Korea’s $350bn US Investment Package: What Is Actually Signed (September 2026)

South Korea has agreed to put $350bn into the United States. That much is signed, in a document both governments published. What is not signed is the part the market has been trading: a reported first project, a reported eight reactors, and a reported signing date that has now come and gone. Between August 26 and September 29, 2026, Korea’s own Ministry of Industry and Trade issued ten separate statements about them, and not one of those statements confirms a project — the most recent is dated September 17, 2026, and its clarification board carried no posting of any kind between September 18 and September 29 (the ministry’s separate press board kept running; see the updates above) — that board resumed on October 2, 2026 with two items, neither about this package. This piece does not argue that either side is right. It sets out what exists on paper, and what does not, as of September 29, 2026.

Updated 6 October 2026: the clarification board has resumed. It posted two items on October 2, 2026 — one from the Semiconductor Division on the execution rate of an AI-chip R&D budget, one from the Petroleum Industry Division on price-cap rulemaking (board items 2,631 and 2,632) — and neither concerns this package. That bounds the sentence that mattered: the board was silent from September 18 to September 29, and its most recent word on this package is still the 17 September statement. The 2026 count of package clarifications is unchanged at fourteen.

Updated September 29, 2026: we re-ran the two record systems this article rests on. The ministry’s clarification board has issued four more statements about the package since our first pass — September 15, 15, 16 and 17 — and a wider census found four earlier ones we had missed, so the count rises from seven to fourteen in 2026. None confirms a project. The reported September 18 signing date passed with no announcement from either government. And across the nineteen days from September 10 to September 29 the four listed Korean nuclear-chain companies filed zero contracts with the United States in the region field. The filing census, the NRC material and the Federal Register counts are dated to their own verification dates in the text; where we could not reach a source again, we say so rather than re-date it.

📋 Four Layers, Two of Them Signed

The single most common error in English-language coverage of this story is treating “the Korea deal” as one thing that is either done or not done. It is at least four things, at four different stages. Sorting them is most of the work.

📰 Which Reported Numbers Have a Primary Source?

Not one project-level figure in this story has a government document behind it — not the $22.3bn, not the 6.3GW, not the eight reactors, not the ~$120bn, not the September 18 signing. The figures that do have one are the framework totals both governments published, $350bn / $200bn / $150bn, and the $20bn annual ceiling.

The split is the point: everything at the level of the framework is documented, and everything at the level of a project is not. The table below is the whole argument of this article in one grid. The left column is what has been reported. The middle column is what we could find in a Korean or American government document. The right column is what the government actually said.

One distinction has to be made before the table, because it is the difference between reading these documents and mis-reading them. Several of the reported figures do appear inside a ministry clarification — the format prints the report’s own claim under the heading <보도 내용>, “content of the report,” before the ministry answers it. The July 28, 2026 clarification, for instance, records that Hankyung and Money Today “reported that a $19.8bn Texas LNG combined-cycle plant had been selected as the first investment project,” and then says the specifics and the timing of any announcement are not decided (Source: motir.go.kr article 172064, Jul 28, 2026). That is a government document containing the figure. It is not a government document stating the figure. A number quoted in order to be declined has not acquired a source, and the column below is asking the second question, not the first.

✍️ What Has Korea Actually Signed With the United States?

A framework, signed November 14, 2025, and a domestic statute to run it, passed March 12, 2026 — and neither one names a single project, sector or counterparty.

Start with the thing that makes this story unusual: Korea’s Ministry of Industry and Trade did not merely announce the memorandum, it published the memorandum, in Korean and in English, as PDF attachments on its own board — the attachment list names both files, 한-미 전략적 투자 MOU (국문본)_수정.pdf and 한-미 전략적 투자 MOU (영문본).pdf. We could not open either. The download endpoint answers with the correct filename in its Content-Disposition header and then serves an HTTP 404 page, on September 10 and again on September 11, 2026. So everything we quote below is from the ministry’s summary, not from the memorandum’s text. That summary describes “strategic investment of $350bn” consisting of “$200bn of investment and $150bn of shipbuilding-cooperation investment including our companies’ direct investment (FDI), guarantees and ship finance.” The signatories were Korea’s Minister of Industry and Trade, Kim Jeong-gwan, and US Secretary of Commerce Howard Lutnick.

The American side published its own account the day before. The White House joint fact sheet, dated November 13, 2025, records “$150 billion of Korean investment in the shipbuilding sector approved by the United States, which is referred to as the Approved Investments” and “$200 billion of additional Korean investment committed pursuant to the Memorandum of Understanding with respect to Strategic Investments (MOU), which is expected to be signed by representatives of the United States and ROK.” That trailing clause is worth keeping: on November 13 the American document treated the memorandum as forthcoming, and the Korean ministry published the signed text the next day.

On nuclear matters the fact sheet says two specific things, and both are hedged in ways the current headlines are not. The first is fenced by two conditions before it begins and a purpose limit at the end: “Consistent with the bilateral 123 agreement and subject to U.S. legal requirements, the United States supports the process that will lead to the ROK’s civil uranium enrichment and spent fuel reprocessing for peaceful uses.” Strip the conditions and it reads as an American commitment to Korean enrichment; keep them and it is support for a process, inside an existing treaty, subject to American law. The second is flat: “The United States has given approval for the ROK to build nuclear-powered attack submarines.” We searched the full text: it does not contain the word “Texas”, the word “reactor”, or any number of power reactors.

Then came the machinery. On March 12, 2026 the National Assembly passed the Special Act on the Operation and Management of Strategic Investment between the Republic of Korea and the United States, by 226 votes to 8 with 8 abstentions of 242 members present — and here the primary record carries more than the wire copy did, because the Assembly’s own roll-call page also prints the denominator: 295 sitting members, so 53 did not vote at all (Source: National Assembly, 의안정보시스템 roll-call record for bill 2217405, vote dated Mar 12, 2026, which prints “재석 242인 / 재적 295인 … 원안가결 242인 (찬성 226인, 반대 8인, 기권 8인)”). The act creates a Korea–US Strategic Investment Corporation with legal capital of $1.5bn (₩2tn — our conversion at ₩1,345.06/$1, the September 7, 2026 rate used throughout this piece; the statute states the won figure only), mandates that it operate for no more than twenty years and then be dissolved, and — the clause that deserves more attention than it has had — obliges the government to consult the United States on adjusting the cash-flow distribution ratio if an individual project cannot recover principal and interest within those twenty years.

That last clause is why the September 10 clarification matters. A statute that provides for renegotiating profit splits presupposes that the splits exist. The ministry’s position, as of its most recent statement on September 17, 2026, is that they do not yet.

💵 How Much Money Is Involved, and How Fast Can It Move?

The strategic-investment ceiling is $200bn, and no more than $20bn of it can be called in a single calendar year — a ceiling the Korean release states outright, and one the American fact sheet frames as an obligation Korea cannot be made to exceed.

The annual cap is the number that decides whether this story is about the next quarter or the next decade. The Korean ministry’s release states it as a ceiling in so many words — “연간 납입한도는 최대 200억 달러이며”, “the annual payment ceiling is at most $20 billion” — and then describes the mechanism: the $200bn is disbursed “by way of capital calls keyed to project milestones, with an annual limit of $20bn, in order to ease the burden on the foreign exchange market,” and Korea may ask for the timing or the size of a payment to be adjusted if the currency market is under strain. The White House text arrives at the same figure from the opposite direction, and the difference in wording is worth keeping: “As trusted partners, the two nations agree that the ROK shall not be required to fund an aggregate amount of U.S. dollars greater than $20 billion in any calendar year.” That declines to require a faster pace. It does not forbid one. The figure appears as an actual limit in Korean law: the March 2026 special act writes it in as a safeguard — “연간 대미 투자 200억 달러 최대 한도”, “a maximum annual limit of $20bn of investment in the United States” (Source: korea.kr policy briefing 148960744, Mar 12, 2026).

One limit on that comparison, and we would rather state it than let it pass. Everything above is each government’s own public account of the clause — a press release, a fact sheet and a summary of a statute. It is not the memorandum’s operative text. The ministry does publish the memorandum itself, in Korean and English, and its attachment list names both files; but the download endpoint returns an HTTP 404 page while echoing the correct filename back in its headers, on both days we tried it, so we have not read how the MOU words this clause. That matters specifically here: if the memorandum’s own Korean text uses the same “shall not be required to fund” construction as the American fact sheet, then the ceiling-versus-obligation distinction drawn above is a distinction between two press summaries and not between the two governments’ commitments. We cannot close that gap from the documents we could open. So read the asymmetry as an asymmetry between two governments’ public characterisations, which is what we actually checked.

Two consequences follow directly, without any forecasting.

First, arithmetic. At the maximum permitted pace, $200bn takes ten calendar years to deploy. Any single-year figure larger than $20bn is inconsistent with the ceiling the Korean documents publish, which is presumably why the ministry moved quickly on September 9 to contradict a report that the totals were being exceeded. Its statement is unusually direct for the genre: the claim that the United States had demanded more than $350bn, or that strategic investment would exceed $200bn, “is not true.”

Second, a units trap that has already produced errors in English-language write-ups. The Korean documents state the annual cap as 200억 달러 — “20 billion dollars.” Read as the more familiar 2,000억 (200 billion), the cap becomes the whole programme in a single year. It is a one-character difference in Korean and a tenfold difference in dollars. If you see a report describing $200bn of annual Korean funding, that is the error.

There is a third, softer condition worth recording because it is the government’s own stated screen. The ministry’s release says Korea will recommend to the US President only investments that are “commercially reasonable” and where “sufficient recovery of the investment is assured.” That is a standard, not a project list, and it is the reason a project can be reported, discussed and still not exist as a commitment.

🗣️ Has the Korean Government Said Anything About the Reported Nuclear Deal?

Yes — fourteen times in 2026, in fourteen separate published documents, and not one of them confirms a project. Nine say the specifics are not decided, four deny a specific reported claim outright, and one does both.

This is the part of the story that is invisible in English. Korea’s Ministry of Industry and Trade runs a board called 보도설명자료 — press clarifications. When a report is wrong or premature, the relevant division posts a document saying so, with the responding division named and the answered article named in the attachment filename. Fourteen of them concern the US investment package in 2026, ten of those issued between August 26 and September 29. They exist only in Korean, as PDF and HWPX attachments. As far as we can determine, no English-language outlet has cited them.

How we counted, because an absence claim is only as good as its method. We paginated the board back to August 2025 — 120 entries — and filtered the titles for 대미 (toward the US), 한미 (Korea–US), 미국, 원전 (nuclear power) and 웨스팅하우스, then read each hit to confirm it concerns this package rather than domestic nuclear-export policy. The filter was tested first against a statement we already knew was in the set, which is the only way to tell a working filter from a silent one. Four entries that survive the keyword filter are excluded as a different subject: the May 6, April 22 and April 13, 2026 clarifications about restructuring Korea’s nuclear-export governance, and a January 28 one about tariffs. Our count is bounded by that filter and by the board’s own titles.

Four things in that table are worth pulling out.

The May 8 entry matters most. “Nothing has been decided” is not a September posture adopted under pressure; the ministry said the same thing about the same first project four and a half months earlier, answering a different broadcaster, and said it again on July 28 to two more. The reported identity of the project has changed repeatedly across those fourteen documents — two finalists and Alaska in May, a $19.8bn Texas LNG plant in July, memory-chip capacity in August, a $22.3bn Texas gas plant in September. The ministry’s answer has not changed once. Whatever else is true, the undecided state is durable, and it is the reported specifics that keep moving.

Only two of the fourteen confirm a figure rather than declining to, and they confirm the same one. On September 9 and again on September 16 the ministry said that strategic investment above $200bn, or a US demand above $350bn, is not true. It is worth noticing why only this figure gets that treatment: the ministry will state a ceiling because the ceiling is in the signed MOU. It will not state a project, because no project is in any signed document. The pattern of what the government will and will not say maps exactly onto what has and has not been signed.

The other denials are about the record, not the deal. Three of the four flat denials — memory chips as the first project, a jointly funded Westinghouse purchase, and a return figure reported to the National Assembly — answer claims about what was decided or said internally, not about the framework’s size. The September 15 one is the sharpest: the ministry’s position is not that the 16% hurdle rate is wrong but that it never gave the National Assembly a return figure at all. That is a denial that a document exists, which is a different and more falsifiable claim than “not yet decided.”

And the attachment filenames are themselves evidence. Each PDF is titled with the response and the answered outlet in parentheses — “(9.7, Money Today)”, “(9.8, Yonhap)”, “(9.15, Hankyung)”, “(9.17, Herald Business)”. The ministry is not making a general statement into the air. It is answering specific articles, by name, and fast: set each clarification’s registration date against the article date in its own filename and the lag is never more than one day across all fourteen — eight landed the same day the article ran, the other six the next day.

🗓️ What Happened on the Reported September 18 Signing Date?

Nothing was announced. September 18, 2026 passed with no statement from either government, no contract filing from any Korean company, and no further word from the ministry — whose clarification board then carried no posting of any kind for the eleven days to September 29.

We record this carefully, because the date’s passing proves less than it looks like it does. September 18 was never a government date. It entered the record through a Money Today report on September 7, which the Nuclear Export Promotion Division answered the same day, and it was repeated in a Hankyoreh report on September 8 that added a first remittance on September 29–30 and “$2.2bn plus alpha” for the year — a report the Korea–US Trade Cooperation Division answered on September 9 and again, in substance, on September 17. A date the government declined to confirm cannot be disproved by its own passing. What it does do is close the one short-dated catalyst the reporting had supplied, and leave the record exactly where the fourteen clarifications left it.

The one thing worth more than the date is the eleven days of quiet that followed it. The ministry’s clarification board answers reports within a day, fourteen times over, whenever a report about this package runs. Between September 18 and September 29 it posted nothing at all — not about this package and not about anything else. That is not a denial, and it is not a confirmation. It is the absence of any report worth answering.

Nineteen days of filings, checked company by company

The second record system is the one that would have to move if a contract were signed. We re-ran the whole nuclear chain for every disclosure of any type between September 10 and September 29, 2026. Eleven filings, four companies, and not one with the United States in it.

The Doosan filing is the one that rewards reading. On September 17, 2026 — the day before the reported signing — Doosan Enerbility filed a 풍문 또는 보도에 대한 해명, a clarification of a media report, and it is the confirming form of one: the version a company must file when it finally resolves an earlier “undetermined” answer. The subject is not this package. It is a May 28, 2024 report that Doosan would supply equipment for NuScale’s small modular reactors, and the resolution is a denial: Doosan states that it holds a right, as a strategic investor in NuScale Power, to supply a certain volume of equipment on projects NuScale wins, but that “the business discussion between NuScale Power and Standard Power referred to in the above media report is not currently proceeding” and that the reported equipment supply “is not true as of now.” The filing lists the six prior undetermined answers it closes out (Source: DART rcpNo 20260917800618, filed Sep 17, 2026). We trace that twenty-eight-month chain filing by filing, alongside the fifty-year runway underneath this whole story, in our timeline of Korea’s nuclear industry from 1975 to 2026.

⛽ Is the Reported “First Project” a Nuclear Project?

On the reporting, no — the first project is described as a gas-fired power plant in Texas, and the ministry says even that has not been decided.

The press account, which we grade as reporting and not as record, runs roughly as follows: the first project under the investment framework would be a combined-cycle gas plant in Texas of about 6.3GW, at around $22.3bn, with nuclear following as a later tranche of some eight reactors and roughly $120bn. The reported figure has also moved: $19.8bn in July 2026, then a reported US request of $25bn, then $22.3bn. The first of those three is anchored, at least as a report: the ministry’s July 28 clarification quotes it — “a $19.8bn Texas LNG combined-cycle plant” — while declining to confirm it (Source: motir.go.kr article 172064). We record the movement because it is part of the reporting record, and we stop there — all three are press figures with no government document stating any of them, and nothing we can read tells us what moved them. The July report and the September one agree on the state and on the plant type — a Texas combined-cycle plant — and the July clarification names no site, so we cannot say from these documents whether the project being described in July is the one being described in September.

We tried to find the project in the places a real American power project leaves traces, and found nothing. There is no matching name in the ERCOT interconnection queue, at the Texas Public Utility Commission, at FERC, in developer press releases, or in Korean corporate filings. The only FERC hit on the reported project name is an unrelated 2007 gas-gathering entity. One check we could not complete: the Texas Commission on Environmental Quality moved its permit search to a JavaScript-only portal on August 31, 2026, and we could not query it programmatically. A large gas plant needs an air permit, so that is a genuine gap in our search rather than a confirmed absence.

What we can say with confidence is narrower and more useful: as of September 29, 2026, the Korean government has not identified a first project, and no Korean listed company has filed a contract for one. Those are two independent record systems, and both are empty. The American infrastructure records above were searched on September 11, 2026 and we have not re-run them since; the two Korean systems were re-run on September 29.

📂 What Do Korean Filings Actually Show About US Orders?

A census of every supply-contract filing by the four listed Korean nuclear-chain companies — 1,199 filings from 1999 to September 10, 2026 — turns up fifteen contracts with an American connection of some kind. Four of the fifteen are nuclear. Three of those four have a region field naming the United States, and the last of the three expired on May 31, 2014. Nothing nuclear with an American site or an American counterparty has been in force since.

That sentence is deliberately built out of three different questions, because "a US contract" is not one thing. The filings answer three separate questions, and they give three different answers — which is why a single number here is always wrong. We set the three out before the table.

Korean listed companies must file a disclosure called 단일판매ㆍ공급계약체결 — a single sale or supply contract — when they win a contract above a size threshold. The filing has structured fields, and one of them is 판매ㆍ공급지역: the sales or supply region. That field is why this question can be answered by counting rather than by searching.

Here is the scope, re-run on September 10, 2026 rather than quoted from an earlier pass. The census window closes on that date; the nineteen days from September 10 to September 29 are covered separately, company by company, in the section above, and they added two contracts, both domestic.

Note what the coverage column permits and forbids. For Doosan Enerbility and KEPCO E&C — the two companies at the centre of any reactor story — the census is complete, and a statement about the absence of US nuclear contracts there rests on all of their filings. For BHI it rests on 82.8% of them, and should be read as such. That distinction matters more than it looks: an absence claim is only as strong as its coverage, and ours has a hole in it that we can name.

Now the fifteen. Every figure below is transcribed from the filing, and where the filing prints its own dollar amount we use the filing’s number rather than recomputing it — the older ones state the exchange rate they applied, and re-deriving the dollars at any other rate invents a figure that is in no document. Amendments are folded into the contract they amend rather than listed separately.

Four observations, none of which require an opinion.

Four of the fifteen are nuclear, and which of them counts as "American" depends on which of the three tests you apply. Run them one at a time and you get three different answers from the same four contracts:

• Region field names the United States: three. Doosan’s 2008 Vogtle order, and BHI’s two 2010 contracts supplying Toshiba, both of which put 국외(미국) — overseas (US) — in the region field and title the scope 원자력 발전소 보조설비 공급, nuclear power plant auxiliary equipment.
• Counterparty is an American entity: two. The 2008 Vogtle order from Westinghouse Electric Company LLC, and the 2007 order printed as from "US Westinghouse" — whose region field reads China, for Sanmen and Haiyang. BHI’s counterparty on both 2010 contracts is Toshiba, a Japanese company, which at the time owned Westinghouse.
• Filing names an actual American place: one of the four. Only the 2008 filing does; the other American place named anywhere in the census, Guam, is on a gas contract. It does not say "reactor equipment"; it says AP1000 — Westinghouse’s design — at Vogtle, in Georgia. BHI’s two filings name no plant and no town; their claim to an American destination rests entirely on that region field.

So the honest sentence is not "one" and not "three" but all three at once, each with its test named. One further filing — the 2007 Bechtel feedwater heaters — names no plant type at all, so it counts in none of the four. Korea’s own reactor, the APR1400, appears in none of them either. Every precedent here is Korea supplying an American reactor programme as a vendor — not Korea exporting a Korean reactor.

And all of it ended twelve years ago. The Vogtle contract ran to March 15, 2014. BHI’s larger Toshiba contract ran to May 31, 2014 and the smaller one to August 30, 2013. That is the whole nuclear record, and its last day was May 31, 2014. So the precise statement is not that Korean companies have no US nuclear contract history — they have one, at two companies — but that no nuclear contract with an American site or an American counterparty has been in force since 2014. It is a weaker-sounding sentence and a stronger fact, because it comes with the dates on it. A keyword search for "zero" would have got it wrong in both directions.

What Korea actually sells into the United States is turbines, boilers and heat exchangers — and the biggest one of all was a gas plant on Guam. Doosan’s 2020 Ukudu contract is the largest amount printed in full anywhere in this table: USD 571,000,000, for a 200MW-class combined-cycle plant and a 25MW battery, on American soil, with the region field reading "Guam, Dededo and Piti districts" (rcpNos 20201221800128, 20220531800931). It finished on October 31, 2024. The two contracts still running are also gas: both 2025–26 Doosan Enerbility filings name the product in the contract title, and it is not a reactor. Both withhold the amount, and both state the reason in the same words: the amount “is 2.5% or more of recent revenue and therefore constitutes a mandatory disclosure item.” That sentence sets a floor, and the floor can be derived from the filing’s own numbers:

No single field gets you the right answer, and the ways it fails are not symmetrical. Count the region field and nothing else and you get thirteen contracts — but you would still miss the July 2007 Doosan contract, whose region field correctly reads China for Sanmen and Haiyang even though the orderer printed in the filing is "US Westinghouse", and you would miss the June 2016 BHI contract, which declines to name any country and prints only 국외, overseas. Count the counterparty field instead and you get nine, dropping BHI’s two nuclear contracts with Toshiba and its 2006 contract with LURGI — the three cases where the goods went to America but the company that ordered them did not live there. Count only filings that name a town or a plant and you get two.

Three of those field values also do not spell the country out in a form any obvious search would match. 국외(미국) reads "overseas (US)" in a parenthesis; the 2001 Doosan filing writes the region as "U S A" with spaces, under a label that reads 공급(판매)지역 rather than the modern 판매ㆍ공급지역; and Guam is written 괌, a word that contains no part of "US", "미국" or "America". All three of those are inside the region field itself, which is why we report the field’s value verbatim in the table instead of a tidy country code.

🏛️ If the Design Is Certified, Why Is Nothing Being Built?

Because a design certification permits construction; it does not order it — and in the almost seven years since Korea’s APR1400 was certified, not one US licence application has referenced it.

The certification is real, and its dates are worth getting right because they are routinely merged. The US Nuclear Regulatory Commission issued the APR1400 design certification as a direct final rule published May 22, 2019 (84 FR 23439), with the applicants named as KEPCO and Korea Hydro & Nuclear Power. A second document published August 16, 2019 (84 FR 41885) confirmed the effective date: September 19, 2019. A separate, earlier standard design approval was issued in October 2018. Three distinct US regulatory actions across two years, and Korean corporate filings themselves print the 2018 approval and the 2019 certification in different sections — which is how they end up conflated in secondary coverage.

One important update changes the shape of this story, and it is recent enough that most secondary coverage has not caught up. Design certifications used to run fifteen years, which would have put the APR1400’s expiry in 2034. A rule published July 2, 2025 and effective September 15, 2025 extended the term to forty years, and it did so retroactively: the rule “replaces the 15-year duration for DCs with a 40-year duration period, both for existing DCs currently in effect and generically for future DCs, including renewals,” and it “does not change the date of issuance or renewal for existing DCs” (Source: 90 FR 28869, direct final rule; effective date confirmed at 90 FR 41503, Aug 26, 2025).

Apply that to the APR1400 and its certification runs to September 19, 2059 — issuance date unchanged, forty years instead of fifteen. That date is our arithmetic on the rule and the original effective date; the NRC does not print it anywhere. Any analysis built on a looming 2034 deadline is working from a superseded rule.

So the certificate is not the constraint. The constraint is that in the United States, building a reactor requires a combined licence — construction and operation for a specific site — and someone has to apply for one. The NRC publishes the list. As read on September 10, 2026 it carried 20 entries, and the referenced designs were AP1000, ESBWR, U.S. EPR, US-APWR, ABWR and Aurora. The APR1400 appeared on none of them, and the string “APR1400” did not appear anywhere on the page. ⚠️ We tried to re-read that page on September 29, 2026 and could not: nrc.gov returned HTTP 403 to us on both a plain and a browser-identified request. So this paragraph is dated September 10 and has not been refreshed. The Federal Register check below was re-run, and it is the independent second route to the same question.

We checked that a second way, independently, and re-ran it on September 29, 2026. When a licence application is docketed, the Federal Register publishes notices. A full-text search of the Federal Register API returns 42 documents mentioning “APR1400” — all rulemakings, fee schedules and information collections, none a licence action. The most recent is dated September 24, 2026: a proposed rule titled “Regulatory Enhancements for Reactor Licensing, Decommissioning, and Operational Oversight.” It is generic licensing reform, and it is the one document that arrived between our first pass and this one; the count was 41 on September 10. A search for “Korea Hydro” returns 37 documents, unchanged, and the most recent is still dated June 19, 2020: a fee schedule. Six years of silence in the register that would have to carry the notice (Source: Federal Register full-text search, both queries re-run via the API on Sep 29, 2026).

The wider context is the least comfortable part of this section for anyone bullish on certification as a signal.

Seven certified designs. Two expired without anyone bothering to renew them. One — the AP1000 — has produced operating American reactors. That is the base rate a certification is worth, and it is the reason “the APR1400 is certified in the US” is a true statement that carries almost no information about whether anything gets built.

⏱️ How Long Does It Take a Headline to Become a Filing?

On the one Korean reactor export we can trace filing by filing, sixteen months and twenty-four days — and we know because the contract filing quotes the original newspaper headline back at itself.

This is a single data point, offered as a baseline rather than a forecast. Korea had exported reactors before the Czech contract, and the wording matters. KHNP’s own English site calls the December 2009 Barakah contract Korea’s “debut in large-scale nuclear power plant exports” — not its first reactor export, and that qualifier is doing real work. The first contract to export Korean reactors was signed on 15 December 1999: KEDO’s turnkey contract with KEPCO for two 1,000MWe Korean Standard units in North Korea, base value $4.182bn, suspended in December 2003 at about 34.5% complete and formally terminated in May 2006. The first reactor Korea actually delivered was the Jordan Research and Training Reactor, handed over on 15 June 2017 — which KAERI itself calls “Korea’s first export of a reactor.” A KEPCO-led consortium signed a contract with the Emirates Nuclear Energy Corporation on December 27, 2009 for four APR1400s at Barakah in Abu Dhabi, and all four have been in commercial operation since the last of them entered service in September 2024. The disclosed amount was $18.6bn — KEPCO’s own SEC Form 20-F for FY2009 prints that figure and its scope: designing and constructing four 1,400MW units, supplying nuclear fuel for three cycles including initial loading, and providing technical support and training. The widely quoted “about $20.4bn” traces to the December 2009 announcement and carries no citation in any source we could reach (Sources: Emirates Nuclear Energy Corporation for the fourth unit entering commercial operation; the contract amount is KEPCO’s own, filed with the SEC as well as with Korea’s FSS, and subject to the escalation provisions the filing names; the December 2009 announcement’s “about $20.4bn” is the figure the World Nuclear Association and the wires carry, and we could find no document behind it. The KEDO figures are from KEDO’s own Turnkey Contract page and 2005 Annual Report — note the 34.5% belongs to the December 2003 suspension, not to the 2006 termination, which is a conflation that has travelled widely. Note also that the Korean filings call the plant “BNPP” and never “Barakah”: searching them for the English name returns nothing, which is its own small lesson in reading primary sources.) Barakah predates the disclosure trail this section uses — what makes the Czech interval measurable at all is that the contract filing reprints its own chain of clarification filings — so Barakah cannot be timed the same way from the filings we read. Korea’s most recent reactor export — and the only one whose headline-to-filing chain can be reconstructed from disclosure filings — is the Czech Dukovany 5 & 6 project. KEPCO E&C’s architect-engineering contract filing for it (rcpNo 20251212801320, filed December 12, 2025) contains this sentence, which we translate from the Korean: “This disclosure is the confirming disclosure of the clarification (undetermined) regarding the report of the content ‘Beat France in Europe, wins ₩24tn Czech nuclear order’ and similar, reported in the Chosun Ilbo and others on July 18, 2024.”

The filing then lists the chain: six separate “clarification of rumour or report (undetermined)” filings, the first of them dated the day the headline itself ran, running to the confirming disclosure of December 12, 2025. First headline to signed contract: sixteen months and twenty-four days. The six dates themselves, and what changed between them, are chronology; they belong to our timeline of this industry, not here.

Two properties of that number are worth stating plainly. It is one observation, so it is a reference point and not a distribution — the true interval for a US project could be much shorter if political will is stronger, or much longer, or infinite. And it describes a project that did happen: this is the interval on a success, not the interval on an average.

What it does establish is that a gap of many months between confident reporting and any filing is normal in this industry, in this country, on this kind of contract — and that a Korean company facing press reports it cannot confirm has a disclosure regime that makes it say so, repeatedly, on the record, with dates. The chronology of the Czech project itself belongs elsewhere in our coverage; here the interval is only a ruler.

🏢 Which Listed Companies Are Actually Exposed, and to What?

Four Korean listed companies sit in this chain with four different exposures — and the largest disclosed US contracts among them are gas turbines, not reactors.

One line each, because each of these companies has a full piece of its own and repeating their financials here would be the wrong article.

• Doosan Enerbility (034020) — the only domestic maker of large gas turbines and the nuclear steam supply system supplier on the Czech project; also the owner of the two withheld US contracts in the census above, both gas turbines, each at least about $301.7M — and, before them, of the largest US-soil contract in the whole census, the USD 571,000,000 Ukudu combined-cycle plant on Guam, which finished in October 2024.
• KEPCO E&C (052690) — it designs the reactor. It is the engineering company behind the APR1400 whose certification sits in Appendix F, and the holder of the €724,958,299 Czech architect-engineering contract.
• KEPCO KPS (051600) — it maintains plants. Its Westinghouse relationship runs the opposite way to the one in the headlines: its own filings label it 기술도입(원자력), nuclear technology intake, for “primary-side major-component maintenance technology cooperation”, and its current term is printed in the filing as ‘25.12.31. ~ ‘35.12.30.(10년) — ten years, from December 31, 2025 to December 30, 2035 — in an agreement the filing describes as “concluded and in operation.” The filing does not use the word “renewal”; that this term replaced an earlier one is our inference and not the document’s statement (Source: DART rcpNo 20260323000091, FY2025 annual report, major technology cooperation agreements table, as of Dec 31, 2025; reprinted unchanged as of Jun 30, 2026 in rcpNo 20260814001938). It is not a royalty negotiation waiting to happen; it is a signed, current, decade-long licence in.
• BHI (083650) — it makes plant equipment, and it has the longest US-region history of the four. It accounts for seven rows above, six of which have a region field naming the United States; the seventh is the unresolved June 2016 one. Three name combined-cycle plants. Two are nuclear — the 2010 Toshiba contracts titled nuclear power plant auxiliary equipment, which makes BHI the second of only two companies here ever to put nuclear scope and a US region field on the same form. One is vessels, tanks and heat exchangers, and one is a 2007 feedwater-heater order from Bechtel; neither states a plant type, so we do not classify those two. The last of them finished on July 7, 2017, and four of the six US-region contracts were voluntary filings — 자율공시 — which a company is free not to make.

And the structural point that matters most to a foreign investor: the counterparty on every Korean reactor export contract is Korea Hydro & Nuclear Power, which is unlisted. It is corp_code 00382001 in DART, a subsidiary of Korea Electric Power (KOSPI 015760). You cannot buy the signing party. You can buy its suppliers, which is a different exposure with different economics. How many contractual steps sit between a foreign investor and the party that actually signs — and why those steps cannot simply be added up — is the subject of our filing map of the Korean nuclear export chain. Here it is enough to note that the entity named in the counterparty field is not investable at any price.

✅ What Would Have to Appear for This to Be Real?

A government document, a Korean contract filing, or a US licence application. As of 1 October 2026 the first of the three exists and the other two do not — which is the whole point of keeping them as three separate rows rather than one verdict.

This is the part of the article designed to outlive it. The next time a report appears saying Korea has agreed to build reactors in America, these are the three places to look, and all three are free and public.

🔴 What We Got Wrong

Four premises we held while researching this piece turned out to be false. Three failed the same way — we took reporting as record without opening the government’s own documents — and the fourth is one we did to ourselves, twice, with the same method error.

We publish these because a piece about the gap between reporting and documents has no standing if it hides its own instances of that gap.

The third and fourth are the instructive failures, and they are the same failure. Asserting that something does not exist is a much stronger claim than asserting that something does, and it requires a census rather than a search. We wrote that sentence about the filings and then broke it on the clarifications a fortnight later. That is why this article now carries a coverage percentage and a named 98-filing hole in one census, and a stated pagination depth, keyword list and exclusion list in the other, instead of the words “none” and “seven.”

📚 Lingo Check

🎯 Why It Matters for K-Export Stars

This site exists because a large amount of decision-relevant information about Korean companies is published, free, in Korean, and never crosses into English. Usually that means a disclosure footnote. This week it means something blunter. Every one of the fourteen documents answers a Korean outlet — KBS, Edaily, Money Today, Herald Business, Yonhap, Hankyoreh, Hankyung, JoongAng Ilbo — named outright in the attachment filename, on a public board, in Korean-only PDF and HWPX files. The earliest is dated May 8, 2026, four months before the story reached English at all. No English-language reader encountering the headline has any way to know they exist.

Nothing in this article required special access or analytical cleverness. It required opening a Korean government board, paginating it to the bottom rather than reading its first page, reading the structured region field in the 1,101 filings whose text we could obtain out of a population of 1,199, and querying two American federal APIs. The wires had the same afternoon we did. The Korea Discount is usually explained in terms of governance and geopolitics; a real part of it is simply that the primary record is in a language most holders of the stock do not read. That is also the part an individual investor can do something about — see our guides on how to access Korean shares, how to read Korean investor flows, and what the Value-Up programme actually commits companies to.

And we will keep score on this one. Every claim in this piece carries the date we checked it — the Korean systems on September 29, 2026, the American ones on September 10 or 11 where we could not reach them again — and each of the three checklist rows has a specific document that would flip it. If a project is announced by either government, the checklist gets its first tick — and, we would guess, the other two boxes will stay empty for a while, which will itself be the story. We correct our published work in public when it turns out wrong; the corrections policy explains how.

Disclaimer: This article is for informational and educational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. Figures are transcribed from primary filings and government documents as cited, each carrying the date on which we verified it — the Korean ministry board and DART on September 29, 2026, the filing census and the American regulatory records on September 10 and 11, 2026; prices, policies and regulatory status change. Statements about what documents do not exist are bounded by the search scope described in the article, including a named 8.2% gap in the filing census and several sources we could not access. K-Export Stars is not affiliated with any company, ministry or agency mentioned. Do your own research and consult a licensed adviser before making investment decisions.