Half-Life

There is a genuine and important story underneath the speculation, and it is worth stating fairly before taking apart the way the market has chosen to express it. Artificial intelligence is extraordinarily power-hungry, and the data centers that run it need vast quantities of electricity that is available around the clock, every day, and — because the hyperscalers have made climate commitments — carbon-free. There is almost only one source that delivers all three at scale: nuclear. As a result, after decades in which nuclear power was unfashionable, expensive, and shrinking in much of the West, it is suddenly the most coveted form of generation on the planet. Global data-center power demand is forecast to roughly quadruple over the next decade, with energy needs climbing from around 400 terawatt-hours in 2024 toward 1,600 or more by 2034, and nuclear is widely seen as an essential part of meeting it. The demand is real, the role of nuclear is real, and the revival is real. The question, as always in this series, is not whether the technology matters but what the market is paying for it, and whether the price bears any relationship to the timeline.

The thirteen-billion-dollar company that sells nothing

Consider Oklo. It is one of the most exciting companies in the nuclear renaissance — a developer of small, advanced reactors, with a celebrated collaboration with Nvidia, a customer "pipeline" it describes as reaching 14 gigawatts anchored by names like Switch, Equinix, and Meta, and a recent regulatory milestone in the form of design-criteria approval. Investors have rewarded this promise with a market capitalization of roughly $13 billion. And here is what Oklo's income statement shows for the full year 2025: zero revenue. Not a little revenue — none. The company reported an operating loss of about $139 million, a net loss of roughly $106 million, and burned through some $82 million of operating cash, while selling not a single watt of electricity, because it does not yet have an operating reactor to sell electricity from. The market is valuing, at thirteen billion dollars, a company whose product does not yet exist, whose revenue is exactly zero, and whose first commercial reactor is years from generating meaningful income.

This is the purest expression of narrative pricing the market currently offers — even more extreme, in its way, than the quantum-computing stocks examined earlier in this series, because at least some of those had a few million dollars of revenue. Oklo has the dream, the demand, the partnerships, the pipeline, and the regulatory progress, and it has built a thirteen-billion-dollar valuation on them — but it has not yet sold anything, and what it is selling is not a product but a future: the promise of cheap, factory-built, modular reactors humming away beside AI data centers sometime in the late 2020s or 2030s. Its peer NuScale, the only company with a fully U.S.-certified small-modular-reactor design and about a billion dollars of liquidity, carries a market cap around $5 billion on a similar basis — early-stage, with real commercial reactor revenue, by most assessments, not arriving until the early 2030s. The market has assigned these companies the valuations of substantial enterprises on the strength of power they have promised to deliver years from now.

The two clocks

The fatal tension in the SMR trade — and the reason the title of this piece is "Half-Life" — is a mismatch between two clocks that tick at wildly different speeds. The first clock is the AI clock, and it is fast and frantic: AI capacity is being built now, this year, this quarter; the hyperscalers are desperate for power immediately; the demand forecasts compound at terrifying rates; and the whole frenzy that has lifted these stocks is driven by the urgency of a power crisis happening in real time. The second clock is the nuclear clock, and it is among the slowest in all of industry. Building a nuclear reactor — even a "small modular" one explicitly designed to be faster — is a years-long ordeal of design certification, regulatory licensing through the multi-year gauntlet of the Nuclear Regulatory Commission, site approval, construction, and commissioning, and the industry's entire history is a catalogue of projects running dramatically over budget and behind schedule. Nuclear is the energy source that takes a decade to deliver, reliably, even when everything goes right.

The SMR stocks are priced on the AI clock. The reactors will arrive on the nuclear clock. That is the whole problem in a sentence. The market is paying, today, valuations that assume these reactors materialize on something like the urgent timeline of the AI demand that is driving the excitement — but the reactors will not, cannot, materialize on that timeline, because nuclear construction simply does not move at the speed of an AI capex cycle. By the time the first commercial SMRs are actually generating power and revenue at scale — the early 2030s, on optimistic assumptions, and the industry's optimistic assumptions have a poor record — the AI demand picture will have evolved in ways no one can predict, other power sources (gas, solar-plus-storage, the existing nuclear fleet) will have filled much of the gap that exists now, and the urgency that inflated these valuations may have long since faded. The narrative that lifted these stocks has a half-life, and there is a real danger that the narrative decays faster than the reactors get built — that the story runs out before the first watt is sold.

The flagship that already failed

The skeptic does not even have to imagine how this goes wrong, because the leading SMR company already provided the template. NuScale's flagship project — its first planned commercial deployment, a development with a Utah municipal-power consortium (the Carbon Free Power Project) that was for years the great hope of the American SMR industry — was cancelled in November 2023, undone by exactly the force that has bedeviled nuclear for decades: cost overruns. The project's estimated cost had spiraled from about $4.2 billion in 2018 to $6.1 billion in 2020 to roughly $9.3 billion by 2023, even after the plant was scaled down — and as the projected price of the power climbed with it, the buyers walked away, and the project that was supposed to prove that small modular reactors could be built affordably instead proved the opposite, collapsing before a shovel finished its work. This is not ancient history or a different technology; it is the single most advanced SMR developer's single most important project, killed by cost, within the last few years.

The cancellation matters because it punctures the central premise of the SMR bull case — that "modular," factory-built reactors will break nuclear's historical curse of cost overruns and delays. Maybe they eventually will; the concept is genuinely promising, and standardized, factory-produced units could in principle drive down costs the way mass production drives down the cost of everything. But "in principle" and "eventually" are doing enormous work in a thirteen-billion-dollar valuation, and the one real-world test of the leading design's economics so far ended in cancellation on cost. Nuclear has had false dawns before — every energy crisis for fifty years has produced a wave of "this time, next-generation nuclear will be cheap and fast" enthusiasm, and every wave has crashed on the same rocks of cost, complexity, regulation, and time. The SMR stocks are a bet that this time is different. It might be. But the base rate for "this time nuclear is cheap and fast" is grim, and the most advanced player's flagship already failed the test.

The blank-check pedigree

There is a detail about how these companies came to be public that belongs in any honest assessment, because it speaks to the kind of vehicle they are. The most prominent SMR stocks reached the public market not through traditional initial public offerings — which require a measure of financial substance and underwriter scrutiny — but through SPACs, the blank-check shell companies that were the defining speculative vehicle of the last cycle's mania. Oklo went public by merging with a special-purpose acquisition company; NuScale did the same. The SPAC structure was purpose-built to bring pre-revenue, story-driven, far-from-profitable companies to public investors faster and with looser disclosure than a conventional IPO would permit, and it produced, in the previous cycle, a graveyard of electric-vehicle startups, space companies, and other moonshots that came public on projections and then collapsed when the projections met reality. That so many of the marquee SMR names share this pedigree is not proof of anything by itself, but it is a meaningful signal: these are, by origin and by structure, speculation vehicles — companies engineered to be sold to the public on a vision rather than a track record, in exactly the way the previous wave of vision-not-substance stocks was sold before most of them disappointed.

It also shapes who owns them and how they trade. Stocks like these tend to attract heavy retail and momentum ownership, drawn by the thrilling narrative and the low nominal share prices, and they move in violent, correlated, sentiment-driven bursts — up 20% on a partnership announcement, down 20% on a dilution or a delay — with a volatility that has far more to do with the mood of the speculative crowd than with any change in the slow, grinding progress of actually licensing and building a reactor. That trading signature is itself a warning. A stock that lurches on press releases and sentiment, divorced from the multi-year physical reality of its business, is a stock being priced by the narrative rather than the fundamentals — which is wonderful on the way up and brutal on the way down, and which has nothing to do with whether the reactors ever get built.

The dilution half-life

There is a financial mechanism that makes the timeline problem worse for shareholders, and it is the same one that afflicts every pre-revenue moonshot in this series: dilution. A company with zero revenue burning eighty or a hundred million dollars a year cannot fund itself from operations — it funds itself by selling stock, repeatedly, for as long as it takes to reach commercial operation, which for an SMR developer is the better part of a decade. And the higher the stock price, the more enthusiasm there is to issue into, which means the soaring valuation is not just a verdict on the company's prospects; it is the raw material of its survival, the thing it converts, through serial share issuance, into the cash that funds another year of pre-revenue losses on the long road to a first reactor. Each round of issuance dilutes the existing holders, and over the many years until revenue arrives, that dilution compounds relentlessly. The shareholder who buys Oklo or NuScale today is buying not only a decade of execution risk on an unbuilt reactor but a decade of dilution as the company prints shares to stay alive until the reactor exists.

This is the cruel arithmetic of pre-revenue narrative stocks, and it is why so many of them, even when the underlying technology eventually succeeds, deliver poor returns to the early shareholders: the value created by the eventual product is shared among a vastly larger number of shares than existed at the start, because the company had to dilute its way across the long desert between the dream and the cash flow. Being right that small modular reactors will one day power the AI economy is entirely compatible with losing money owning the SMR stocks today, because the years of dilution between now and then can transfer most of the eventual gains away from the people who funded the journey. The half-life of the narrative and the half-life of the share count are working against the holder simultaneously.

The reactor is not the only slow thing

Even granting the optimistic case that the SMR developers execute, build their reactors, and bring them online on schedule, there is a further bottleneck that the bull narrative tends to skip, and it compounds the timeline problem rather than relieving it: the grid. A reactor that generates power is useless until that power reaches the data center that needs it, and connecting new generation to the electrical grid — or wiring it directly to a co-located facility — is its own multi-year ordeal of interconnection queues, transmission permitting, regulatory approval, and infrastructure construction. The interconnection backlog in much of the United States already stretches for years, clogged with projects waiting for permission to plug in, and building the high-voltage transmission to move power from where it is generated to where it is consumed is among the slowest, most litigated, most NIMBY-afflicted activities in the entire energy economy. The reactor is not the only slow thing; the wires are slow too.

This matters because the SMR pitch is often sold on the promise of speed and flexibility — small, modular units that can be dropped in near a data center to power it directly. But "drop a nuclear reactor next to a data center" is a sentence that contains multitudes of regulatory, safety, security, and grid-integration complexity, each layer of which adds time. The notion that AI's power problem will be solved by a quick proliferation of small reactors collides not only with the slow clock of building the reactors but with the slow clock of connecting them, licensing their unusual co-location arrangements, and satisfying the safety regime that — appropriately — surrounds anything nuclear. Every one of these steps is measured in years, and they do not all run in parallel. Stack the reactor timeline on top of the interconnection timeline on top of the licensing timeline, and the gap between today's valuations and tomorrow's first commercial watt widens further still — which is precisely the gap in which a narrative-priced, pre-revenue stock has the most room to disappoint.

Where the demand actually lands

The most useful thing a forensic eye can offer here is the distinction the hype obscures: between the companies that benefit from the nuclear renaissance now, with real cash flows, and the companies that might benefit later, on a promise. The nuclear revival is genuine, and it is already flowing into real money — but that money is landing on the established players, not the pre-revenue startups. Uranium, the fuel, has surged to a roughly fourteen-year high around $86 a pound on genuine supply tightness and real demand, and Cameco, the great Western uranium miner — a real company with real mines, real sales, and real profit — saw its earnings jump on the order of 87% year over year, its market value climbing toward $47 billion on the back of actual, current cash flows. The established reactor operators, examined in this series' chapter on AI power, are likewise generating real power and real revenue today. The nuclear renaissance is paying off — for the miners who sell the fuel and the utilities that run the existing fleet, right now, in cash.

The SMR developers, by contrast, offer exposure to the same theme with none of the current cash flow and all of the timeline, dilution, and execution risk. That is the choice the trade actually presents: you can own the nuclear revival through assets that are generating money today, at valuations grounded in real earnings, or you can own it through pre-revenue startups priced for a reactor fleet that does not yet exist and may not exist at scale for a decade. Both are bets on the same powerful, real, AI-driven demand for nuclear energy. One is anchored to cash flows; the other is anchored to a story and a clock that ticks too slowly to keep pace with the price. None of this means the SMR companies will fail, or that Oklo and NuScale will not eventually build their reactors and reward the patient — small modular reactors may indeed become a cornerstone of the AI-powered grid, and today's developers may be among the winners. But the valuations have run to where they price the triumph as imminent and certain, when it is, on the most honest reading, distant and contingent, gated by a regulatory and construction process that takes a decade and an industry track record of missing its promises. The atom is patient; its half-life is measured in years and millennia. The market is not patient; the half-life of a narrative-priced, pre-revenue moonshot is measured in the time it takes for the story to outrun the timeline — and that, history suggests, is a good deal shorter than the years it takes to build a reactor. The demand is real. The reactors are coming. But the price is on the fast clock, and the atoms are on the slow one, and the gap between them decays in only one direction.