Foreign Flow Watch: Korea’s Nuclear Chain Rose Up to 41% and Foreigners Sold All of It — Week of September 11, 2026

The KOSPI rose 3.33% this week, and 81.1% of that gain came from two stocks. Foreigners sold $2.05bn of those two combined. Two days before that window closed we published a four-part decode of Korea’s nuclear export chain, with a hub that landed on the Friday. Every one of those four companies rose this week — one of them by 41% — and foreigners were net sellers of the basket over five sessions and over twenty. We are not claiming those facts are connected. We are claiming that if you read only the index print, or only one window of flow data, you would have got this week exactly backwards.

This is Foreign Flow Watch, our weekly read of where global capital actually moved in Korean equities. Every figure below is computed from settled exchange data for the 2,549 KOSPI and KOSDAQ common stocks in our published universe. Week of 7–11 September 2026, five trading sessions. Previous issue: Issue #8, week of 4 September. The full series lives at the Foreign Flow Watch hub.

📈 Both boards rose, and almost all of it was Monday

🧮 Two stocks were 81.1% of the rally — and foreigners sold both

Aggregate market value of the 828 KOSPI common stocks present in our reference set on both dates rose from $3,980.2bn to $4,112.8bn, a gain of $132.6bn or 3.33% — the index’s figure to two decimal places. That the two agree is our check that the reference set tracks the index. The set excludes preferred shares, which is why it is 828 names and not the 943 KOSPI lines a price screen shows.

The pair added $107.6bn of the index’s $132.6bn — 81.1% of the rise, on 51.6% of the market value. Over the same five sessions foreigners were net sellers of $2,051.2M across the two, which was 90.9% of their $2,255.5M net selling across both boards. Two issues ago the pair was 196% of a decline; last week it was 16.9% of one. The useful conclusion is still not any single one of those numbers. It is that the KOSPI’s weekly print carries almost no fixed information about what happened to Korean companies, because roughly half of it is two memory stocks whose relationship to the other half changes week to week. We set out why that concentration exists in Korea’s chip supercycle.

🔁 Did foreigners sell Samsung and SK hynix this week?

Yes on the week and no on the first half of it: foreigners were net buyers of both stocks through Tuesday and turned into heavy net sellers from Wednesday, so the weekly net figure describes neither half. This is the same failure mode a single flow window always has, compressed into five days instead of twenty.

Read only the weekly number and you have a story about foreigners exiting Korean memory. Read the sessions and you have foreigners adding roughly $2.5bn across the pair into Monday’s 4.61% rip and selling roughly $4.6bn back out as the index rolled over from Wednesday’s high. Both descriptions come from the same five rows. The second one is the one a position-sizing decision actually depends on.

☢️ The nuclear chain rose up to 41%, and foreigners sold all of it

On 9 and 11 September we published a five-part map of Korea’s nuclear export chain: a hub organised by how many steps each company sits from the entity that signs the contract, plus one decode each of Doosan Enerbility (the fabricator), KEPCO E&C (the designer), KEPCO KPS (maintenance) and BHI (balance-of-plant equipment). The four company pieces went out on 9 September, before the week’s flow data was complete — two of the five sessions and the last of the twenty were still to come — and the hub on 11 September. The basket was therefore named in public before we could have seen the full week, which is the only reason it is worth reporting a basket number at all; it is not a basket assembled after the fact to fit a result.

Three of the four were net sold in both windows. The fourth, KEPCO KPS, was net bought by 1,870 shares over five sessions — $0.1M, which on a $1.7bn company is indistinguishable from zero — and net sold over twenty. There is no window in this data in which foreigners were meaningfully buying this basket, and every name in it rose in both windows, by between +6.71% and +52.30%.

One more thing the filing record says about this week, because it is the part no price chart contains. Across the four companies, DART carried no new nuclear order between 1 and 13 September 2026. The only single-supply-contract filings in the window were amendments to legacy non-nuclear work: Doosan Enerbility re-filed its Jafurah cogeneration contract on 1 September to fix the additional-scope end date at 31 March 2027 (rcpNo 20260901800761) — a gas cogeneration plant in Saudi Arabia, awarded 22 September 2022, counterparty Korea Electric Power Corporation, ₩536,466,320,000, which the filing itself states as USD 384,950,000 at the contract-date rate of ₩1,393.60/USD and 4.75% of 2021 revenue. BHI re-filed its Long Phu 1 contract on 10 September to fix an end date of 28 January 2027 after a counterparty change at 50% completion (rcpNo 20260910900425) — a Vietnamese power plant awarded in 2016 — the filing names the counterparty as the Long Phu 1 project management board and the scope as power-generation equipment, and does not state the fuel — ₩9,062,933,216, stated in the filing as USD 7,791,380 at ₩1,163.20. We quote the dollar figures the filings print rather than reconverting them, because the companies fixed those rates on the contract dates and a fresh conversion would invent a number that appears nowhere.

🔬 The chip-equipment selling was one session, not one week

We have a second published basket to test this against, and it cuts the other way. On 6–8 September we mapped the Korean semiconductor supply chain across ten filings, covering nine listed suppliers. Over the five sessions foreigners net sold $531.2M of that basket. Over twenty sessions they net sold only $254.8M — meaning across the fifteen sessions before this week they were net buyers of roughly $276M.

$492.6M of the basket’s $531.2M of five-session selling — 93% of it — happened on Thursday 10 September. Thursday was also larger than the whole week across both boards: all-market foreign net that day was −$2,870.8M, against −$2,255.5M for the five sessions combined, because Monday and Tuesday were net buying. Two of the nine names, PSK and HPSP, are net bought over twenty sessions and net sold over five. If you had written “foreigners are exiting Korean chip equipment” from the weekly number, the twenty-session column would have told you they had been accumulating it until Thursday, and the daily column would have told you Thursday was one event.

We have made precisely this mistake in this series before. Issue #3 declared a rotation out of SK hynix and into its holding company SK Square on a single week’s net figures, and a longer window reversed it. For the record, that pair passes the test this week: foreigners net bought SK Square +$261.4M over five sessions and +$316.6M over twenty, while net selling SK hynix in both. Two agreeing windows is still not a rotation — it is two facts that agree — but it is at least the minimum we now require before saying anything at all.

📊 The export-champion scoreboard: 5 sessions against 20

Our fixed twelve-name export universe, so issues stay comparable. The right-hand column is the one to read first.

There is a tidy story available in the top half of that table — money leaving memory and arriving in defence and bio — and four of the twelve rows refuse it. Hanwha Aerospace and Korea Aerospace Industries, the two largest defence names here, were bought over five sessions and sold over twenty. Samyang Foods and APR were the exact opposite. So the five-session window says “rotation into defence”; the twenty-session window says foreigners have been reducing two of the three defence names all month and were buying consumer names until this week. Only LIG Nex1 supports the rotation reading in both windows, and at +$10.8M over five sessions it is too small to carry the claim on its own.

🏦 Is the fourth buyer still buying?

Yes, at an undiminished pace, and both programmes have now passed their halfway marks about a third of the way into their filed windows. Korea’s exchange feeds publish three investor categories — foreigners, institutions and individuals — and their net purchases do not sum to zero. The remainder, which the exchange calls other corporations, includes an issuer buying its own stock, and it is the only way to see a Korean buyback executing in near-real time.

Two things changed this week and both are worth flagging against ourselves. First, Samsung has now bought back more than half of its programme on won amounts and 55.55% on share counts in 15 of roughly 63 sessions; the share-count figure runs ahead of the won figure because the filing’s share count was struck at a lower reference price than Samsung has been paying this week. Second, SK hynix’s daily band fell below 620,000 shares for the first time, printing 600,643 on Thursday and 603,890 on Friday against a floor of 632,027 last week. That is a 5% slowdown, not a stop, and it is the first directional wobble in 17 sessions. If it continues it is the leading edge of the test we set ourselves last week; one week is not enough to call it.

We checked DART for every filing by both companies from 5 to 13 September 2026. Neither has amended, suspended or terminated its programme, and neither has filed an execution report — the authoritative quarterly document that would confirm or demolish the figures above. Both windows remain in force as filed.

🗂️ A $1.45bn purchase our flow data will never see

On 9 September 2026 Samsung Electronics’ chairman, Lee Jae-yong, filed an insider trading-plan report (rcpNo 20260909000516, DART). The contents, verbatim from the form:

Two notes on that table. Every figure in it is as filed; the percentages the filing prints are all against total shares of 6,648,649,811, which is common plus preferred, not the 5,846,278,608 common shares a Korea Exchange screen shows. On the common-share count the purchase is 0.123% — our arithmetic, not the filing’s. We flag it because the common-share denominator is 12.1% smaller, so any percentage computed on it comes out 13.7% larger: the chairman’s stake reads 1.47% of total shares or 1.67% of common, and quoting one against the other is the kind of mismatch that quietly wrecks a comparison.

Korea requires directors and major shareholders to pre-announce trades of this kind, with a quantity, a price basis and a settlement date, before they happen. Most markets learn about a transaction like this afterwards. Here it is a calendar entry: 12 October 2026.

The reason it belongs in a flow column rather than a governance one is the first row of that table. This is an off-exchange transfer, so it will not appear in any KRX investor-flow series — including the dataset every number in this issue is built from. A $1.45bn change of ownership in the largest company in Korea will pass through October leaving no trace in the four-category flow data at all. That is a hard limit on what this series can see, it applies to our residual method as much as to the headline numbers, and we would rather state it in the week it became concrete than in a footnote later. Samsung separately filed a change in shares held by the largest shareholder on 8 September (rcpNo 20260908800624) reporting Samsung Life’s special account down 225,066 shares to 1,151,617,799, or 17.32% of total shares — a separate, much smaller matter, and a reminder that ownership disclosure and flow data are two different records that only sometimes agree.

One further line from the same week, continuing an item from Issue #8. On 9 September SK hynix answered a Korea Exchange inquiry-disclosure demand dated 10 August about reports that it was pursuing a sale of a stake in its ₩4tn ($2.99bn) Chongqing packaging plant (rcpNo 20260909800559). The answer: the company is “reviewing various measures to strengthen the competitiveness of the packaging business, but nothing has been determined as of now,” with a commitment to re-disclose by 8 December 2026. Issue #8 reported a separate undetermined answer on Solidigm with a re-disclosure date of 3 December 2026. SK hynix shareholders now have two dated obligations to say more, five days apart, on two different structural questions. Neither is news. Both are appointments.

🧾 Scorecard: how Issue #8 held up

Issue #8 published five testable conditions. One resolved mostly in the data’s favour, one fired against us for the second week running, two remain open, and the weak number got fixed at source while the underlying exposure grew.

Our running record of grading ourselves: Issue #4 scored Issue #3 at two failures of three; Issue #5 scored Issue #4 at three holds and one open; Issue #6 scored Issue #5 at two failures of three; Issue #7 scored Issue #6 at three tests that could not run; Issue #8 recorded one of its own tests as badly written. This week that same test failed again, for a different reason, after we rewrote it. We print the bad weeks at the same size as the good ones.

🧩 Prove this issue wrong

• If the nuclear basket’s foreign flow turns net positive over both windows within four weeks while the prices hold, then “sold in both windows” was a two-week artefact of the late-August tape and not a description of foreign positioning at all. Testable in every issue from here.
• Third restatement of the third-name test, now with a size floor: the residual counts as anomalous only if it exceeds 1% of shares outstanding and the company’s market value is above $1bn and no large-holding report, treasury-share filing or largest-shareholder change appears within five business days. Under this version, this week produces four candidates rather than twenty-two — DL E&C (+1.59% of shares), Hyundai Department Store (+1.28%), Lotte Shopping (+1.03%) and SK IET (−1.09%) — and the 3 September POSCO names would have triggered correctly. None of the four is yet resolved, and we are not going to pretend otherwise. Two have filings inside the window that do not explain the residual: Lotte Shopping‘s 11 September filing (rcpNo 20260911000512) is Shin Dong-bin selling on-exchange as an individual, −203,150 shares and 9.81%→9.09%. Our residual for the same week is the +1.03% above — a purchase, by the corporate category, and a figure from our flow data that appears nowhere in that filing. Wrong investor category and wrong direction, so the filing does not explain it; SK IET‘s 10 September filing (rcpNo 20260910800520) is a disposal of shares in another company, which says nothing about who traded SK IET’s own stock and is not one of the three filing types the test names. For DL E&C and Hyundai Department Store the five-business-day window runs to 18 September 2026 and has not elapsed. All four get scored in Issue #10.
• If SK hynix’s daily residual band stays below 620,000 shares for two more weeks, the run-rate projection of 36.7 sessions is stale and the completion date moves later. This is the specific, dated version of the November test, and it is checkable next Saturday.
• If either execution report, when filed, shows materially less than ₩18.52tn ($13.84bn, SK hynix) or ₩7.72tn ($5.77bn, Samsung) bought through 11 September, our residual attribution is an upper bound we leaned on too hard. Still the authoritative check; still not yet available.
• If the Lee Jae-yong purchase settles on 12 October and any visible trace of it appears in the four-category flow data, then our claim that off-exchange transfers are invisible to this dataset is wrong, and the residual method is measuring something broader than we think. We will look on 12 and 13 October and publish the answer either way.
• Our weakest number this week is the 20-session dollar column, for the reason in the scorecard: the won moved 5.4% across the window, so a 20-session dollar figure is method-dependent by about 4% for a stock whose flow holds one sign and by up to 17% for one whose sign flips inside the window (PSK, 16.8%). Share counts and won figures are exact; five-session dollar figures carry about 1.3% of the same exposure.

📚 Lingo Check

🎯 Why it matters for K-Export Stars

A single KOSPI print told you Korea rose 3.33% this week. Underneath it, four-fifths of that gain came from two companies, of which foreigners sold $2.05bn combined; retail sold $7.15bn; institutions bought $3.06bn; and the largest bid in the market was again two corporate treasuries executing programmes they disclosed, in Korean, in August. Meanwhile a nuclear supply chain rose between 7.64% and 41.26% on a week whose disclosure record contains two end-date corrections on a Saudi cogeneration plant and a Vietnamese power plant, and nothing nuclear at all.

None of that requires a view on memory pricing or on reactor exports. It requires reading the filings, matching them to settled flow data, and printing the disagreements instead of the tidy version. The filings state the won committed, the start date, the purpose, the counterparty and — for an insider’s plan — the settlement date. The flow data states what got bought on the exchange each session. The gap between the two is where most of the useful information about Korea sits, and it exists because Korea makes issuers and insiders say in advance what most markets only reveal afterwards. For how these flow categories are built and why they mislead as often as they inform, start with how foreign and institutional flows move Korean stocks; for the reactor chain in filing terms, the nuclear export chain map and our longer-standing Doosan Enerbility valuation; for the two stocks at the centre of the index in dollar terms, Samsung and SK hynix stock prices in USD; and for what a cancellation actually does to a share count, our Value-Up programme guide and the Celltrion case.

Disclosure of interest. The author holds positions in Samsung Electronics (005930) and SK hynix (000660), two of the stocks this issue discusses, and in other Korean equities not named above. The author holds no position in any of the four nuclear-chain companies or the nine semiconductor suppliers listed in this issue. No position was opened or closed in any stock named above in connection with this article.

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. Figures are as of 11 September 2026 and go stale quickly; prices, flows and programme-completion estimates in particular. Do your own research and consult a licensed adviser before investing.