Nuclear Power Wins As Uranium Supply Turns Tight

I kept coming back to one sentence from a crowded hall in London this September, and it was not the one the slide decks wanted me to remember. A uranium executive, the kind of person who has sat through a decade of polite indifference, said the industry no longer has to sell the story. It has to turn the story into concrete, steel, and fuel. That sounds like victory. Sit with it for a minute and it starts to sound like a warning.

For most of the 2010s, the annual gathering of the world nuclear trade body felt like group therapy with name badges. Fukushima was still the ghost in the room. Germany was switching the lights off on purpose. Uranium had spent years in a bear market so long that even the optimists stopped checking the quote. Delegates explained, carefully, why the technology deserved to exist. This year the room was different. Attendance was up about 17 percent. More than a thousand people came from fifty countries. Thirty-five organizations joined the association for the first time. Nobody was apologizing.

The Argument Is Won, and That Is the Awkward Part

Analysts who walked the floor came away with the same reading. Nuclear power has largely won the policy fight and the market-acceptance fight. Attention has moved to execution at scale. The question is no longer whether the world builds reactors. It is whether anyone in the West can build them on time and on budget. On that score, the industry’s recent history does not need a long footnote. It needs a wince.

I’ve found that industries talk this way right before the hard part starts. The narrative gets clean. The capital starts nodding. Then somebody has to pour the first concrete and live with the overrun. That is where this symposium actually got interesting.

Everyone Wants Bronze, Nobody Wants the Gun

The line that stuck was almost comic. Everyone is racing for bronze. Most utilities would rather be the third owner of a new large reactor than the first. A large American power company said the same thing from a different panel: prefer third, not first. You can dress that up as prudence. It is also how you get a race in which every runner stays on the blocks and watches the others breathe.

We do not need to sell the story anymore. We need to convert it to action.

Chief executive of a major uranium producer, speaking in London

There is a catch inside that confidence. The same executive, whose company owns just under half of a flagship Western reactor vendor, was candid about it. In the West, nobody wants to go first. Capital is no longer the missing ingredient in the speech. Nerve is.

Perhaps the most interesting aspect is how normal that hesitation has become. Boards remember Vogtle. They remember European projects that turned a decade into a punchline. They remember political cycles that can rewrite a permit. Being third is a strategy. It is also a queue, and queues have a way of never moving if the person at the front keeps checking their phone.

What the Official Outlook Finally Admitted

The London meetings lined up with the latest long-range outlook from the international atomic agency. Projections were raised for the sixth year in a row. Under the high case, global capacity reaches about 1,045 gigawatts-electric by 2050, up from 992 in the prior edition, and about 1,284 by 2060. That is roughly 3.4 times the 377 gigawatts operating at the end of 2025. Even the low case now roughly doubles capacity by 2060.

The more useful shift, to my eye, is where the revision landed. The low case moved the most: up about 14 percent, to 641 gigawatts, against a 5 percent lift in the high case. That is what de-risking looks like on a chart. The floor rose. North America’s low case flipped from contraction last year to growth this year, around 136 gigawatts by 2050 versus 98 before. When the pessimistic path stops being a decline, the argument has changed shape.

Two points from the sell-side notes are worth keeping, and not because they flatter the trade.

• The forecast is built from the bottom up. It is not reverse-engineered from a net-zero slogan. Nobody started with 1.5 degrees and worked backward until the reactors appeared. That makes the numbers less theatrical and, frankly, more usable.
• The driver has shifted from climate branding to energy security. Zero-carbon still matters. It is just no longer the only reason a finance ministry picks up the phone.

Climate politics ebb and flow with elections. Security of supply does not, or at least it ebbs more slowly. Germany’s own economy minister calling for a nuclear rethink while power prices surged was late. It was also hard to argue with. Better late than a second winter of industrial apologies.

Small Reactors Stopped Being a Footnote

Last year the agency’s range for small modular reactors was so wide it was almost useless: 5 to 24 percent of new capacity by 2050, with the low end a paltry 16 gigawatts. This year the band tightened to 23 to 28 percent, or something like 120 to 285 gigawatts. In North America, roughly 60 percent of new nuclear capacity is expected to come from these smaller designs.

I have been skeptical of renderings that look like appliance catalogs. A reactor is not a shipping container with a logo, no matter how many pitch decks say otherwise. Still, the direction of the revision matters. Modular, and in some cases behind-the-meter, is where a lot of the new power demand actually lives. Data centers do not want a white paper. They want megawatts that show up before the next training run.

One hyperscaler is already putting real money behind a high-temperature gas design. Another has sketched a path to more than triple its own data-center capacity, from about 12 gigawatts toward more than 38 by 2032, and that figure leaves out rented capacity at specialist cloud firms. Those gigawatts have to come from somewhere. A power-purchase agreement is a piece of paper. A wire is a queue.

There is no point signing a power contract if the connection takes seven years.

A large technology buyer, on interconnection delays

That sentence should be taped above a regulator’s desk. It is the grid queue in one breath. It is also why on-site generation keeps coming up in rooms that used to talk only about wholesale markets. If the public network cannot move fast enough, the load will try to bring its own plant. Some of that will be gas. Some of it, if the paperwork ever catches the ambition, will be nuclear.

The Fleet Is Old, and Standing Still Is Already a Build

Here is the part that gets less applause. Two-thirds of operating reactors are more than 30 years old. Forty-six percent are over 40. Even in the high case, about a third of today’s capacity retires by 2060. The industry has to keep building just to stay flat. In 2025 it did not quite manage that. Seven reactors, about 2.8 gigawatts, were retired. Only three, about 3.0 gigawatts, were connected. Nuclear generation grew 1.1 percent. Overall electricity generation grew 2.7 percent. Nuclear’s share of global power slipped to 8.4 percent from 8.7.

Read that again if you are tempted by the victory lap. The story won. The share of the grid fell. Extensions and uprates can buy time. They are not a substitute for new steel, and they do not refill a mine that has already been worked out.

Tables like that are easy to skim and easier to misuse. The point is not that nuclear is failing. The point is that a won argument and a built fleet are different objects. One lives in communiques. The other lives in a construction schedule.

Six Trillion, and the Money Is No Longer the Punchline

Day one was a finance summit, and the headline number was large enough to make a room go quiet. Meeting the political pledge to triple nuclear would take something like $6 trillion over 25 years. Annual nuclear capital spending would have to rise from about $80 billion today to more than $250 billion a year through 2050. Current national targets, added up, point even higher: around 1,457 gigawatts by 2050, closer to a quadrupling than a tripling.

A few years ago the question in these rooms was whether the money existed. A panel of vendors and fuel-cycle firms said that question has flipped. The money is there. Discussions, according to people who sat in them, are less about whether institutions will invest and more about when. The association even launched an investment guide, an attempt to drag a specialist energy bet into the costume of mainstream infrastructure.

The goal is tidy: strip out design risk, licensing risk, regulatory risk, and supply-chain risk, and leave only execution risk. Based on history, execution risk is plenty. I would not bet a career on a model that assumes the last risk is the small one.

Customers Are Blunter Than the Vendors

The most useful comments came from the people who would actually buy the power. A major oil company, not a natural cheerleader for reactors, said the industry does not have a technology problem. It has a deployment problem. The risk model is wrong. In liquefied gas or a combined-cycle plant, nobody expects the equipment vendor to carry most of the project risk. The owner-operator does. That company is preparing a report on a different delivery model. Whether anyone adopts it is another matter. The diagnosis is hard to dismiss.

Demand, in other words, is not the mystery. Microsoft-scale load growth is not a rumor. Interconnection is. Permitting is. The willingness to be the first name on a large Western project is. You can have a full order book of intentions and still pour no concrete.

China Builds While the West Watches the Start Line

While Western utilities wait to see who blinks, China explained how it does the unglamorous version. A senior figure from the state nuclear technology group said the country had 62 reactors operating, about 66 gigawatts, and 58 under construction, about 69 gigawatts, as of July. Targets sit at 110 gigawatts by 2030 and 150 by 2035. The method is dull, and it works: standardization, specialization, centralization.

The targets they recited are the kind of numbers a project manager can actually use. Eighty percent design reuse. A 56-month construction schedule. Factory acceptance rates above 93 percent. Centralized procurement. Some Western projects have spent longer than 56 months on permitting alone. That comparison is unfair in a dozen legal ways. It is also the comparison the market is making.

Counting rules matter, and it is worth being honest about them. The Chinese figure of 58 includes projects Beijing treats as underway. The trade association counts a reactor only after first concrete. On that stricter definition, a monthly tracker put China at 37 reactors under construction as of mid-August. That is more than twice India and Russia combined, and nearly as many as the other sixteen countries on the list put together. The United States does not appear on the chart. China 37. United States zero. That is the more telling number.

This is where racing for bronze leads. The United States has not qualified for the large-reactor race yet. China is already building the next set. A speaker from a major American utility said the quiet part in public: the West should apply the China model. Not the politics. The repetition. License once, review once, build many.

There is movement, uneven and late. Finland is cutting binding regulatory requirements from something like 8,000 down toward 1,500. The American nuclear regulator is being pushed, under recent legislation, to change its culture rather than just its pamphlets. A European effort aims to harmonize design reviews across borders so a plant accepted in one capital is not treated as science fiction in the next. None of that pours concrete this quarter. All of it is the precondition for anyone agreeing to go first.

A World That Is No Longer Waiting for Permission

Japan is back in a way that would have sounded reckless a decade ago. The operator of the Kashiwazaki-Kariwa station restarted Unit 6 in February after fourteen years offline. Commercial operation followed in April 2026, returning a 1.36-gigawatt advanced boiling-water reactor to the grid. The seven-unit site holds 8.2 gigawatts, the largest power plant on earth by that measure. A western Japanese utility said fifteen reactors have now restarted since Fukushima. Policy language has moved from reducing nuclear to maximizing it, with a 20 percent share targeted by 2040 and eleven to fourteen new large reactors eyed by the 2050s.

Was new Japanese build inevitable? A year ago that was still a real question. The answer, sitting in that room, looked like yes. Energy-import dependence has a way of settling arguments that ideology prefers to keep open.

• Poland has lined up about 17 billion euros for its first plant, with three large American-design units in preparation.
• The Czech Republic is moving toward 80-year operating lives and is building new units alongside a British small-reactor design.
• In the United States, an executive order from May 2025 called for ten large reactors under construction by 2030. A federal long-lead procurement program of about $17.5 billion has letters of intent with seven unnamed counterparties. Some attendees think one or two site announcements could land before the midterms. A sodium-cooled project in Wyoming already has a construction permit and an active job site.
• In Africa, Togo’s president signed the tripling pledge and will host a regional nuclear summit in 2027. The ask was partnerships and investment, not technology handouts. That distinction is sharper than it sounds.
• A Korean shipbuilder presented on nuclear-powered commercial vessels, a follow-on to Washington laying groundwork for offshore nuclear. The risk disclosure was admirably plain: it might not work.

That last line is the healthiest sentence in the whole symposium. Nuclear marketing has a habit of sanding off the failure modes. A builder who says the thing might not work is a builder you can underwrite. A builder who promises a plug-and-play miracle is selling a rendering.

You Cannot Print a Uranium Deposit

This is where the meetings got most interesting for anyone who actually allocates capital. Reactors can be standardized, financed, and, eventually, built. The fuel to run them is a different problem. Uranium mining is the bottleneck. Deposits are not reproducible. Lead times from discovery to production run as long as twenty years. Conversion, enrichment, and fabrication can be expanded with capital and permits. A mine cannot be willed into existence because a data center signed a memorandum.

The head of the Kazakh producer put it more bluntly. After years of underinvestment, producers will prioritize customers who give clear, long-term demand signals. The company will no longer turn down attractive eastern deals while it waits to see whether Western utilities mean it this time. Western buyers who have been sitting on the sidelines for a softer spot print may find the inventory has already been spoken for.

Timing made the point sharper. Russia banned sulphuric-acid exports through year-end, a squeeze that threatens something like three million pounds of Kazakh output in 2027. The Kazakh producer also agreed to sell uranium into the Russian fuel system. Coincidences happen. Supply chains notice them anyway.

Run the arithmetic and it stops being abstract. At roughly 0.4 to 0.5 million pounds per gigawatt-electric per year, the agency’s 641 to 1,045 gigawatt range implies 288 to 470 million pounds of annual uranium demand by 2050, against about 168 million pounds today. That is before you replace mines that deplete. Primary supply does not scale like a software license.

A rough fuel sketch, not a forecast:
  Today's uranium demand .............. ~168 million lb/year
  2050 low-case call ................... ~288 million lb/year
  2050 high-case call .................. ~470 million lb/year
  Discovery-to-mine lead time .......... up to 20 years
  You do not 3D-print the missing pounds.

Enrichment Is Already Short in the West

Strip out Russian and Chinese separative work and the Western picture is already tight. Demand sits around 28.5 million separative work units against about 24.8 million of supply from the main Western enricher. That is a deficit before the buildout, not after it. In the high case, demand rises toward 57 million. The gap is on the order of 32 million units, and it can stretch toward 69 million if small reactors running on higher-assay fuel take a large share.

The enricher confirmed the demand from the other side of the table. Its order book has grown from 8.7 billion euros in 2021 to 21.3 billion. It is adding 4.6 million units of capacity, almost a third more, based on a market assessment rather than a stack of signed orders. That is confidence. It is also a bet that customers will still be there when the centrifuges spin up.

Elsewhere in the fuel chain, a large American utility said fuel enriched up to 10 percent, sometimes called LEU-plus, will allow 24-month reload cycles across its fleet. A laser-enrichment venture expects a license near an old gaseous-diffusion site by 2027. Earlier this year the American energy department put billions behind domestic enrichment. The money, increasingly, is going to the fuel chain rather than to another rendering of a reactor that does not yet have a site.

Is enrichment the whole trade? No. Asked to choose between enrichment and uranium, at least one long-only desk picked the mined metal. Both need the same atoms. Only one of them takes a generation to find.

Prices Are Not Waiting for the First Concrete

The long-term contract price for uranium hit an all-time high of $96.50 a pound, above the prior 2007 peak of $95, even while many nuclear equities have acted as if they missed the memo. Desk forecasts still point higher: about $104 in 2026 and nearly $130 in 2027, which would be a fresh record, before a long-term anchor nearer $85. Treat those numbers as a house view, not destiny. The direction is the point.

Spot has been trading below term. That used to be a curiosity. The more important change is who is buying. Producers’ share of term selling fell from 92 percent in 2023 to 76 percent in 2025. Rising utility participation in the term market is what pushes the price, because churn between traders gets replaced by end-user demand that actually burns the fuel. Less musical chairs. More consumption.

For anyone who thinks the move is finished, pull up a 38-year chart and sit with it. On some measures uranium is still not in the mania of the last cycle, even though the demand story is broader than it was in 2007. Then it was a China build and a supply shock. Now it is a China build, a Western restart, a Japanese return, life extensions, and a data-center load that did not exist as a planning category. A cumulative supply deficit measured in the billions of pounds through the 2040s is the sort of figure that makes term buyers nervous and spot tourists late.

I do not think the market has fully priced the physical constraint. Equities have priced the presentation. Those are not the same trade.

Where the Value Chain Actually Pays

Sell-side desks agree on direction and disagree on where to stand. One remains constructive on American power producers with nuclear fleets, on a couple of reactor developers, and on fuel-cycle names. Another cut price targets on several small-reactor developers to reflect a higher cost of capital, which is a polite way of saying more dilution, and a slower buildout between 2030 and 2035. That desk prefers companies that make money now: fuel, components, enrichment, in-situ miners.

That second reading feels closer to what the symposium actually said. The long-term story is intact and, if anything, improving. Official forecasts are up again. Hyperscalers are writing checks. Japan is back. The near-term constraints are physical: fuel, forgings, welders, and grid connections. Reactor developers have to wait for the first Western mover to deliver on time and on budget before the multiple expands for real. Uranium miners and enrichers do not. Both sides of the reactor debate burn the same fuel.

1. Policy acceptance is no longer the scarce resource. A construction company that finishes is.
2. Being third is rational for a utility and fatal for a timeline.
3. The existing fleet is old enough that replacement demand arrives even if not a single new policy is passed.
4. Interconnection queues can erase a power contract. Behind-the-meter is a workaround, not a slogan.
5. Mines take up to twenty years. Centrifuges take capital. Only one of those can be rushed with a check.
6. Term buying by utilities matters more than another spike in speculative spot volume.

What Disciplined Vision Actually Means

A delegate from a Japanese utility described the next phase as disciplined vision, and as a return to meeting in person, more analog in a digital world. That is a fair description of uranium too. It is a physical asset. It cannot be replicated by a model update. It is needed for plants that run through the night, and for the data halls that would prefer not to. It is in short supply relative to the build the speeches now assume.

Disciplined vision is an easy phrase. In practice it means fewer hero projects and more repeated designs. It means owners carrying owner risk instead of hoping a vendor’s balance sheet will absorb a decade of delay. It means utilities signing term contracts before the eastern buyer does. It means regulators cutting requirements that do not buy safety and keeping the ones that do. Finland’s cut from roughly eight thousand binding rules toward fifteen hundred is a case study in that distinction, not a call to get casual around a core.

Does the West actually do this? Parts of it might. The American long-lead program is an attempt to buy time on vessels and forgings that have multi-year queues of their own. Letters of intent are not purchase orders. A construction permit in Wyoming is not a fleet. Ten large reactors under construction by 2030 was a political target. The industry, judging by who wants to be third, has not yet volunteered a full starting grid.

The Data-Center Distraction, and Why It Is Not One

It is tempting to file the whole nuclear revival under artificial intelligence and move on. That would be too neat. Life extensions were coming back before the latest training clusters. Japan’s restart calendar was political, not algorithmic. Poland and the Czech Republic are not building for chatbots. Energy security after a European gas shock did more to change ministerial language than any keynote about tokens.

Still, ignore the load and you miss the urgency. A buyer who needs tens of gigawatts inside a planning cycle measured in years will not wait for a 2038 commercial operation date and call it a strategy. That buyer will sign gas. That buyer will look at on-site nuclear if someone can license it faster than a transmission line. The seven-year interconnection complaint is not color. It is the constraint that decides which technologies get a check.

Behind-the-meter will not be the whole answer. It might be the only answer that fits the clock for a slice of new demand. If small reactors really do take the majority of new North American nuclear capacity, it will be because the large-plant queue stayed polite and the load did not.

A Note on What Can Still Go Wrong

None of this is a straight line, and anyone selling it as one is selling something else. Construction can slip. A single high-profile overrun can freeze the bronze-medalists for another cycle. Politics can reverse a restart, as Japan has already learned. Enrichment projects can miss their license dates. A sulphuric-acid ban can nick supply, and a deeper trade rupture could do worse. Higher prices bring on supply, eventually, including from projects that look marginal at $60 and obvious at $120. The twenty-year lead time cuts both ways: it supports the price, and it means the response is slow.

There is also the share-of-grid problem. If electricity demand grows faster than nuclear output, the technology can win every conference and still lose mix. That is what 2025 already showed in miniature. Generation up. Share down. A tripling of capacity inside a world that also triples everything else is a different outcome from a tripling that actually moves the percentage.

And the human constraint is easy to leave off the slide. Welders, nuclear-qualified engineers, project managers who have seen a plant through hot functional testing: these are not job posts you fill from a general contractor’s bench. China built that bench by building plants. The West let the bench age. You can fund a program in a quarter. You cannot train a crew that has never closed a containment in the same quarter.

How to Read the Next Two Years

If I had to watch a short list, it would not be another outlook revision. Those will keep drifting up until a project slips badly enough to pull them down. I would watch first concrete in the West, not announcements of intent. I would watch term contracting volumes, and whether utilities are the ones signing. I would watch whether the Kazakh producer keeps preferring buyers who show up with duration. I would watch enrichment expansions against the Western deficit that already exists. I would watch Japan’s new-build language turn into a site, and Poland’s financing turn into a schedule that survives contact with a regulator.

Site announcements before an American midterm would be a tell. So would silence. TerraPower’s Wyoming job site is a tell of a different kind: a design that is not the standard light-water fleet, with a permit and people on the ground. One project does not make a renaissance. Zero projects make a slogan.

The investment guide launched in London is an attempt to make this legible to infrastructure allocators who do not want to become nuclear specialists. Taking design and licensing risk off the table is the right sales pitch. Leaving execution risk on the table is the honest one. History says that is the risk that bites. Anyone underwriting a Western large reactor as if it were a toll road with a reactor-shaped logo is using the wrong comp set.

Picks, Shovels, and the Cost of Waiting

There is an old mining line about selling picks and shovels during a rush. It maps cleanly here, with one edit. The shovel is the fuel, the forging, the vessel, the enrichment cascade. The pick is the construction crew. The gold nugget everyone puts on the cover is the reactor developer with a beautiful design and no reference plant in the West. Some of those designs will work. The cost of capital between here and a first commercial unit is the part the target-price cuts were trying to admit.

Power producers that already own operating plants sit in a different seat. They collect the scarcity rent of a fleet that is hard to replicate, they can extend lives, and they do not need the next AP1000 to finish before the cash flow exists. That is not a free option. Outages, politics, and wholesale prices still matter. It is a different risk from funding a first-of-a-kind.

Fuel-cycle exposure is the one that does not require the bronze medal to be awarded. A reactor in China, a restart in Japan, a life extension in the American Midwest, and a small unit behind a data-center fence all ask for the same pounds and, in many cases, the same separative work. That is why the bottleneck talk in London mattered more than the attendance figure. A full room is a sentiment indicator. A twenty-year mine lead time is a physical one.

The Story Converted, or Just Repeated

So did nuclear win the argument? Yes. I think that part is settled, and arguing it again is how you waste the next five years. The symposium’s own mood said as much. Record attendance. New members. Customers from oil and from cloud computing in the same building. An official outlook that raised the floor, not just the ceiling. Japan using the word maximize. China publishing a construction cadence measured in months, not in election cycles.

Winning the argument is the cheap part. Converting it is the plant, the permit, the crew, and the fuel contract signed before someone else takes the pounds. The West’s preference for third place is understandable. It is also a choice with a scoreboard, and the scoreboard currently shows a long list of reactors under construction with the United States absent from it.

Maybe that changes with the long-lead program, with a culture shift at the regulator, with one utility that decides bronze is a consolation prize. Maybe it does not, and the build stays concentrated where the state can order repetition. Either path still runs through uranium. Either path still runs through a fuel cycle the West let thin out. That is the part of the London week that does not need a new narrative. It needs a calendar, a contract, and a deposit that already exists.

The executive was right that the selling is over. The awkward sequel is that action has a first mover, and right now the first movers are not the ones giving the fireside chats about how ready everyone feels. Watch the concrete. Watch the term book. The story can take care of itself.