The Forge at Muroran
There is no shortage of writing about the nuclear renaissance in 2026.
There is almost none about its physical limit.
Every serious investor in the space can recite the demand. Hyperscaler PPAs. Polish, Bulgarian, British, French, Swedish, Finnish, Korean, Japanese restart calendars. The Westinghouse-Hyundai consortium. Doosan and Fermi America.
Constellation buying Three Mile Island back for Microsoft. Amazon at Susquehanna. The first EPR2 final investment decision expected in France by year-end 2026.
Almost nothing about whether the equipment to build any of it actually exists.
The chip-equivalent of this thesis is not the fab. It is the forge.
Every gigawatt-class reactor pressure vessel in the Western world begins as a single piece of forged steel — a ring weighing several hundred tons, hot-formed under a hydraulic press operating in five places on Earth.
Three of those presses actually matter. One of them, until very recently, was the only one that mattered at all.
That press lives in Muroran, on the southern coast of Hokkaido, inside the heavy machinery shop of Japan Steel Works.
I have been watching this dependency for four years. The story is hard to operationalize, which is the polite way of saying it has not been operationalized.
JSW the stock has run — roughly fourfold since the end of 2023, on a one-line consensus that “nuclear is back and JSW makes the steel.” The one-liner is not wrong. It is also not the trade.
The trade is what happens when the order book exceeds the press hours, when the geography of the cap-ex shifts, and when the consensus misses where the steel actually gets forged in 2027 and 2028.
This is the piece that goes all the way down to the press and stays there.
Yamato, in the same shop
Japan Steel Works was formed in 1907 for one purpose: to forge gun barrels for the Imperial Japanese Navy.
The Muroran plant was the place where the largest barrels were made. The forty-six-centimeter cannons mounted on the battleship Yamato — the largest naval guns ever put to sea — were forged at Muroran in the late 1930s.
The mill survived the war largely because the topography of Hokkaido placed it outside the main bombing target lists.
Stand on the access road above the plant today, and you can still see the layout the Navy laid down. The shore line. The deep-water berth. The rail spur running into the heavy bay.
The reason JSW is in Muroran, and not in Kobe or Osaka, is that in 1907 the Navy needed a place where it could put a sixty-thousand-ton press, draw the cooling water it needed from the Pacific, and ship the finished barrels out by sea without crossing any other prefecture’s politics.
The same logic now applies to reactor pressure vessels.
After 1945, JSW converted slowly. Plastics machinery, paper machinery, civilian ship-class forgings.
In 1974, the company began forging components to nuclear standards. Quietly, over the next two decades, it accumulated something more important than a casting bay: process knowledge.
By the early 2000s, JSW’s Muroran plant was the only Western-bloc facility consistently capable of forging the largest single-piece components a Generation III+ reactor pressure vessel required.
JSW claimed, at the peak of the pre-Fukushima cycle, roughly 80% of the global market for large forged nuclear plant components.
The 526-ton RPV that arrived at Olkiluoto 3 was forged at Muroran. So was the one at Flamanville 3.
The press, and what it actually protects
A reactor pressure vessel is a thick-walled steel cylinder roughly twelve to fifteen meters tall, holding the fuel assemblies and the primary coolant at about 155 bar and 320°C, for sixty years.
You do not want welds in it.
The Western industry’s slow technical preference, over forty years, has been toward fewer welds and larger single-piece forgings. The Czochralski-equivalent in this story is the monobloc — a single ingot of nuclear-grade steel, five to six hundred tons, forged in one pass into a complete cylindrical RPV course without a longitudinal weld.
That requires a press of, conservatively, 14,000 metric tons of force, fed by a casting line that can pour 600-ton ingots without segregation.
At Muroran, JSW has both.
The 14,000-ton hydraulic press has been the only press in the developed world consistently qualified, for two decades, to forge monobloc RPV courses for Generation III+ reactors.
The press itself is, by industrial standards, not exotic.
The forty years of accumulated process documentation behind every JSW course — heat numbers traceable, inspection reports countersigned, regulator dossiers running thousands of pages — is.
That distinction is the entire story.
A capital monopoly is broken by spending money.
A regulatory monopoly is broken by spending years.
On the timescale investors actually care about, the two are not the same thing.
Toshio Matsuo, April 2022
Toshio Matsuo became president and chief executive of Japan Steel Works on April 1, 2022.
Two months earlier, in February, a whistleblower at JSW’s wholly owned subsidiary Japan Steel Works M&E — also in Muroran — had reported that inspection data on forged and cast steel products had been routinely falsified.
The internal investigation that followed found 449 confirmed cases of fraud, stretching back, at the latest, to 1998. Twenty of the affected components were for nuclear-industry use, the oldest dating to 2013.
The committee report came out in November 2022.
The mechanism was, in the dry phrasing of the report, that “if products failed to meet the standards during inspections, the product department instructed the inspection department to fake the data so that they would pass.”
Twenty-four years.
There is a reading of this scandal in which it is contained.
Most of the falsified components were not nuclear. None has, to public knowledge, failed in service. The French nuclear regulator examined the implicated components and concluded that the safety case was not compromised. Trading on the stock barely flinched.
There is another reading.
JSW M&E was, for much of those twenty-four years, the only facility outside South Korea and China consistently qualified to forge a monobloc EPR-class reactor pressure vessel course — and during that entire window, the inspection regime that was the regulatory expression of that qualification was, in part, fictional.
Matsuo took the company over the week the second reading became public.
His first eighteen months were spent on the rebuild — new quality organization, new whistleblower lines, new external audit posture. The April 2026 board reshuffle, which clarified roles around quality and export control, is the formal end of that rebuild.
The substance of it remains unfinished.
A regulatory monopoly built on inspection data has a particular kind of fragility when the inspection data is, intermittently, made up.
The fabs and utilities did not de-qualify Muroran. The order book did not collapse. Matsuo did the work.
But the second reading is the one that matters for sizing, and the consensus has chosen to forget it ever happened.
What Doosan did while Muroran was distracted
In 2017, Doosan Enerbility — South Korea’s national heavy industries champion, headquartered at the Changwon complex on the southern coast of the peninsula — completed installation of a 17,000-ton hydraulic forging press.
The investment was approximately 200 billion won. The press is 23 meters tall and 8 meters wide.
It is the largest hydraulic forging press in the world.
It was built, in operational terms, while JSW was working through Fukushima dormancy and the early stages of what would become the M&E scandal.
Doosan was not the obvious winner of the global RPV market in 2017. The Korean industry had supplied APR1400 RPVs to Barakah in the UAE, but had no Western fleet exposure. The press was a strategic bet on a market that had not yet returned.
The market has now returned.
The press is, in 2026, suddenly central.
Doosan does not have JSW’s full qualification dossier. It also does not need one for the markets currently placing orders.
Hyundai Engineering & Construction won the Bulgarian Kozloduy AP1000 contract in 2024 with Westinghouse; Doosan is the contracted RPV and steam generator supplier.
Doosan is the Fermi America forging partner for the four AP1000s at Project Matador in Texas, announced in late 2025 ahead of the Trump-Yoon summit.
Doosan started forging NuScale SMR components in 2023.
Hyundai is the Westinghouse partner for the Sweden and Finland AP1000 opportunities. The entire Westinghouse export pipeline — Poland, Bulgaria, Ukraine, the proposed US ten-unit fleet — is being underwritten on Doosan steel.
JSW, meanwhile, holds the legacy EPR franchise.
Olkiluoto 3 came from Muroran. Flamanville 3 came from Muroran. The replacement-parts pipeline for the existing French, Finnish, Chinese-EPR, and Japanese fleets — call it forty years of forward demand at a quietly attractive margin — runs partly through Muroran.
JSW has disclosed receiving EPR2 component orders in fiscal 2023 and 2024; the headline EPR2 RPV courses are being forged at Le Creusot.
These are different businesses.
One is the AP1000 fleet build. The other is the EPR fleet’s steady-state plus EPR2’s build.
The market has confused them.
The market has bought JSW as the levered play on the AP1000 buildout, because JSW is the recognizable name and the headlines say “nuclear is back.”
JSW is not, on any rigorous read of the order book, the levered play on the AP1000 buildout.
The levered play on the AP1000 buildout is in Changwon.
And the third forge, briefly
I will not relitigate the Areva quality scandal at Le Creusot.
Wikipedia has more than enough on the 2016 disclosure that 400 forgings produced since 1965 had carbon-content irregularities, on the two-and-a-quarter-year regulatory shutdown of the forge from December 2015 to January 2018, on the Wall Street Journal’s characterization of the affair as a “decades-long coverup,” and on the company’s subsequent acknowledgement that documents had been falsified.
I mention it for one reason.
The two facilities responsible for every Generation III+ RPV course forged in the Western world both went through documented multi-decade falsification scandals in the same ten-year window.
Both were caught.
Both were rebuilt.
Both are operating now.
The barriers to entry that protect JSW and Le Creusot are real. They are also opaque enough to have, on occasion, protected the fraud.
You should not invest in this sector without internalizing that.
The consensus model — two qualified Western producers, technical moats so deep that newcomers are decades behind — is correct on the moat and wrong on the moral.
Opacity has a price, and the holders of the order book pay it in the next scandal, not this one.
That price is one of the things the paid section sizes.
Why this matters now
Three things are happening at the same time, and the order of the things is the story.
First, the EPR2 program in France is moving toward a final investment decision expected by year-end 2026.
The first six reactors — three pairs at Penly, Gravelines, and Bugey — represent a forge order of at least eighteen monobloc RPV courses through the back half of the decade.
Le Creusot handles the headline courses. JSW handles ancillary components and, depending on capacity, certain backup roles.
The orders that land in the next twelve months set the run-rate at both shops through 2032.
Second, the Westinghouse AP1000 export fleet has gone from speculative to operational.
Bulgaria’s Kozloduy 7 and 8 are under contract. Poland’s Lubiatowo-Kopalino has the engineering services contract signed and U.S. ExIm Bank financing initiated. Ukraine has selected AP1000 for nine planned units.
The Sweden-Finland Westinghouse-Hyundai partnership has begun supplier engagement. Fermi America’s Project Matador is at site selection in Amarillo.
Every one of those reactors begins as a monobloc RPV course.
Doosan is contracted to forge the Texan ones. Hyundai-Doosan is contracted to forge the Bulgarian ones.
The pattern is now visible.
The AP1000 fleet supply chain in 2026 is Korean.
Third, the SMR ramp is real but is a different forge.
SMR pressure vessels are smaller — typically 200 to 350 tons of forged steel versus 500-plus for a Generation III+ — and can be forged on presses in the 8,000 to 12,000-ton class.
That class of press is not the Muroran press.
It is a different competitive landscape, with a different set of qualified suppliers, a different geographic distribution, and at least one sovereign government quietly recapitalizing a forge it acquired in 2021 for an entirely different program.
I will name it, and the listed equity proxy, in the paid section.
You are already long this trade
Every hyperscaler power purchase agreement signed in 2024 and 2025 is, structurally, a bet that gigawatt-scale clean dispatchable power will be available on the grid five to ten years from now.
There are exactly three sources of that power in the technology stack as it stands: existing nuclear, new nuclear, and gas.
Existing nuclear is being repurposed as fast as it can be re-permitted; Three Mile Island, Palisades, Duane Arnold.
New gas has the carbon footprint the hyperscalers’ published commitments will not accept past 2030.
That leaves new nuclear.
Every gigawatt of new nuclear announced in the West begins, mechanically, as a monobloc RPV course.
Every monobloc RPV course is forged at one of three sites on Earth — and on the timescale of the announcements being made, the three sites are not interchangeable.
You are already long this trade.
Your NVIDIA position is long this trade. Your CEG, VST, NRG positions are long this trade.
The question the rest of this edition exists to answer is whether the exposure you currently own captures the right end of the supply chain, or the wrong one.
The consensus answer is JSW.
The right answer is more complicated.
What the consensus is missing
The consensus reading of JSW is: “long-standing dominant forge for large nuclear components, leveraged to the nuclear revival, stock has run but order book is filling, hold or accumulate on pullbacks.”
This reading is half-correct.
The order book is filling — but its composition is shifting toward EPR2 and Japanese restarts, not toward the AP1000 fleet.
The AP1000 fleet is the larger market by units announced through 2030.
JSW is not the levered play on that market.
The levered play on that market is in Korea, and it is a different equity with a different margin profile, a different geopolitical exposure, and a different timing structure.
Saying this out loud will lose me some readers who bought JSW at ¥4,000 and have watched it triple.
I am not asking them to sell.
I am asking them to re-underwrite what they own.
I am also asking the readers who do not own JSW, and have been waiting for an entry, to consider whether the position they actually want is the AP1000-fleet position, not the JSW position.
That is the trade the rest of this edition exists to make operational.
If you have read this far
You already know more about the structure of the global RPV forge market than the people writing most of the nuclear-revival notes you read every week.
I am not being flattering. I have read those notes. They are not paying attention.
JSW appears in them as a name. Doosan appears, when it appears at all, as a regional industrial conglomerate.
The forge in Hokkaido and the forge in Changwon are different businesses, with different customer concentrations, different qualification regimes, and different margins.
The note that explains this does not yet exist in the English-language sell-side universe.
The thing you do not yet have is the operational answer — how to position around the actual order book shift, which equities to be long and short of, and what to do with the JSW position you may already own.
The paid portion does the harder thing.
Behind the paywall, in this edition:
The full breakdown of JSW’s nuclear forgings backlog by customer and program, with my estimate of how much of the recent revenue acceleration is EPR2 versus Japanese restart versus residual AP1000 — and what that means for the multiple the market is currently paying.
The Doosan Enerbility thesis: why the Changwon press is the cleanest expression of the AP1000 export fleet, what the consensus price target misses, the timing catalysts to position before (one of them is dated, in the back half of 2026), and the specific risk to that thesis I think is most likely to materialize.
The Hyundai Engineering & Construction angle — the EPC partner Westinghouse keeps choosing — and why the listed parent’s diversified business mix dilutes the exposure more than most readers will assume. Two specific subsidiary listings give cleaner exposure. I name them.
The SMR forge as a separate trade: the sovereign forge in the UK that Rolls-Royce SMR now depends on, the listed Rolls-Royce exposure that captures it, and the one US-listed name with the cleanest defense-and-civil SMR forge profile.
The Korean nuclear ecosystem beyond Doosan — the steam generator specialists, the instrumentation suppliers, the heavy lift logistics — and which of them are listed.
The pair trade I am running personally on the consensus mispricing between JSW and Doosan, with sizing, entry conditions, and the option overlay that defines downside.
The catalyst calendar: every dated event from now to the end of 2027 that I think moves the JSW / Doosan / Hyundai / SMR complex by more than five percent, with notes on which way.
Six pages of operational detail, on a supply chain the market has badly mis-mapped, ahead of catalysts that arrive within the next twelve months.
The trade is positionable before the EPR2 final investment decision lands in France and before the next round of AP1000 export contracts is announced in the back half of 2026.
After those announcements, the asymmetry compresses materially…