Nuclear and SMRs
Palisades, Three Mile Island and Duane Arnold reopening for buyers who are new to the power business, and why SMR announcements are in gigawatts while operating reactors are in single units.
Words the industry uses are marked like this at the sentence that defines them, so you can tell a term you will hear on a desk from a phrase this book happens to be using.
In 2024 a software company agreed to buy the entire power output of a nuclear reactor that had been shut down, at Three Mile Island, the most notorious address in American nuclear power. Microsoft took all 835 megawatts of Unit 1 under a deal with Constellation worth around 1.6 billion dollars, and the station is being restarted as the Crane Clean Energy Center.
Palisades in Michigan became the first American reactor to return from retirement, supported by a federal loan guarantee of up to 1.52 billion dollars. Duane Arnold in Iowa is following under an arrangement between NextEra and Google.
A generation of plants closed on economics is being reopened by buyers who are new to the power business.
Why nuclear specifically
Chapter 25 described a load that runs continuously, cannot easily be moved, and belongs to companies with public commitments about carbon.
Solar and wind are cheap and match a 24-hour load only with storage that Chapter 5 showed does not extend to seasons. Nat gas is dispatchable and carries emissions. Nuclear is the only widely available generation that is firm, dispatchable and effectively carbon-free at the same time, and it produces at a flat output that happens to match a data centre's consumption profile almost exactly.
Nuclear restarts are also relatively cost efficient. Reviving an existing station means recommissioning equipment, requalifying staff and satisfying the regulator, at a cost measured in the low billions for something close to a gigawatt. Building new costs many times that per gigawatt in Western countries. Restarts are the cheapest firm clean megawatts available anywhere, which is why they went first.
They are also finite. The number of recently shut reactors in recoverable condition is small, and after Palisades, Crane and Duane Arnold the list is close to exhausted.
The half that has not happened yet
Small modular reactors (SMRs), a technology deployed in submarines and aircraft carriers for decades, are the nuclear answer everybody reaches for. There is, however, a gap between press releases and SMR hardware.
On commitments, software companies have collectively contracted for more than 10 gigawatts of new nuclear capacity. Amazon, for example, is investing around 500 million dollars into X-energy, targeting up to 5 gigawatts by 2039.
There are four ways for nuclear projects to be permitted. First, a US NRC (Nuclear Regulatory Commission) design approval clears a design generically, to be built afterwards at any suitable site, and NuScale holds the only SMR one, uprated to 77 megawatts electric per module in May 2025. Second, a construction permit takes the older route of licensing one plant at one site, and needs no approved design at all: TerraPower’s Natrium at Kemmerer in Wyoming received the first one ever issued to a commercial Generation IV reactor, on 4 March 2026.
Third, reactors built on Department of Energy land are authorised by the department rather than licensed by the NRC, which is a research pathway rather than a commercial one, and that is how Oklo’s Aurora and Aalo’s Aalo-X are under construction at Idaho National Laboratory. Both companies intend to go to the NRC afterwards for commercial units. Fourth, the reactor may not be in the United States at all. GE Hitachi’s BWRX-300 at Darlington in Ontario, Canada, the first SMR under construction in North America, is regulated by the Canadian Nuclear Safety Commission, which licenses a plant at a site and has no separate design approval step to be first through. China’s Linglong One is on track to be the first land-based commercial SMR anywhere.
The announcements are denominated in gigawatts and the operating reactors are denominated in single units. The distance between those two denominations is the entire SMR question: the world needs gigawatts, not just megawatts.
The argument, on both sides
The case for SMRs rests on manufacturing. A large reactor is a construction project executed once, on site, by a workforce assembled for it, which is how two conventional large reactor projects, Vogtle in the US state of Georgia and Hinkley Point C in Somerset, England, arrived years late at multiples of their budgets. A small reactor built repeatedly in a factory should behave like a manufactured product, with a learning curve, standardised regulatory review and predictable cost.
The case against is that this argument has been made before. Nuclear construction is one of the few industries to have exhibited a negative learning curve, becoming more expensive per unit as more was built, in several countries and across several decades. Smaller reactors also give up economies of scale directly, since the cost per megawatt of a reactor generally falls as it gets larger, so the manufacturing gains have to outweigh a real penalty before the first unit is competitive.
The question of SMR scaling requires somebody to build several dozen units of one design. That has not happened in the West, and the first genuine evidence will come from Darlington in Ontario, Canada, and from China.
For a reader tracking this, the number to watch is the cost of the fourth and fifth units of a design rather than the first. The first unit tests whether the thing works. The fourth tests whether the SMR cost economics scale.