Korea’s Nuclear Export Chain: What the Filings Show

A reactor is ordered by a government. It is built by companies you can buy. Between those two facts sits a chain of contracts, and every link in it is printed on a filing — including the ones that do not point where you would expect. This is a map of that chain for Korea’s nuclear export business, drawn only from what the filings themselves report, and honest about where it stops being able to see.

🪜 How Many Steps Are There Between You and the Company That Signed?

At least one, and that one cannot be bought: on Korea’s reactor export contracts the signing counterparty is Korea Hydro & Nuclear Power, an unlisted state company — so every listed name a foreign investor can actually own sits at least one contractual step away from the signature.

This is a different question from the one a supply-chain map answers, and it deserves its own map. Our semiconductor filing map arranged ten companies by what they sell to whom: memory maker, equipment, substrates, test interface. That works because in chips the buyer at the top of the chain is itself a listed company you can own.

Nuclear does not work that way. The buyer at the top is a state. The order is placed by a utility. And the filings that record the resulting contracts have a field — 계약상대, the counterparty — which prints exactly who signed with whom. That field is what this map is built from. It is not inference. It is a column.

In fact it is two columns, and the second one is the reason this map can tell a group relationship from an arm’s-length one. Directly beneath the counterparty on the same form sits 회사와의 관계 — relationship to the company — and the filers fill it in. On KEPCO E&C’s two contracts with the utility it reads 계열회사, affiliate. On KEPCO E&C’s contract with Doosan Enerbility it reads –: no relationship, an outside customer. In the periodic reports the same relationship is spelled out at more length, as 최대주주의 특수관계인 — a related party of the largest shareholder. So the chain below is not just a set of names in a row; each link is labelled by the filer as either inside the group or outside it.

So the organising axis here is distance from the ordering party, measured in contractual hops:

Read that last row twice, because it is the honest limit of every filing map anyone builds. Korea’s single-contract disclosure rule is proportional: the trigger is a share of the filer’s own recent revenue. A big filer therefore has a high floor in dollars and a small filer has a low one. The map gets more detailed as the companies get smaller and less consequential, and it blurs exactly where the money is largest. We quantify that below.

✍️ Who Actually Signed Korea’s Czech Reactor Contracts?

Korea Hydro & Nuclear Power signed them — a company that is not listed on any exchange, and that no investor can buy directly.

This is the single most useful structural fact about the whole business, and it is the easiest one to lose in a headline. When a news wire reports that “Korea won a reactor order,” the entity that won it is a state generation company. In DART’s corporate registry it carries code 00382001, and it has no stock code. So does KEPCO Nuclear Fuel (00160515), the fuel supplier. Neither is purchasable.

The nearest listed thing is one level up in the ownership chain rather than one step out in the contract chain: Korea Electric Power Corporation, code 015760 on the Korea Exchange, is the parent utility, and it says so in its own periodic report: KHNP appears there as the subject company of KEPCO’s nuclear power generation segment. That is a different animal from an exporter — a regulated domestic power business whose earnings are driven by tariffs and fuel costs, not by an export order book. Owning the parent is not owning the export contract, and this map does not treat it as a step in the chain, because it does not appear in the 계약상대 field of any of the contracts below.

The parent’s report matters for a different reason, and we use it twice later on. Because KEPCO consolidates the signing entity, KHNP’s own order book is printed inside a listed company’s filing — which is the only place the project’s total value appears at all, and the only place you can watch the ordering party decline to disclose things its suppliers disclose freely.

Which leaves the actual question this article exists to answer: if the signing party is off the table, who is on it, and how far out?

1️⃣ Step One: Which Export Contracts Name the Utility as Counterparty?

Six contracts, from three listed companies — Doosan Enerbility, KEPCO E&C and KEPCO KPS — name Korea Hydro & Nuclear Power as the counterparty on nuclear work outside Korea, across the Czech, Romanian and UAE projects. Their execution runs from 0% to 96.7%, which is the single most useful thing a chain map can tell you.

These are the closest positions to the signature that a public-market investor can hold. Two boundaries define the set, and both are worth stating before the table rather than after it:

• The counterparty field reads “Korea Hydro & Nuclear Power.” That is the filing’s own 계약상대 field, not our classification.
• The work is outside Korea. This is an article about an export chain, so domestic nuclear contracts with the same counterparty are excluded — and we should say plainly that the exclusion is not a size filter. KEPCO KPS’s largest current KHNP contract is domestic: ₩567,700 million ($422.1M) of multi-unit maintenance work running July 2025 to June 2027, bigger than either of its two export rows below. Excluding it is a scope decision, not a ranking.

Dollar values converted from each filing’s own won figure at ₩1,345.06/$1. Won as filed: ₩4,928,968,079,043 · ₩711,106,288,572 · ₩1,250,850,298,678 (the filing also prints EUR 724,958,299 at ₩1,725.41 per euro) · ₩467,532 million · ₩485,000 million. Execution figures are as of the June 30, 2026 half-year reports. Deliberately absent: a total. Five of these six rows carry a won or euro amount that could be added, and the sixth does not — the Romania CTRF design contract is filed in euro with no rate, so a dollar column for it would be manufactured. More importantly, adding across a chain is the error this article is about; the section after next explains why.

Four things in that table are worth reading slowly.

The end date is the same on the three Czech rows: April 18, 2038. Not approximately the same — identical. Three separate filings from two separate companies print one project clock. That is a thirteen-year exposure signed in December 2025, which is a very different thing from a 2026 news event, and it is the sort of fact that only shows up when you line the filings up beside each other.

The percentages the filings print are ratios to one year of revenue, not to the project. Doosan’s NSSS contract is disclosed at 30.36% and its turbine-generator contract at 4.38%, both against the same denominator — FY2024 revenue of ₩16,233,054,771,559 ($12.07bn), which the filings themselves label as consolidated. KEPCO E&C’s architect-engineering contract is disclosed at 226.0% of its FY2024 revenue of ₩553,362,978,026 ($411.4M). All four reconcile against those two denominators to the decimal, which is a useful check that we are reading the same fields the companies filled in. But 226.0% does not mean the contract is 2.26 times the company. It means a thirteen-year contract total is 2.26 times one year of revenue. Get the basis wrong and the number tells the opposite story — and note that the two denominators are not the same kind of thing either, one being consolidated and the other the filer’s own top line.

The two KEPCO KPS rows are a different kind of measurement, and we have flagged them rather than blended them. Both come from the company’s long-term-supply-contract disclosure, which is cumulative over the whole contract term rather than annual, and which measures against separate-basis (not consolidated) total revenue of ₩1,545,300 million ($1.149bn) at the last fiscal year end. On that basis Barakah is disclosed at 30.25% and Romania at 31.39%. Neither is an annual revenue share and neither is a new order. We include them because of what their counterparty field says: for a reactor in Abu Dhabi and a refurbishment in Romania, both maintained by a Korean contractor, the counterparty is still Korea Hydro & Nuclear Power. Even the Gulf and the Danube route through the same unlisted signature.

And the execution column is the one that repays the whole exercise, because the four filers disclose it four different ways on three different projects in the same quarter — and the differences do not track the projects. Doosan prints a progress rate itself — 진행률, 6.43% on the NSSS contract, against an order total of ₩5,063,505 million — $3.765bn at the cluster rate; note that this is not the ₩4.93tn ($3.664bn) on its contract filing, because the order book is revalued at the reporting date while the contract filing froze the signing-day exchange rate. Its turbine-generator row prints 1.24%, and only in the separate-basis table, because the contract clears 5% of separate revenue but not of consolidated. KEPCO E&C prints base, completed and balance in euro and leaves you to divide. KEPCO KPS prints base, completed and balance in won and leaves you to divide. And the ordering party, as we will see, prints one number and stops.

One more step-one contract deserves a sentence rather than a row, because it shows how ordinary this relationship is. KEPCO E&C also holds a ₩54,056,294,600 contract — $40.2M — to develop licensing documentation, counterparty Korea Hydro & Nuclear Power, region domestic (receipt no. 20260727800450, itself an amended filing). Regulatory paperwork is a revenue line. One label on it matters and is easy to miss: that filing states the amount is VAT-inclusive, while all four Czech contracts above state theirs VAT-exclusive — so this figure is not on the same basis as the table and must not be added to it. Note carefully that this one is for the APR1000, a design aimed at European requirements — not the APR1400, and nothing to do with the United States. The two designations differ by one digit and are routinely swapped.

2️⃣ Step Two: Why Does the Same Company Appear Twice at Different Distances?

Because distance belongs to the contract, not to the company: KEPCO E&C’s reactor-system design contract on the Czech project names Doosan Enerbility as its counterparty, not the utility — so on one project the same listed company is simultaneously one step out and two steps out.

This is the finding that made us build the map this way, and it is not an interpretation. It is receipt no. 20251224800396, filed December 24, 2025, twelve days after the architect-engineering contract:

So the chain, drawn strictly from the counterparty fields of seven contracts — six at step one, one at step two — looks like this. Seven contracts, six filings: KEPCO KPS’s two export rows are both printed inside one half-year report (receipt no. 20260814001938), which is why the receipt column above repeats. Count the contracts and you get seven; count the documents and you get six. Execution percentages are as of the June 2026 half-year reports:

The euro figures cross-check. The two KEPCO E&C design contracts add to EUR 724,958,299 + EUR 214,653,247 = EUR 939,611,546 — the whole of the company’s disclosed Czech design scope, in two pieces with two different counterparties. Converted at the cluster rate from each filing’s own won amount, that is $1.208bn, and we should say plainly that this particular dollar total is a sum of two conversions struck twelve days apart at two different euro rates. The euro sum is exact; the dollar sum is a convenience.

How much of it has been earned? KEPCO E&C’s half-year report (receipt no. 20260814001293) prints the order-book line for the architect-engineering contract: base contract value EUR 724,958,299, completed work EUR 23,213,165, contract balance EUR 701,745,134. The three reconcile exactly. Dividing the first two gives 3.20% complete — and that division is ours, not the company’s, because no progress rate is printed for this contract.

Say that precisely, because it is a claim about the contract and not about the document. The word 진행률 — progress rate — appears fifteen times in the same half-year report, and the report carries a dedicated five-contract progress-rate table that prints the figure to two decimals. The Czech rows are simply not in it. We come back to that table at the end of the next section but one, because it turns out to say something about the whole chain.

Those euro figures stay in euro on purpose. The half-year report prints them without a conversion rate, and applying a rate from a different filing on a different date would manufacture a number that no document supports. This is the one place in this article where we break the house rule of showing money in dollars, and this is why.

➕ Can You Add the Filings Up?

No — and this is the arithmetic rule that makes a chain map different from a supply-chain map: adding contract values along a chain double-counts, because a subcontract is a slice of the contract above it.

The four Czech filings disclose ₩4,928,968,079,043 + ₩711,106,288,572 + ₩1,250,850,298,678 + ₩373,449,426,066 = ₩7,264,374,092,359, which is $5.401bn at the cluster rate. That number is arithmetically correct and economically wrong as a project figure. KEPCO E&C’s $277.6M reactor-system design contract is work performed for Doosan Enerbility — and Doosan’s own $3.664bn NSSS contract is the scope that pays for it. Sum them and you have counted the same euros twice.

The correct statements are narrower, and we will use only these:

• Doosan Enerbility’s two Czech contracts total $4.193bn, or 34.7% of one year’s consolidated revenue — our subtotal, from the filings’ own won amounts and denominator.
• KEPCO E&C’s two Czech contracts total EUR 939,611,546, or 293.5% of one year’s revenue — again our subtotal, and again a thirteen-year total against a single year.
• The three Czech step-one filings — the ones that do not overlap — total $5.123bn. That is the largest Czech figure in this article we are willing to add up, and it is still four separate contract scopes rather than a project.

So is there a filed number for the whole project?

Yes — one, and it is not in any of the four contract filings. It is in the periodic report of the listed parent, and it is $19.23bn.

KEPCO’s half-year report (receipt no. 20260814002198) carries the order book of its nuclear generation segment — that is, of Korea Hydro & Nuclear Power, the signing entity. One row reads “Czech Dukovany units 5,6 new nuclear plant project contract,” order total 258,608 — the report’s unit is 100 million won, so that is ₩25.8608 trillion, and at the cluster rate $19.226bn. The date beside it is June 17, 2025, and the column it sits in is headed 수주일자: order date, which is the date the order was booked and not necessarily the date a contract was signed. (The sibling table for the listed engineering subsidiaries, two pages later in the same report, uses a different header — 최초계약일, first contract date. The two are not interchangeable and the report does not treat them as such.) The trillion figure and the dollar figure are both ours; 258,608 is what the page prints. This is the prime contractor’s own booked total, and it is the denominator the press coverage is reaching for when it quotes a single headline number.

Put the two together and the chain finally has a scale:

Both percentages are our division: the won sums above over the ₩25.8608 trillion implied by the filing’s 258,608.
Read the bottom row as a floor on the listed suppliers’ disclosed share, not as their economics — a contract value is revenue over thirteen years, not profit, and the shares of the project held by unlisted Korean and Czech participants are not disclosed anywhere.

So roughly a quarter of the prime’s booked total is visible as contracts on listed companies’ filings. That is a genuinely useful ratio and it is also a ceiling on how much of this project a public-market investor can be attached to — but it is worth saying what it is not. It is not a statement that the other three quarters go to someone else in a way that could be found. It is the share we can see.

We made that mistake first. An early version of the chain diagram on this page added the four Czech contracts into one project total, and this section exists because of it. We publish it for the same reason we published the three failed claims in the semiconductor filing map: a map that never shows its own corrections is asking to be trusted rather than checked.

This matters beyond bookkeeping. Press coverage of reactor exports routinely quotes a single headline value for a project and then lists the Korean companies “involved,” which invites a reader to imagine each company holding a share of that one number. The filings describe something different: overlapping contracts, at different distances, with different counterparties and one shared deadline.

🔍 Where Does This Map Stop Being Able to See?

At the disclosure floor — and because that floor is a percentage of each filer’s own revenue, it converts into a completely different dollar amount at each company, leaving the largest contracts the least visible.

Korea’s single sales or supply contract disclosure is triggered proportionally, and Doosan Enerbility’s filings state the test in so many words: the value qualifies because it is at least 2.5% of most recent revenue. Apply that to the revenue base each filing prints and you get a floor in dollars. The two withheld US filings below cite the same FY2024 consolidated revenue the step-one table used, which puts their floor at $301.7M. Below the floor, a single contract need not be filed at all.

A proportional floor moves every year, so it has to be quoted with a base. Doosan’s filings of August 21 and September 1, 2026 cite a newer one — ₩17,057,876,465,803 — and still tick the large corporation box that carries the 2.5% test, so the floor today is $317.0M (receipt nos. 20260821800022, 20260901800387). That $317.0M is our arithmetic, not a figure any filing prints. Every $301.7M in this article is the FY2024 one, which is the right floor for the two contracts below because it is the number those filings were tested against. The arithmetic for both is written out operand by operand in the Doosan article, which also shows why consolidated is the load-bearing word: Doosan Bobcat, construction equipment, is 52.25% of that revenue base, so a good year selling compact loaders raises the smallest reactor contract Doosan is obliged to tell you about.

Two live contracts sit in that shadow — a “large gas turbine package supply contract” filed December 17, 2025 (receipt no. 20251217800425) and a second on March 6, 2026 (20260306801165). On both the sales region reads United States; on both the value and the counterparty are withheld for business confidentiality. Withholding a value does not withhold everything. A filing that exists has cleared the threshold, so each of these is at least $301.7M — and neither is permanently dark, because both carry a 유보기한, a deferral deadline undertaking re-disclosure, on September 30, 2028 and November 30, 2029. The point for the map is smaller than the money: these are gas turbines, so the largest live US exposure in the whole group of four companies is not nuclear. Neither was the largest ever disclosed in full — Doosan’s Ukudu combined-cycle plant on Guam, USD 571,000,000, which ran to October 2024 (receipt nos. 20201221800128, 20220531800931).

Now the other end of the chain. BHI filed a US combined-cycle contract in January 2016 and printed its value in both currencies — ₩6,404,520,000 and USD 5,300,000 at its own stated rate of ₩1,208.4 — disclosing it as 1.3% of FY2014 consolidated revenue. At Doosan it would not have been filed at all: Doosan’s current floor is 66.6 times that contract — a ratio we take in won on both sides on purpose, because dividing our dollar floor by the filing’s own 2016-rate dollar figure mixes two exchange rates and shrinks the multiple to 59.8. So the resolution really is inverted. And note which filing was compulsory: the two Doosan filings above are mandatory forms, while BHI’s was voluntary — 자율공시 — as are four of the six BHI contracts whose region field names the United States. A company that files by choice can also decline to, so the low floor at a small filer is not a guarantee of visibility; it is an observed courtesy.

And one more limit, which is about counterparty type rather than size: the two filings above withhold both value and counterparty because the buyer is a private commercial customer. The Czech contracts print everything to the won because the counterparty is a state utility on a government export project. Distance from the buyer is not the only thing that decides what you are told. Who the buyer is decides as much.

Which is why the very top of the chain is the blurriest point on the whole map, and the reason is printed rather than inferred. KHNP’s Czech row in KEPCO’s half-year report gives an order total and then stops: the delivery date, the progress rate, unbilled work and construction receivables are all blank. A footnote on the table explains why, and the explanation is the most quotable sentence in this entire article. Our translation of it: the contract with the client bars disclosure of contract details without the client’s consent, and the client does not consent to disclosure of the delivery date, progress rate, unbilled work and construction receivables; because a breach would carry a high likelihood of a damages claim, only the order total is stated. The notes to the consolidated financial statements repeat the point for four overseas nuclear projects at once — the UAE, El-Dabaa in Egypt, Cernavoda 1 in Romania and Czech Dukovany — add the accounting authority (the Korea-specific paragraph 129.2(2) of K-IFRS 1115) and record that the omission was reported to the audit committee.

Note carefully what that does not mean. It is not a governance complaint: a confidentiality clause in a cross-border reactor contract is ordinary commercial practice, and the filer disclosed the existence of the omission, its reason and its accounting basis, which is what the rule asks of it. It means something narrower and more useful — the one row with the project’s whole value on it is the one row whose progress you cannot track, so anyone monitoring this project for delivery has to do it through the suppliers, whose disclosure is better precisely because their numbers are smaller.

With one qualification that spoils the neat version of that gradient: the same clause reaches down the chain. Remember the five-contract progress-rate table in KEPCO E&C’s half-year report. The note beneath it reads, in our translation: “owing to relevant law and to contractual reasons with the ordering party, contract items falling under paragraph 129.2(1),(2) of K-IFRS 1115 have been omitted.” Doosan’s order-book table carries its own version, and its version is countable — the delivery date is omitted for one contract on the consolidated table and for two on the separate-basis one, on the ground that the client’s contract forbids publication (receipt nos. 20260814001293, 20260814003782). The suppliers are not simply more open than the ordering party; they invoke the same exemption, more narrowly.

⚠️ The limit of that observation, stated plainly: KEPCO E&C’s note does not say which contracts it omitted. What we can say is the shape of the absence — none of the company’s foreign-currency contracts appears in that progress-rate table: not the two Czech rows, not Romania, not the two UAE dollar rows. A competing reading is that some of those rows never met the table’s own materiality test, and the filing does not rule it out. It is a strained reading for a contract worth 226% of the company’s annual revenue, but it is not impossible, so treat this paragraph as an observation and not as a finding.

🏷️ Where Does Each Listed Name Sit on This Map?

Four listed companies file contracts in this business, they sit at three different distances from the utility, and each one is decoded in full in its own article — this map places them, it does not describe them.

That division is deliberate. Everything about what these companies make, what their revenue is composed of, how they are owned and how foreign money has moved through them belongs in the company articles, where it can be examined properly and where it stays true whether or not any given project happens. Here they get one line each.

The BHI row is the one that needs a word of defence, because “step 3 or beyond” looks like a guess. It is not: it is the absence of evidence stated precisely. BHI does not appear in the counterparty field of any of the four Czech filings, and it has filed no Dukovany contract of its own. Its own disclosed counterparties are engineering and construction firms and power producers — Bechtel Power Corporation (20070905900016, filed under the company’s former name 범우이엔지), Calpine, Amec Foster Wheeler — that is, the parties who assemble a plant, not the utility that orders one. So the filings place BHI in the equipment layer below an assembler, and decline to place it on this project at all. That is a weaker claim than a step number, and it is the claim the documents support.

For readers who want the older, valuation-led view of the largest name in the group, our earlier piece on Doosan Enerbility’s valuation and SMR positioning remains available; the 2026 company article above supersedes its figures.

🇺🇸 Is There a US Link in This Chain at All?

Not today — and not since 2014. No US utility, developer or federal body appears as the counterparty on any Korean nuclear contract filing now in force, so for the United States there is no first link, no chain and no distance to measure. Three contracts once put nuclear scope and an American destination on the same form; the last of them expired on May 31, 2014.

This is a narrow claim and we want to keep it narrow. It is not a claim about whether an American reactor programme will happen, nor about the terms of any investment package, nor an argument about why a certified design has gone unbuilt for seven years — those are political and commercial questions that contract filings cannot settle, and this map does not try to. It is a claim about specific fields on a specific form: no filing in force today carries both nuclear scope and an American counterparty or an American region, so there is no distance to measure. The claim is about today, and it is bounded by a date rather than by "never".

The support for that claim is our own census, not an American record. Across 1,198 supply-contract filings from these four companies — of which we could read 1,100, a coverage rate of 91.8% whose gap is set out in full below — not one filing in force today pairs nuclear scope with a United States counterparty or a United States site. Filings once did: three contracts, at two companies, all expired by May 31, 2014. Which of them counts as "American" depends on whether you are asking about the region field, the counterparty field or a named place — three questions that pick out three different sets, and the reason a single number is always wrong here. The set-by-set count, with every filing transcribed, belongs to the companion piece on the investment framework and is set out there in full. What matters for this map is narrower: there is no live American link to measure a distance from.

The absence is worth something only because we can show you the presence. An absence claim is weak if nobody knows what the thing being claimed absent would look like — so here is the shape, from the same set of filings. KEPCO E&C’s order book carries two contracts whose client is Nawah Energy Company, both long-term engineering support for the operating Barakah plant and both running January 2018 to January 2031 (receipt no. 20260814001293; executions are our division):

Both rows reconcile: base less completed equals the printed balance, to the dollar. The two are separate lines in separate segments of the same order book — the larger under architect-engineering, the NSSS-scope one under the reactor division — so they are two contracts, not one contract counted twice. Dollar amounts are the filing’s own; no conversion is involved.

Nawah is neither Korean nor Korea Hydro & Nuclear Power. It is a foreign nuclear operator contracting a Korean listed company directly, in dollars, for work on a reactor it runs itself — and at $310M on the larger of the two lines, it is not a token. It proves the row is possible. A step-one nuclear relationship with a foreign utility is a thing Korean filings are perfectly capable of recording. For the United States, no such row exists.

Note what that does to the shape of the claim, and to its opposite. It is not that Korean disclosure would hide an American nuclear contract, or that the counterparty field somehow cannot hold an American name — the 2008 filing below holds one. It is that the field is empty, on a form that demonstrably gets filled in when there is something to fill it with.

What exists in America is a certification, and a certification names no counterparty — so it creates no step. The APR1400 design certification was published at 84 FR 23439 on May 22, 2019 (document 2019-10715), naming KEPCO and KHNP as applicants, and took effect on September 19, 2019, that date being confirmed separately at 84 FR 41885. It is codified as Appendix F to 10 CFR Part 52. A 2025 NRC rule replaced the fifteen-year duration with a forty-year one, in its own words “both for existing DCs currently in effect and generically for future DCs,” while stating that it “does not change the date of issuance or renewal for existing DCs” (90 FR 28869, effective September 15, 2025; effective date confirmed at 90 FR 41503). On that basis the certification runs to September 19, 2059, an end date that is our arithmetic from the rule and the effective date rather than a figure the NRC prints for this design. What it grants is permission for someone to apply. It is not an order, not a site, and not a party you can stand one step away from.

Which brings the 2008 contract back into view, because it is the only contract in the census that names an American place alongside nuclear scope — Vogtle, in Georgia. (Two BHI contracts from 2010 also pair nuclear scope with a US region field but name no plant, and one further Doosan contract has an American counterparty and a Chinese site; all three are transcribed in the companion piece.) Doosan Heavy Industries — now Doosan Enerbility — filed a contract on May 8, 2008 to supply core equipment for Vogtle units 3 and 4. The counterparty was Westinghouse Electric Company LLC, the region field read United States, and the value was printed as USD 247,029,000 (₩252,216,609,000 at ₩1,021.00 per dollar, as stated in the filing itself — we do not reconvert it). The contract ran to March 15, 2014. Note the shape of it: even that contract put a Korean supplier one step behind an American reactor vendor, not in front of a utility. Receipt no. 20080506800108.

🛠️ How This Map Was Built

By reading every supply-contract filing these four companies have made since 1999 — 1,198 of them — rather than by searching for the ones we expected to find.

The method matters because the alternative fails in a specific way. Searching produces the filings that match your search terms; counting produces the filings that exist. Only the second one lets you say something is absent.

One clarification about that number before the table, because it is easy to over-read. 1,198 is a count of filings, not of contracts. Korean issuers re-file a supply-contract disclosure whenever the value or the term changes, and the amendment carries the [기재정정] prefix and its own receipt number. We counted those prefixes across the whole census rather than sampling them: 557 of the 1,198 filings are amendments, or 46.5% — 259 at BHI, 177 at KEPCO KPS, 98 at KEPCO E&C and 23 at Doosan Enerbility. A further 82 are progress-update notices on an existing contract (74 at BHI, 8 at Doosan Enerbility, filed under the other material matters heading; one of the 82 is also an amendment, so the two counts overlap by one) and 13 are contract terminations — 6 at BHI, 4 at KEPCO KPS, 2 at Doosan, 1 at KEPCO E&C. So the same underlying contract can appear several times, and not every appearance is a new deal. This does not weaken any absence claim, because an absence claim is about what no filing contains and re-filings only add chances to find something. But it does mean the census is broader than the number of distinct deals, and anyone quoting 1,198 as a deal count would be quoting it wrong.

Read the last two columns against each other and you have the coverage rate: full text in hand for 1,100 of 1,198 supply-contract filings, or 91.8%. Per company it is 100% at Doosan Enerbility, 100% at KEPCO E&C, 99.0% at KEPCO KPS and 82.8% at BHI. Those percentages are ours, computed from the two columns above; they are weighted by filing count and so cannot be averaged across companies. And 91.8% is the ceiling on every absence claim in this article: about the remaining 98 filings we cannot say anything at all.

The 98 filings we could not read are the honest boundary of every absence claim in this article. They are 95 BHI filings and 3 KEPCO KPS filings, clustered in 2015 to 2020, and the reason is mundane: for some older exchange filings the DART open API returns “file does not exist,” and the web viewer that serves them was unreachable from our address while we worked. So the range of our claims is not uniform:

• For Doosan Enerbility and KEPCO E&C — the two companies this map’s chain actually runs through — coverage is 100%, and “there is no current US nuclear contract” is a census statement.
• For BHI, coverage is 82.8%, and the same statement is a statement about 456 of 551 filings. Two of the US-region contracts we found at BHI are nuclear — both filed in 2010, both titled 원자력 발전소 보조설비 공급, nuclear power plant auxiliary equipment, both with Toshiba as counterparty and both finished by 2014. The rest are combined-cycle or state no plant type. We cannot rule out something else again in the 95 we did not read.
• The window is January 1, 1999 to September 9, 2026, with pagination exhausted rather than sampled. Starting in 2015, as a shorter search would, misses the Vogtle contract entirely.

Three traps were live in this exercise and are worth naming, because each one would have produced a confident wrong answer:

The region field does not always name a country — and a counterparty’s name is not a jurisdiction. Of the 1,198 filings, 346 have a sales region that names no country: 170 read 국외 (overseas), 23 read 해외 (abroad), 5 print a bare dash, and 148 show no region field under the modern label. That is our own count, read out of each filing’s own text rather than taken from a summary, and it is a floor: filings we could not read cannot be classified either way, so the true number is 346 or higher. Roughly one filing in three, then, and the region column alone cannot be trusted.

That last group is worse than "no field" sounds, and it cost us a contract. Pre-2002 forms carry the same information under a different label — 공급(판매)지역 instead of 판매ㆍ공급지역 — so a scan keyed to the modern wording reports them as blank when they are not. One of them, Doosan’s September 2001 filing, has a region field reading "U S A" and a contract value of ₩734,272,073,594 (receipt no. 20010926000031). It is the largest US-region contract in the census and our first pass did not see it. Every filing body we could read has now been re-scanned for US-related tokens under both labels, company names and US territories included — which is a fix, not a guarantee, because the token list is still something a person writes by hand.

But that fix introduces a second trap, and we walked into it. BHI filed two contracts six months apart with almost the same counterparty string. The January 2016 one (receipt no. 20160129900118) names Amec Foster Wheeler North America and its region field reads United States. The June 2016 one (receipt no. 20160622900039) names only Amec Foster Wheeler and its region field reads overseas. We initially counted the second as American on the strength of the name — but Amec Foster Wheeler plc was a British company, registered in England and listed in London, and the filing body names no plant, no site and no country. So the honest classification is unresolved, and our US count is nine confirmed plus one we cannot place. A name match is evidence of a corporate group, not of a country.

Older filings are not in modern encoding. Exchange filings from the 2000s are EUC-KR or CP949, not UTF-8. Read them as UTF-8 and the counterparty names dissolve into noise. Decoded in the correct order, zero characters were corrupted. The related trap is vocabulary: KEPCO KPS’s reports call the Emirati project “BNPP” and never once write “Barakah,” so a search on the familiar name returns nothing and looks like absence.

Some of these filings were later amended, and amendment is not the same as market code. DART’s list interface carries a remarks flag whose values mix market designation with amendment status. We pulled that flag for every filing cited in this article. Every Doosan Enerbility and KEPCO E&C filing cited here carries a market flag only — no amendment. Five of BHI’s US-region filings and one of Doosan’s carry an amendment marker. We read the amendment chains for BHI’s 2006 LURGI contract and its 2010 Toshiba contract, and for Doosan’s 2020 Guam contract, and those amendments are reflected here. Three we could not read — the 2014 Calpine contract (20140923900036) and both 2016 Amec Foster Wheeler contracts, whose amended texts sit inside the unread 2015–2020 block. Anywhere this article characterises those three, that caveat applies.

Finally, the census has a date on it. It was counted on September 10, 2026 and closed at the end of September 9, and we can show you what that boundary costs, because a filing landed on the wrong side of it while we were writing. On September 10 BHI filed an amended supply-contract disclosure (receipt no. 20260910900425): USD 7,791,380 of power-plant equipment for the Petrovietnam Long Phu 1 project, region overseas, and neither American nor nuclear. It changes no claim in this article, and we read it rather than assuming that. That is the routine — a chain map is only current as of the day someone counted, and the honest version of “current” is a date plus the next filing you checked.

📚 Lingo Check

🎯 Why It Matters for K-Export Stars

Everything in this map was public before we wrote it, and nearly all of it was public only in Korean. The counterparty field on a Korean contract filing is not translated by any wire service. Neither is the sentence that establishes a $301.7M floor under a contract whose value is withheld, nor the relationship field that marks one counterparty an affiliate and the next an outsider, nor the footnote in which the ordering party says its client refuses to let it publish a progress rate, nor the remarks flag that quietly tells you which filings were later amended.

That is the arbitrage this site exists to run, and a chain map is where it pays best. A foreign investor reading an English headline about a reactor order gets a country, a number and a list of company names. A reader of the filings gets who signed with whom, for how long, for how much, how far along it is, and — most usefully — which of those things the documents decline to say. The first version invites you to buy a theme. The second lets you locate yourself in a structure.

It also sets the standard we want to be held to. The absence claims in this article carry a coverage percentage and a date, because an absence without a range is a guess wearing a suit. If a filing we could not read turns out to contain a US nuclear contract, this map is wrong in a way you can check — which is the only kind of wrong worth publishing.

Disclaimer: This article is for informational and educational purposes only and is not investment advice, a recommendation, or an offer to buy or sell any security. Every figure is sourced to a public filing or government record as of September 10, 2026, and figures drawn from filings become stale as new filings are made. Currency conversions are our own at ₩1,345.06/$1 unless a filing’s own printed rate is stated. Readers should verify all figures against the primary sources linked above and consult a licensed adviser before making investment decisions. See our corrections policy.

Flow footnote: in the week after this map was published, foreigners were net sellers of all four nuclear names in both the 5- and 20-session windows — while every one of them rose. Foreign Flow Watch, week of September 11, 2026.