Restart Math, SMR Cost Curves, and Where the Risk Sits

Let me now get concrete about the numbers, because the economics differ sharply between the two strategies, and the difference tells you where each fits in a buyer's portfolio.

Start with restart math, which is the most favorable in the sector. When you restart an existing reactor, you skip the single most dangerous cost: the greenfield construction of the nuclear island itself, built years or decades ago and largely paid off. Your costs are refurbishment, replacing aged components, refurbishing turbines, refueling, requalifying systems, and satisfying the regulator that the plant is fit to return to service. Those costs are real and can run into the low billions, but they are a fraction of building new, and crucially they carry far less schedule risk because you are restoring known equipment rather than pioneering a new design. This is why the restart PPAs pencil out at prices a data-center buyer will accept. The buyer is paying a premium over historical wholesale rates, but for firm clean power delivered this decade, that premium is cheap relative to the alternative of not having the power at all.

Now the harder case: new-build SMRs. Here the economics rest entirely on a wager about learning curves. The levelized cost of the first SMR will be high, plausibly higher per megawatt than a conventional large reactor, because you lose the economies of scale that come with building big, and you pay all the first-of-a-kind costs of a design that has never been built. The SMR thesis is that you claw that back through economies of volume: by standardizing the design and manufacturing many identical units in a factory, each successive unit gets cheaper as the workforce and supply chain climb the learning curve. In manufacturing this is a well-documented phenomenon. In nuclear it is, so far, a hypothesis. The industry has never built enough identical reactors to prove the curve bends the way the models promise.

That is the central financial uncertainty of the SMR sector, and it produces a chicken-and-egg problem worth naming. To get the cost reductions you need volume. To get volume you need customers willing to place large orders. But early customers face the high first-of-a-kind cost, so someone has to be willing to overpay for the early units to seed the volume that eventually lowers the price. This is exactly why the involvement of hyperscalers matters beyond any single deal. A company like Amazon or Google, with an enormous balance sheet and a strategic need for future firm power, can afford to anchor early orders at unfavorable early-unit economics precisely because it is buying an option on cheap power later and a hedge against the power constraint. In effect, these buyers are subsidizing the descent down the cost curve, and they are willing to because the strategic value of secured power exceeds the premium.

So where does the risk sit in each case? For restarts, the risk is mostly finite and bounded: refurbishment could cost more than expected or take longer, the regulator could impose additional requirements, but the fundamental asset works and the path is understood. For SMRs, the risk is existential to the thesis: if the learning curve does not bend, if first-of-a-kind costs prove sticky and the tenth unit is not meaningfully cheaper than the first, the whole economic rationale weakens and the sector may stall the way NuScale's Utah project did when its costs rose past what subscribers would bear. That cancellation is the empirical warning the sector has already received, and it is why sober analysts treat SMR cost projections as aspirations to be verified rather than facts to be banked.

There is also a subtler risk that applies to both: the possibility that something else solves the power constraint first. If grid-scale batteries get cheap and durable enough to firm renewables across multi-day gaps, or if enhanced geothermal delivers firm clean baseload at lower cost, or if gas with carbon capture proves practical, the specific niche that makes nuclear attractive could narrow. Nuclear's current opportunity exists because it sits at the unique intersection of firm, clean, and dense with a customer desperate enough to pay for it. Anything that fills that intersection more cheaply is a competitive threat. The empiricist holds the nuclear thesis firmly but not religiously: it is winning a specific competition under specific conditions, and you should keep watching whether those conditions hold.

My bottom line on the economics. Restarts are the near-certain, finite, this-decade play, financeable today because the offtake fixes the revenue risk on an asset whose construction risk is already retired. SMRs are the larger, longer, higher-variance bet whose payoff depends on a manufacturing learning curve that has not yet been demonstrated at scale, financeable now only because strategic buyers will underwrite unfavorable early economics to secure future power. Read every deal by asking which of those two logics it runs on, and you will price its risk correctly.