Nuclear, in units small enough to finance

Nuclear power has an excellent safety record, produces no carbon during operation and runs continuously regardless of weather. It also has a construction record in Western countries that is genuinely dismal: projects delivered years late at multiples of their budget, several of which contributed to the near-collapse of their builders.

The diagnosis behind small modular reactors is that the problem is not the reactor, it is the project. A gigawatt plant is a bespoke construction site with tens of thousands of workers, first-of-a-kind engineering, and a decade of interest accruing before the first kilowatt hour is sold.

The theory of the case

Build a reactor small enough to be manufactured in a factory and shipped to the site. Factories have learning curves, quality control and repeatability that construction sites do not. Build the tenth unit and it is cheaper than the first. Financing improves because the capital is smaller, arrives later and returns sooner.

Small reactors also allow simpler safety. Below a certain power density, decay heat can be removed by natural convection with no pumps and no external power, which removes the failure mode behind the most serious accidents. That is a genuine engineering advantage and it is not marketing.

The argument against, which is equally strong

Nuclear economics have always been about scale. A reactor a tenth the size needs a control room, a containment, a security perimeter, an operations crew and a regulatory licence, none of which shrink proportionally. Cost per megawatt therefore starts much higher, and the entire case depends on factory learning outweighing that penalty.

Nobody has built enough units to know. That is not scepticism; it is the state of the evidence.

The cautionary case is instructive: one prominent American project was cancelled in 2023 after its projected cost rose to a level that lost its customers, before construction began. The failure was economic, not technical.

Regulation, which is the real timeline

Nuclear licensing frameworks were written for large light-water reactors. A design with passive cooling, a different coolant or a different fuel form does not fit the templates, and every deviation requires new analysis, new precedent and new review. Several regulators are building pathways for advanced designs; those pathways are themselves years of work.

A design certified in one country must generally be recertified in another, which undermines the factory-product argument at exactly the point where volume would come from.

Why technology companies are signing contracts

Data centers need firm power, they need it in large blocks, and they are willing to sign twenty-year agreements at prices above market. That is exactly the customer nuclear has lacked for forty years: one that values reliability over cost and can sign before construction.

Read those agreements carefully. A power purchase agreement for delivery in the 2030s from a design not yet licensed is a funding commitment that helps a developer raise capital. It is not megawatts anyone is counting on this decade.

What would actually prove it

One thing: a second and third unit of the same design delivered at lower cost than the first. That is the entire thesis, and no Western developer has demonstrated it. Until then, small modular reactors are a plausible answer to a real problem, funded on the strength of the diagnosis rather than the cure.

What to watch

Watch cost per unit on repeat orders, not first-of-a-kind. Watch regulatory approvals for non-light-water designs. And watch whether any developer builds a factory before it has an order book, which would be the first real sign of confidence.