How the UK could build 3 gigawatts of new nuclear power by July 2027

Building new nuclear power in the UK is deeply broken—it is slow to approve; too slow to build, too expensive, and all too often asks for state subsidy. This is not the state of nature: other countries build reactors many times more quickly and cheaply, and the UK used to be able to. Since the UK last added a nuclear reactor, it has turned off 26 reactors and the rest of the world has added 148.

The Government has recognised there is a problem, and announced their intention to “rip up the rules to fire up nuclear power”. To build nuclear power as quickly as possible; smaller is better. Giga-scale reactors would take a minimum of 5 years, Small Modular Reactors (roughly 500 to 100 megawatts) take 2 to 5 years, but Micro Modular Reactors (from 1 to 50 megawatts) can be built in less than 2 years. I’ve spoken to developers who say this would be possible, but others have expressed scepticism. Setting this aside and being charitable to the companies; what would the Government need to do to make this possible?

Working backwards, the fabrication of micro nuclear reactors would need to begin at the end of this year; and so the three pillars of nuclear approvals — licensing, permitting, and planning — would need to be reformed to allow this. This would probably require primary legislation within the next three months.

What would that primary legislation have to do?

1. Create a regulatory sandbox, administered by the Nuclear Regulatory Taskforce, with authority to license Micro Modular Reactors. The sandbox would need to:

   • Give the licensing team who run the sandbox permission to decide which conditions are proportionate for micro reactors. There are 36 license conditions and 909 goals in the UK’s current Nuclear Site License process, and there is no success criteria. The sandbox would allow for ‘technology-first’ approvals; which first consider the reactor’s design basis and decide on the suitability of other principles.

   • Align the Basic Safety Objective to the same level as background radiation in Cornwall. This would not change the developer’s responsibility to minimise radiation, but it would stop requiring paperwork once the developer has proved that being next to the reactor if it were damaged, is safer than living in Cornwall.

   • Remove cost recovery mechanisms for the regulator. The current system has bad incentives for the regulator to extend ‘pre-application consultation’.

2. Grant planning permission and replace environmental permits for Micro Nuclear Reactors, within designated areas, provided specific environmental conditions are met, and neither the Secretary of State nor the local planning authority objects within a specified time.

3. Incorporate the ‘regulatory justification’ — which currently sits within the Department for the Environment, Food, and Rural Affairs, and takes two years — into the planning decision.

Licensing reform

The best way to regulate nuclear reactors is a goals-based approach. Rather than the regulator specifying how the reactor needs to be made safe (“rules-based”), the developers just have to prove that their reactor is safe. While the UK has a goals-based approach in theory, it doesn’t work like this in practice. The regulator doesn’t set out criteria for meeting these goals in advance, and has such narrowly specified success criteria, based on what they are already familiar with, that it is de facto rules-based. For example, at Hinkley Point C the regulator required EDF add an all-analog quadruple-backup to the control room (as in, four sets of spare equipment) despite other international nuclear regulators deeming one digital backup to be sufficient. In total, the regulator required that EDF make 7,000 design changes to a design that was already operational in France and Finland. This is de facto rules-based, without specifying the success criteria.

Why did this happen?

1. The regulator is only incentivised to prevent risks from nuclear reactors, not to balance the costs and benefits of nuclear power construction.

The regulator’s website lists its mission as “to protect society by securing safe nuclear operations”. This is no expectation that it will promote, enable, or ensure the development of nuclear power. Its five statutory purposes¹ are all risk-based, meaning their goals could, in theory, be achieved without any nuclear activity at all.

There is no positive force, pulling towards nuclear getting built, to act as a counterweight. At no point is the full cost-benefit analysis happening—considering whether the benefits of additional regulation (to ‘safety’) outweigh the costs of it becoming prohibitively difficult to build.

2. The regulator has an expansive mandate and there is no oversight.

This might sound like an overstatement, but quite literally, Clause 78 of the Act which created the regulator says that the ‘Principal Function’ is that:

“The ONR must do whatever it considers appropriate for the ONR’s purposes.”

The nuclear regulator sits within the Department for Work and Pensions, so it is hardly reasonable to imagine that the Secretary of State—otherwise busy with their responsibility for all benefits and the state pension—would provide suitable oversight to regulator’s performance. In South Korea, where they build nuclear cheaply and quickly, the Nuclear Safety and Security Commission reports directly to the Prime Minister.

The combined effect of this incentive misalignment and expansive mandate means that companies would reasonably struggle to get the regulator to be proportionate. In the regulator’s response to EDF publicly saying that they were required to make 7,000 design changes to Hinkley Point C, they said:

“EDF and AREVA did not make any arguments of gross disproportion during or after the [Generic Design Assessment].”

To whom were EDF supposed to complain if the regulator was being grossly disproportionate? The Work and Pensions Secretary? The regulator clearly holds all the cards, and so the developer is incentivised to go along with any changes they ask for, lest it damage their chances of getting a licence.

3. There are no recent successful UK nuclear projects to provide a model for goals-based regulation.

The aforementioned reasons are compounded by the fact that the UK has not built a new reactor in 30 years. There are no examples of what constitutes meeting the contemporary set of goals, and the precedents from Hinkley Point C is evidently a bad guide.

This means we have to turn to case law precedent. In the UK, the law is that a marginal safety feature must be added, unless it can be deemed to be ‘grossly disproportionate’ in costs relative to benefits. This is operationalised as when the costs to the reactor outweigh the benefits by a factor of 10 to 1.² In practice, this means the regulator would need to simulate the view of the judge; and as is in line with their incentives, will likely default to adding more features.

By comparison, South Korea builds reactors in fleets (many at a time) which means they have very clear examples of what constitutes passing the regulations.

In aggregate, these factors lead to ‘goals-based’ regulation morphing into a proscriptive and ever-ratching regulatory regime that prevents new construction. Consequently, in the last 35 years, the UK has decommissioned 30 reactors and added just one. While it is no individual’s fault, the nuclear regulator’s core competency has become shutting down reactors, not licensing them.

How do we fix this?

We need a reset; to wind back the ratchet.

1. Give the licensing team who run the sandbox permission to decide which conditions are proportionate for micro reactors.

There are 36 license conditions and 909 goals that developers need to prove to get a Nuclear Site License. Prima facie, one might expect these will be about reactor design but they can often be organisational. For example, goal 59 is:

“The value of safety as an integral part of good business and management practice should be reinforced through interactions between directors, managers, leaders and staff, including contractors, to establish a common purpose and collective social responsibility.”

And license condition 3(1) is:

The licensee shall make and implement adequate arrangements to control all property transactions affecting the site or any part of the site to ensure that the licensee remains in overall control of the site.

(Translation, if I understand correctly: prove you won’t accidentally sell the site.)

The regulator has said, “[W]e don’t license technologies. We license organisations to undertake a nuclear activity on a particular site.” There is some rationale to this approach, for example, it is important that the operator is capable of competently refueling their reactor. However, the organisational and site requirements would differ greatly if Radiant—who make a 1 megawatt reactor—wanted to get a license in the UK, compared to EDF building two 1650 megawatt reactors at Sizewell C.

To account for this, the licensing team should do technology-first approvals, where they first consider the design basis (as in, what the reactor is actually going to do), and then decide which of the 909 goals need to be verified based on this. At the moment, different aspects of the license application proceed in parallel, without considering whether the reactor design requires this. Taking a technology-first approach would support safety in practice rather than a box checking approach.

2. Align the Basic Safety Objective to the same level as background radiation in Cornwall.

The Basic Safety Objective (BSO) is the point where the regulator considers that, “beyond which further consideration of the safety case would not be a reasonable use of [the regulator’s] resources”.³ In the UK, this is set to 0.02 mSv of radiation per year.⁴ This is a wholly unscientific point to choose—it is about the same as an annual roundtrip from London to New York, or eating 5.4 bananas per day for a year—it is just a quirk of the rules.

We should move the burden of proof to the Cornwall Standard.⁵ This would mean that—under the most unfavourable assumptions of a reactor in meltdown—the question regulators would have to answer is, ‘Would people receive more radiation over the course of a year, than they would receive from being in Cornwall?’

Past this threshold, it should be deemed that it is a waste of public money to prove any further. This does not change the developers’ responsibility to minimise radiation; it just means the regulators stop evaluating data. The Cornwall Standard would still be extremely conservative—by nearly a factor of 100 from my conversations with experts—on the effects on ionising radiation, but it would still be a lot of progress from the current standard.

3. Remove cost recovery mechanisms for the regulator.

At present, the regulator can charge companies for ‘pre-application engagement’ and companies cannot enter licensing until the regulator allows (known as being ‘license-ready’). The pre-application engagement is private, but as an example of this kind of gating; here is a quote from Rolls Royce’s Generic Design Assessment:

“Rolls-Royce SMR Ltd should: Demonstrate the adequacy of their organisational arrangements to support the development of the E3S case for GDA. This should include roles and responsibilities, relevant processes, governance and oversight of the case.”

(Translation, if I understand correctly: prove that your company is able to write documents.)

This might go some way to explaining why Rolls Royce—a 52 billion-pound company—still needs to receive £210 million in taxpayer subsidy for its SMR. Both sides of this regulatory engagement are funded by the taxpayer, and so the cost vortex is being sustained whilst both sides think that ‘the other’ is paying.

Instead of continuing with this broken system, we should end ‘cost recovery’ to align the regulator’s incentives with moving at pace to license reactors, not gate access to the licensing process; and perhaps the need for subsidising the approval process will go away.

Planning and environmental permission

Why is reform necessary?

The time from initial consultation to starting construction at Sizewell C took 11 years, but for the same basic reactor in France, it took just two years.⁶ During this time, there were four rounds of public consultation and the Environmental Impact Assessment produced was 44,260 pages long. This is clearly broken. To achieve our goal by 2027, we would need to create a new mechanism for planning, as even a Development Consent Order takes two years and still faces substantial risk of Judicial Review.

What do we need to do?

1. Give conditional planning permission to specific sites in the legislation, provided that specified environmental conditions are met, and neither the Secretary of State nor local planning authority objects. This would take inspiration from the ‘Renewable Acceleration Areas’ created in Spain and Germany, to substitute for the Environmental Impact Assessment. We have written about this previously here.

2. To ensure the new power plants have a positive environmental effect, require generous contributions to the Nature Restoration Fund. This would actually improve the efficacy of environmental mitigations: the assessment for Hinkley Point C required that EDF build an ‘acoustic fish deterrent’ with 288 underwater speakers to prevent about a trawler’s worth of fish in total from being drawn into the water pumps.
   
   This is £100m-bat-tunnel-levels-of-ridiculous and it would clearly be more efficacious to allocate this money to preserving fish populations elsewhere.

3. Allow the local authority to keep 100% of the business rates from the new power plant to align their incentives with construction.⁷

4. Allow the developer to make payments to people who live near the reactor to compensate for the inconvenience of construction.⁸

Regulatory Justification

‘Regulatory justification’ — from the Justification of Practices Involving Ionising Radiation Regulations 2004 — requires that the benefits of using ionising radiation must outweigh the costs. The practical application of this in the UK, each new nuclear reactor design must show the benefits outweigh the costs, rather than saying that ‘nuclear power’ overall must outweigh the costs. The Department for the Environment, Food, and Rural Affairs takes two years to provide a decision for each reactor, which entirely duplicates the planning process. (What is planning, if not to consider whether the benefits outweigh the costs?) Both France and Germany incorporate regulatory justification, which stems from a 1996 EU directive, into their planning process. We have written about the need to reform regulatory justification before, as have Lexology, Britain Remade, the Tony Blair Institute, and UK DayOne.

Regulatory justification should be incorporated into the planning verdict.

What remains?

We have just considered what would need to be true from a planning, permitting, and licensing perspective; to enable 3 gigawatts of micro nuclear reactors by 2027. There are still other considerations—as we noted at the beginning, whether the developers are capable of delivering this; or whether the fuel supply or supply chain of skill would be able to make it in time. One concern I would not have is whether there is commercial demand for this power—I find it essentially impossible to imagine that demand for this electricity would not respond to these changes, as there is a trillion dollar wave of capital expenditure for AI that is principally bottlenecked by access to energy. Demand is elastic to the boldness of reforms.

So then, what is scarce? I would suggest the most scarce resource is urgency and political will. The UK was capable of getting a vaccine in under a year, and I see no reason why the same should not be true for building nuclear power by the middle of 2027. With regard to economic growth, if we had grown at 2% since 2008, and then fallen to our current level, it would be a drop tantamount to the Great Depression. The UK is deindustrialising because it has the highest industrial electricity prices of any country measured by the International Energy Agency, exceeding the US by a factor of 4. With regard to AI, there will be models that match human-level capabilities within the next 5 years, with an effective explosion in the cognitive workforce. And we have committed to be 95% net zero by 2030.

A good litmus test is to imagine that OpenAI wanted to build a nuclear reactor in the UK. Sam Altman has written about Greg Brockman, his cofounder, that “an average email response time of about 5 minutes to anything”, and Sam has said previously that he has written a script to see how quickly the billion dollar founders of tech companies respond to his emails versus “bad founders”; he notes, “It was a difference of minutes versus days on average response times.” The important question to ask is: does the regulator match the operating pace of companies who want to build in the UK? Currently, an industry source tells me that it can take months to get a meeting with the regulator.

To make things more concrete, people at the AI labs would respond to an email at 10pm on a Sunday; they’d work 60 hours a week; and they’d work directly from the office of their counterparty, if they needed to, until the work was done. It seems worthwhile to consider what it would take for the state to share this level of intensity too.

Slowness is a policy choice.

1. The statutory purposes are: nuclear safety, nuclear site health and safety, nuclear security, nuclear safeguards, and radioactive transport safety.

2. HSE principles for Cost Benefit Analysis (CBA) in support of ALARP decisions. Note this source is not nuclear-specific, but the ALARP principle applies across industries.

3. 701, Safety Assessment Principles, Office for Nuclear Regulation, January 2020.

4. Note that this is for members of the public. For workers, the level is 0.1 mSv.

5. According to the Nuclear Decommissioning Authority, the average background radiation in Cornwall is 7.8 mSv per year.

6. While the pre-construction phase for Flammanville 3 took just 2 years, this should not imply that construction was also quick: all EPR construction has been very slow. Construction at Flammanville took 16.5 years. At Olkiluoto 3 in Finland; approval took 4.5 years, and construction took nearly 18 years. At Taishan, in China, the first EPRs were projected to take 3 years and 10 months to build, but actually took 9 years and 10 months.

7. Inspiration is taken from the authors of Foundations, in their piece on datacentres.

8. Inspiration is taken from the ‘Street Votes’ housing policy and from Looking For Growth.