The US nuclear groundswell

Kann picks the word groundswell deliberately, because a groundswell sometimes becomes a wave and sometimes does not, and the United States has completed exactly one new nuclear project in decades. Colbert, who was recruited into the 2007 renaissance and watched it fail, locates the difference in what broke last time. Then, regulated utilities were being asked to underwrite $17 billion to $20 billion projects against market capitalizations of $20 billion to $30 billion, on the twin expectations that demand would grow and gas would stay expensive. Neither happened. Today gas prices are near all-time lows, which by the old logic should be fatal, and yet demand is there anyway, because hyperscalers want reliable, clean baseload contracted for a long time.

The load-bearing change is not the offtake contract itself. Nuclear development costs hundreds of millions of dollars before a final investment decision, against tens of millions for renewables, and no power purchase agreement funds that. What took a year and a half to sink in with the hyperscalers, on Colbert’s account, is that they had to put money into development, not just into technology, and the Google deal is his example of that money arriving at the front of the process. That reframes the entire business. Elementl inverts the renewables development sequence: find the customer first, then the site, then work with the utility the customer already knows, then choose the technology last. An early site permit is technology-agnostic, so the site can be proven up while the vendor field sorts itself out, and only the later construction permit is design-specific.

Which is convenient, because Colbert’s honest position on vendors is that nobody has arrived. Billions have gone into both Generation III lightwater designs and Generation IV designs such as high-temperature gas and sodium-cooled reactors, and all are near the end of development while each still solves for something, whether supply chain, licensing or cost estimation. He cannot say which will succeed and says so plainly, betting only that three or four at minimum will. He does separate technology risk from execution risk sharply: lightwater is 50 to 60 year old technology, the Navy has run it since the fifties, and the EPR, APR1400 and AP1000 all struggled to the finish line and then worked. So for lightwater there is, in his framing, no technology risk left, and the challenge is entirely in de-risking the front end and executing in the field.

On cost his thesis is repeatability, and his analogy is assembling four identical pieces of IKEA furniture, where the first one breaks and the fourth is quick. He points to Vogtle unit four coming in roughly 40% below unit three. This is why the Google deal is three projects rather than one, and why the reactor size lands between 75 and 300 electric megawatts rather than gigawatt scale: you cannot write a check big enough to order enough large reactors to get repetitions, so smaller units buy more swings at the bat. Underneath sits the unglamorous second level, meaning a finished design, a supply chain lined up and above all a labor plan. Labor is the input he calls most uncontrollable, since crews can walk to the data center project down the road, and his answer is to pay the premium rather than eat the delay. One siting inversion is worth carrying too. Colbert expects it to be easier to permit and build a nuclear plant than a new transmission line, everywhere he has worked, because transmission crosses people’s property without leaving economic benefit behind. So sites go where transmission already exists, meaning former coal plants and large industrial sites, and finding sites is not the hard part. The customer decides where, and hyperscalers choose more by which utility they trust than by geography.