s. Maybe, then, construction times globally are not increasing because countries such as China and South Korea are building quickly.
Let’s take a look at how build times vary by country.
Which countries are fast- and slow- builders?
Japan has been the fastest builder.
In the table I’ve shown the median and mean build times for some countries with large nuclear fleets. They’re sorted by median time, from fastest to slowest. On average, it took the Japanese less than 5 years to build one. In South Korea and China, less than 6 years. East Asian economies build fast.
France, the UK and the US are not that far behind – it took them a few years longer, on average. Note that most of these were built several decades ago. We’ll come on to newer builds soon.
It has taken India 10 years, on average.
However, the maximum and minimum times also tell an important story. India’s average is high, but it can build quickly: its two units in Tarapur took just 5 years. Those were built in the 1960s, but it has a more recent success too: the third unit of its Kaigai fleet took just 5 years in the early 2000s.
And France, the UK, and the US have all built quickly in the past too.
What’s also clear is that all countries have had plants with significant over-runs. Even Japan had plants that took almost a decade (which is double its average time).
In conclusion: Japan, South Korea, and China are faster than the Europeans and Americans. But most countries can build quickly when they want to. And all countries are at risk of significant delays.
I’ve not yet answered the question, though, because this is looking at average build times since the 1950s. That means it’s comparing US builds in the 1970s with Chinese and Korean builds in the 2000s. It doesn’t tell us about the different regional speeds over time.
The problem here is that we just don’t have much recent data for the US, the UK and France. They haven’t built much. That could be a signal in its own right: maybe they’re not building because they don’t have a lot of faith that they can build time and cost-effectively.
But we can look at the build times of a few reactors that are still under construction. In the US, two reactors – Vogtle Units 3 and 4 – are due to come online this year. Unit 3 in May or June, and Unit 4 is estimated for the end of the year. Construction of both started in 2013, and I estimate that their construction time will be 120 and 122 months (about 10 years). That’s longer than the US average, although faster than Vogtle Units 1 and 2 which were built in the 1970s and 1980s. Those took 130 months.
The UK is building Hinkley Point C – two reactors that started construction in 2018 and 2019. Their projected timeline is to be online in 2027 and 2028, respectively (which is several years later than originally planned). That would give them a construction time of 110 to 120 months, if they are on-schedule, which is a big ‘if’. Again, that’s longer than the UK average but definitely not its slowest build to date.
Finally, we have France’s Flamanville, which makes a mockery of its nuclear dominance in the 1970s. Construction started at the end of 2007. It’s not estimated to come online until 2024. By the end of 2024, it will have taken around 200 months. Its longest build to date.
What explains differences in construction times?
It’s not clear why the construction times of reactors vary so widely.
It’s not about the size or the type of the reactor – at the end of this post I show that they have little impact on construction times.
What should also be clear is that nuclear energy as a technology is not inherently slow: we know we can build it quickly (and safely). That means it’s the political and economic context that matters most.1
One argument I find fairly convincing is that countries build quickly when they have to. When the country’s energy demands are rising, they need to deliver electricity. In the 1960s and 1970s, electricity demand was rising quickly in France, the UK, US and other rich countries. They built quickly because delays meant blackouts. Countries such as China and South Korea have been in that position more recently2 The urgency is not the same across Europe and the US anymore (other than the urgency of decarbonising their electricity grids…) which might explain why they build very little nuclear and when they do it is slower than it used to be.
Another explanation is that there are now intense regulatory hoops and costs to jump through. This seems reasonable to me, but I’d be surprised if it was the only explanation.
A final explanation is that these countries have regressed in their ability to build stuff more broadly. Time over-runs are not just an artifact of nuclear plants. When we build the same reactors in succession, we learn from the process. The next one becomes easier and quicker. Several decades out-of-the-game surely comes at some experience cost.
Modular reactors could be built much faster, and reduce the risk of large projects
I recently read the book How Big Things Get Done by Bent Flyvbjerg and Dan Gardner.
They argued – quite convincingly with data on thousands of projects – that nuclear reactors have some of the largest cost and time over-runs. They are defined by long tails, which we saw earlier. That is a symptom of big projects in general.
Long and uncertain construction times are a big risk for energy planners. It matters whether a big chunk of your energy grid will come online in 5, 10, 15, or 20 years.
The other downside to big projects is that there are fewer opportunities to ‘learn’. Many emerging technologies follow a consistent pattern where their costs fall and construction improves as we build more. We learn as we go. This is called a ‘learning curve’. We see it for solar PV and batteries, for example.
The problem with big projects is that you have fewer data points to learn from. If you’ve only built a handful of nuclear plants – or haven’t built them in decades – the amount of learning you can do is limited. This is even worse when you build different reactor designs each time, which is not uncommon. To build quickly, you need to pick a standardised design and implement it over and over.
Jessica Lovering and Jameson McBride found that ‘learning rates’ for nuclear reactors were positive in France in the 1950s and 1960s, and in South Korea more recently. In both cases the reactors designs were standardised. They learned by doing the same thing over and over. In most other countries and periods, learning rates were negative. This data focused on costs but it is probably true for construction times too.
In short, big projects are riskier than smaller ones, and they give us far fewer opportunities to learn how to build them better.
That’s why the future of nuclear power could be modular reactors. These are smaller than conventional reactors, and are often built off-site in a location such as a factory. They are then transported and operated at a separate site.
The amount of electricity you generate will be smaller than a large plant, but you just build more of them. A delay in one small part of a large nuclear construction can hold the whole thing up. With modular designs, each unit is independent of the rest – you can keep building the others if one is put on hold.
It also means we would build thousands (or millions) of them. That gives us lots of opportunity to learn how to do better: this gives them a shot of a better ‘learning curve’ than the large nuclear designs we’ve had in the past.
If nuclear fission technology is to have a renaissance, it’s going to be modular. Some countries are still building big nuclear, but others – such as the UK and the US – have turned their backs on them. The bad press and longer construction times of recent reactors in these countries make a comeback even less likely.
Additional notes
Does the size (capacity) of the plant matter for construction times?
I thought that to do these comparisons I would need to convert construction times to months per unit of power added. I thought that the size of the nuclear reactor would matter a lot: the biggest reactors would take much longer.
But I looked at construction times versus the capacity of each reactor and found no clear pattern. Many medium-sized reactors took much longer than big ones. We see this in the chart.
Unless we’re talking about very small reactors (which ventures towards the ‘modular’ category), the size of reactor doesn’t have much of an impact on construction time.
How does build time vary by reactor type?
Similarly, I thought the type of reactor design might make a difference. Maybe pressurised water reactors would be faster than boiling water ones (or vice versa).
I looked at the data and found no consistent pattern. I mostly found that boiling water reactors used to be popular and have gone out of fashion. Pressurised water reactors have been the most common since the 1990s.
Breetz, H., Mildenberger, M., & Stokes, L. (2018). The political logics of clean energy transitions. Business and Politics, 20(4), 492-522.
Lovering, J. R., & McBride, J. R. (2020). Chasing cheap nuclear: Economic trade-offs for small modular reactors. National Academy of Engineering, 50(3), 38-44.
The construction of the World's first grid scale nuclear power station - Calder Hall in North East England - was commenced by Taylor Woodrow Construction in 1953 and was officially opened on 17 October 1956, using 1950s engineering and construction techniques..
Originally designed for a life of 20 years from respectively 1956-1959, the plant was after 40 years until July 1996 granted an operation licence for a further ten years.
The station was closed on 31 March 2003, the first reactor having been in use for nearly 47 years.
Seems to me we have gone backwards.
Always love your work. Thanks for injecting empirical reality into what is too often a political discussion.