SK hynix (000660): What It Actually Makes, Who Buys It, and Where the Money Goes (2026)

Every English-language story about SK hynix starts in the middle. It is the HBM company, the one whose profits went vertical and whose share price then halved. Almost none of them stop to say what the company actually manufactures, who writes the cheques, or where the factories are. This article is the part everyone skips — built from the Korean-language filings, because that is where the answers live.

SK hynix (KRX: 000660; ADRs SKHY on Nasdaq since July 2026) is the world’s largest supplier of high-bandwidth memory. That much is in every wire story. What follows is the layer underneath — sourced from DART, Korea’s mandatory electronic disclosure system, and from Korea Exchange settled data.

A note on units. Every money figure in this article is shown in US dollars. Won is converted at ₩1,375.67/$ throughout — the rate carried in our data as of 28 August 2026. Share prices are the one exception: they are shown in dollars with the won price alongside, because a reader checking a Korean quote screen needs the won number.

Correction, 7 September 2026. An earlier version of this article said SK hynix had never filed a Value-Up plan. That was wrong. The company filed 기업가치제고계획 on 27 November 2024 and an implementation update on 27 November 2025. The passage below has been rewritten around what those filings actually say. We are leaving this note because the original sentence invited readers to check it.

🏢 What Does SK hynix Actually Make and Sell?

It makes two kinds of memory chip — DRAM and NAND flash — and essentially nothing else; the company’s own annual report describes semiconductors as more than 90% of revenue and reports the whole company as a single segment.

The distinction between its two products is the one thing a new investor has to hold onto. DRAM is volatile memory: fast, and it forgets everything when the power goes off. NAND flash is non-volatile: slower, and it remembers. Your phone uses both — DRAM to run apps, NAND to store photos. The annual report puts it in exactly those terms (Source: DART annual report FY2025, rcpNo 20260317000635, filed 17 Mar 2026).

The annual report adds one more line of business that the revenue table does not name: alongside memory, the company “also runs a foundry business” — contract manufacturing of other companies’ chip designs. That is the most likely home of the small “Other” line in the revenue table below, although the filing does not say so explicitly. The product that made the company famous, HBM, is a variety of DRAM — many DRAM dies stacked vertically and wired together so an AI accelerator can pull data through a much wider pipe. SK hynix was first to mass-produce HBM3 in 2022 and first to ship 12-layer HBM3E in 2024. If you want the technology roadmap rather than the corporate structure, our earlier HBM4E outlook covers it, and the Korea chip supercycle overview explains where memory sits in the wider Korean market.

The corporate basics, from the same filing: headquarters at 2091 Gyeongchung-daero, Bubal-eup, Icheon, Gyeonggi Province, about an hour southeast of Seoul. Chief executive Kwak Noh-jung. Listed on the KOSPI since 26 December 1996. Fifty-three consolidated subsidiaries at end-2025, rising to 57 by mid-2026. The semiconductor workforce of the filing entity numbered 34,549 at end-2025 — that is SK hynix Inc.’s own headcount, not the consolidated group’s, so overseas subsidiaries such as Solidigm are not counted in it — with average tenure of 12.3 years for men and 15.7 for women — long, by the standards of most technology companies. Background entity reference: Wikipedia; the company’s own investor materials are at skhynix.com.

📊 Where Does SK hynix’s Revenue Actually Come From?

Three-quarters DRAM, a quarter NAND — and you can only find that split in the Korean-language interim filing, because the annual report presents revenue as a single undifferentiated line.

This is worth dwelling on, because it is the clearest example on this site of why we read the primary source. The FY2025 annual report’s product table reads, in full: business division “semiconductor”, item “DRAM, NAND Flash etc.”, revenue ₩97,146,675 million, share “100%”. One row. No split. An investor reading only that document cannot tell whether SK hynix is 95% DRAM or 60% DRAM.

The half-year report filed on 14 August 2026 contains a segment note that does split it. Here is what it says.

Source: DART half-year report, segment note, rcpNo 20260814003509, filed 14 Aug 2026. Won figures: DRAM ₩97,641,379m, NAND ₩33,534,133m, other ₩719,521m, total ₩131,895,033m. Rows sum exactly to the disclosed total. Converted at ₩1,375.67/$.

Two things follow. First, NAND is a quarter of this company, not a rounding error — and its share rose during the AI boom, from 20.0% to 25.4%, because NAND revenue quadrupled while DRAM revenue tripled. Second, this is a partial answer to a question we left open in our analysis of the second quarter, which asked why a DRAM leader trades in line with pure NAND names. A quarter of the top line is one reason. It is not the whole answer: the segment note is organised by product, not by legal entity, so it still does not tell you how much of that NAND comes from Solidigm, the former Intel business SK hynix owns outright. That split remains undisclosed.

And by geography?

Nearly two-thirds of SK hynix’s revenue is billed to customers in the United States and about a quarter to customers in China; roughly half of one percent comes from Korea.

The same note breaks revenue down by customer location. This is where the company’s exposure looks least like a Korean company’s.

Source: DART half-year report rcpNo 20260814003509 and annual report rcpNo 20260317000635. Regions sum exactly to disclosed totals in every period shown. Caveat: IFRS geographic disclosure is by customer location — that is, the entity billed — not by where the chip physically ends up. A part invoiced to a US corporation may be installed in a data centre anywhere.

About one dollar in every two hundred of SK hynix’s revenue comes from Korean customers. This is a Korean company only in the sense that its factories, its workforce and its listing are Korean. Its income statement is a claim on American and Chinese capital spending.

Is SK hynix actually a global leader, or only a Korean one?

In DRAM, yes — its world market share has run between 29.9% and 39.1% over the past three years and was 34.3% in the most recent quarter disclosed; in NAND it is a mid-sized player whose share has round-tripped rather than climbed. The annual report gives five observations on a revenue basis, sourced by the company to IDC: DRAM 29.9% (2023) · 33.4% (2024) · 36.3% / 39.1% / 34.3% (2025 Q1–Q3), and NAND 19.6% · 21.6% · 16.1% / 22.0% / 19.7%.

The instructive contrast is not the size of the swing — in percentage points the 2025 quarters moved 5.9pp for NAND against 4.8pp for DRAM, which is similar. It is the direction relative to where each started. Every 2025 DRAM quarter sat above the 2024 annual figure of 33.4%. NAND spent the first quarter of 2025 at 16.1% — below not only 2024 (21.6%) but 2023 (19.6%) — and by the third quarter had returned to 19.7%, roughly where it began in 2023. Measured against its own average, NAND’s 2025 swing was about 31% versus 13% for DRAM. So the quarter of the company that is NAND is also the quarter whose competitive position has gone sideways across three years while DRAM’s advanced. We published the full quarterly table in our earnings-quality analysis; the disclosure is in the FY2025 annual report, rcpNo 20260317000635 (DRAM data IDC Nov 2025, NAND data IDC Dec 2025). These are company-reported figures; we cannot independently verify IDC’s underlying data.

👥 Who Actually Buys SK hynix Memory?

Two customers each accounted for about 13% of first-half revenue — and that is materially less concentrated than a year earlier, when a single customer was 27.3%.

Korean disclosure requires a company to report any single external customer exceeding 10% of consolidated revenue, but permits it to anonymise them. The half-year note reads, in the original: “당반기 중 단일 외부고객으로부터의 매출액이 연결회사 전체 매출액의 10%를 상회하는 고객 (가)로부터 발생한 매출액은 17,608,702백만원 (13.35%)이며, 고객 (나)로부터 발생한 매출액은 17,187,421백만원 (13.03%)입니다.” — “Revenue from Customer (A), a single external customer exceeding 10% of the consolidated group’s total revenue, was ₩17,608,702 million (13.35%); revenue from Customer (B) was ₩17,187,421 million (13.03%).”

Two honest limits. The filing does not name either customer, and it does not state whether “Customer (A)” in 2026 is the same legal entity as “Customer (A)” in 2025 — the labels are per-period. So the correct reading is about structure, not about any named buyer: a year ago more than a quarter of the company depended on one relationship, and now the top two are roughly equal and each about half that size. We are not going to tell you which accelerator vendors those are. The filing does not say, and inferring it from press coverage would be guessing dressed as analysis.

🗓️ How Did SK hynix End Up Inside the SK Group?

It was a construction company, then Hyundai’s chip arm, then a ward of its creditors after a near-collapse that wiped out 87% of every share ever issued — and SK bought the survivor in 2012.

The corporate history section of the annual report begins in a place no English profile mentions. The company was founded in October 1949 as Kukdo Construction. It became Hyundai Electronics Industries in February 1983 — that was the entry into semiconductors — and Hynix Semiconductor in March 2001, after the Hyundai group broke apart.

What happened next is recorded in the share-count table rather than in any narrative. Of the 5,721,980,209 shares the company has issued in its lifetime, 4,990,449,799 were extinguished in a single capital reduction on 31 March 2003 — a 21-to-1 reverse split, disclosed in the annual report as “감자, 2003.03.31, 주식병합(21:1)”. That is 87.2% of all shares ever issued, erased in one line. It is the accounting scar of the creditor workout that kept the company alive. SK Group acquired control in March 2012, from creditors, of what is now one of the two most valuable companies in Korea. We are not going to tell you the 2003 reduction caused the 2012 price — the filings record the two events, not the negotiation between them.

Source: annual report corporate-history and share-count sections, rcpNo 20260317000635; half-year report share-count section, rcpNo 20260814003509.

One consequence of that January-2026 to July-2026 stretch is worth flagging for anyone reading old coverage. On 28 January 2026 the company answered a Korea Exchange query about press reports of a US listing by saying it was reviewing options but that nothing had been decided. Six months later it listed on Nasdaq. Both statements were accurate when made; the sequence is a useful reminder of how little a Korean “no decision yet” filing tells you about the timeline. We covered the listing itself in the Nasdaq debut piece, and what the ADR does and does not track in the ADR decoupling piece.

🏭 Where Does SK hynix Build All This, and What Is It Spending?

Four production bases, three R&D entities, fabs running at 100% utilisation 365 days a year — and $60.4bn of new construction approved by the board in 2026 alone.

The annual report describes the footprint as “four production bases and three research-and-development corporations, plus sales entities and offices in the United States, China, Singapore, Taiwan and Hong Kong”. The Korean sites are Icheon (headquarters, home of the M16 fab completed in February 2021) and Cheongju; the Chinese operations include Wuxi, Chongqing and a Dalian fab acquired from Intel in December 2021.

On utilisation, the annual report is unusually blunt: the fabs run on a four-crew, three-shift pattern, operated 365 days including public holidays in 2025, at an average utilisation rate of 100% — 267,670,560 available hours against 267,670,560 actual hours. Production capacity, measured at average cost rather than at selling price, was $30.0bn in 2025 against $25.6bn in 2024 and $25.3bn in 2023. Read the 100% with care: the report states that utilisation is computed from each fab’s staffing and yield, so it is the company’s own operating measure, not an independent statement that no tool was ever idle.

The forward spending is where the filings get interesting, because Korean rules require a board resolution on major facility investment to be disclosed individually, with an amount and an end date. Four such filings landed in 2026.

Sources, in order: DART rcpNo 20260225801974 (₩21,608,100m), rcpNo 20260722800829 (₩7,093,100m), rcpNo 20260807800602 (₩19,100,000m), rcpNo 20260807800613 (₩35,224,600m). The four won amounts sum exactly to the total shown. Each filing states that the amount and the dates may change as the projects proceed.

For scale: actual property, plant and equipment additions in the whole of FY2025 were $21.9bn. The board has committed nearly three years of that run-rate in eight months. Net PP&E stood at $64.6bn at 30 June 2026, up 14.7% in six months, with contracted-but-unbilled equipment purchase commitments of $44.7bn — up from $4.8bn six months earlier, a 9.2-fold increase in half a year (Source: DART half-year report commitments note, rcpNo 20260814003509; consolidated basis, ₩61,484,533m at 30 Jun 2026 against ₩6,667,863m at 31 Dec 2025). That is equipment ordered and not yet invoiced — spending already contracted for, on top of the board resolutions above.

Separately, on 29 June 2026 the company filed a forward-looking business plan sketching a much larger long-run programme: roughly ₩600tn for the Yongin cluster with a fourth fab targeted for 2033, ₩100tn for Cheongju, and ₩400tn for a wholly new “Southwest cluster” — about ₩1,100tn, or roughly $800bn in total [Company guidance — not a committed figure]. The filing labels itself as forward-looking, states that the numbers are “a guideline provided to aid understanding”, and says specific schedules will be disclosed when individual board approvals happen. Treat it accordingly: the $60.4bn above is money a board has voted on, and the $800bn is an ambition (Source: DART rcpNo 20260629800981, 29 Jun 2026).

🔍 The Clock in the Filings That Nobody Is Watching: China

The conditions China’s competition regulator attached to SK hynix’s purchase of Intel’s NAND business run for five years from December 2021 — meaning they expire in December 2026, four months from now — while a separate Chinese antitrust investigation, which the annual report still described as unpredictable, is no longer disclosed at all in the half-year report.

This is the section that exists because we read the Korean notes. Buried in the contingent-liabilities note of the half-year report is this, on the Intel NAND deal:

“연결회사는 2021년 중 완료된 인텔 NAND 사업 인수 1차 종결과 관련하여 중국 경쟁당국(중국 국가시장감독관리총국)으로부터 … 향후 5년간 중국 eSSD 시장에서 합리적인 가격정책을 유지하고 생산량을 확대할 의무 및 중국 eSSD 시장에 제3의 경쟁사가 진입하는 것을 지원해 줄 의무 등을 주요 내용으로 하는 조건들을 부과받은 바 있습니다. 따라서, 연결회사는 2021년 12월부터 5년간 동 의무들을 준수하여야 하며, 5년 후 동 의무들을 면제해 줄 것을 신청할 수 있습니다.”

“In connection with the first closing of the Intel NAND acquisition completed during 2021, the group had conditions imposed on it by China’s competition authority (the State Administration for Market Regulation), the principal contents of which are obligations to maintain reasonable pricing policy in the Chinese eSSD market and expand output over the following five years, and an obligation to assist the entry of a third competitor into the Chinese eSSD market. Accordingly, the group must comply with these obligations for five years from December 2021, and after five years may apply for their release.”

Read that carefully. SK hynix has been under a Chinese regulatory obligation to help a competitor enter a market it operates in — and to hold pricing “reasonable” — since December 2021. That window closes in December 2026. Release is not automatic: the filing says the group may apply, and that SAMR will decide by reference to the competitive situation in the Chinese eSSD market at that time.

Alongside it sits a second item, and what happened to it is the more interesting half. The FY2025 annual report, drawn up as of 31 December 2025, records that in May 2018 SAMR opened an investigation into whether the major DRAM makers violated Chinese antitrust law in their Chinese sales, and states “당기말 현재 연결회사는 동 조사의 최종결과를 예측할 수 없습니다” — “as of the end of the current period the group cannot predict the final outcome of this investigation.”

The 14 August half-year report no longer carries that disclosure at all. Its litigation note has been replaced by a generic paragraph about intellectual-property disputes, closing with the statement that no related liability is recognised. We searched the full Korean text of the half-year filing for the terms the annual report itself uses: “2018년 5월” and “예측할 수 없” return zero hits, against three and four respectively in the annual report. “반독점” survives twice, but both are the merger-remedy passage quoted above, not the investigation. We are not going to tell you why it was dropped — a company may remove a contingency because it was resolved, because it became immaterial, or because presentation changed, and the filing does not say which. What is checkable is the sequence: disclosed as unpredictable at the end of December, absent by the middle of August, and no provision ever recognised in between.

Neither of these is a prediction on our part, and neither is a reason to be bearish on its own. What they are is a pair of dated, checkable items sitting underneath a revenue line where China has grown from 18.5% to 24.6% of the group in twelve months. December 2026 is a date on which something either happens or is disclosed as not having happened. That is a better thing to watch than a narrative.

📈 Is the Business Getting Better or Worse — and Why You Must Not Use ROE Here?

Better, spectacularly — but the same set of numbers produced a $6.6bn loss three years ago, which is why return on equity is the wrong ratio for this company at any point in its cycle.

Source: DART consolidated financial statements — annual reports for FY2022–FY2025 and half-year report rcpNo 20260814003509 for H1 2026. Equity is period-end total equity; ROE is period net income over period-end equity, not average equity. The H1 2026 row is a six-month figure against a six-month-end balance sheet, which is why it is not comparable to the annual rows — annualising it gives an ROE above 100%.

The ROE column is the point. Over four consecutive full years this company reported 3.5%, −17.1%, 26.8% and 35.6% — a ten-fold spread between the two positive troughs and peaks, and a sign change in between. Nothing about the business’s competitive quality changed by ten times in those four years. What changed was the memory price cycle.

Worse, the denominator is unstable too. Total equity went from $87.7bn at the end of 2025 to $191.0bn six months later, because a very large unrealised valuation gain ran through the income statement and into retained earnings. Annualising the first half against that period-end equity produces an ROE above 100% — a number that tells you the ratio has broken, not that the company earned it. For a cyclical manufacturer the useful comparisons are operating margin through a full cycle, capacity, and balance-sheet capacity to survive the trough. ROE is not one of them.

💰 What Do You Actually Pay for SK hynix Today?

About $878bn of market value, which is 28.1× last year’s audited earnings, 8.4× annualised first-half operating profit after tax, and 4.5× annualised first-half total earnings — three honest answers to three different questions.

SK hynix closed at $1,202 (₩1,653,000) on 28 August 2026, up 153.9% year to date and 43.4% below its 52-week closing high of ₩2,919,000; the intraday high over the same period was ₩2,987,000. On 730,492,365 shares that is a market capitalisation of $877.8bn (Source: Korea Exchange settled close, 28 Aug 2026).

Our house valuation convention separates core value — what the operating business earns — from diluted value, which includes non-recurring and non-operating items. For SK hynix that distinction is not academic. It is the whole argument.

All three divide the 28 August 2026 market capitalisation by the stated earnings base and by 730,492,365 shares. 26.4% is Korea’s top corporate rate including the local surtax. Reconciling with our August article, which quoted 22.9×: that figure divided by the FY2025 disclosed EPS of ₩62,044, which rests on a weighted-average share count. On the same market-capitalisation basis used here it was 24.2× at the 7 August close, and the move from ₩1,422,000 to ₩1,653,000 takes it to 28.1×. Annualising a half-year at a cyclical peak flatters any memory company — we show the first two rows because the market quotes them, not because we lean on them. Book value at 30 June 2026 was $190.7bn attributable to owners, giving 4.6× book; adding the July ADR proceeds of ₩39.89tn — which the offering itself reported as $26.5bn at its own pricing rate, and which settled after the balance-sheet date — gives roughly 4.0× on a won basis.

The gap between 4.5× and 8.4× is the non-operating income we wrote about in August. Which number is right depends entirely on whether you think that income recurs. We do not think it does, which is why we anchor on the core-value row.

How does that compare with its peers?

SK hynix is the smallest of the three by market value and the highest-margin by a wide distance — 48.6% operating margin in FY2025, against 13.1% for Samsung Electronics as a group.

SK hynix and Samsung Electronics: Korea Exchange settled closes of 28 August 2026 and DART FY2025 audited statements. Micron: public market data as of 28 August 2026; we do not hold it as a primary source, so treat it as indicative to the nearest point. Samsung’s margin is a group figure spanning phones, displays and foundry as well as memory, so it is not a like-for-like memory margin — the comparison shows the difference in business mix, not in memory execution.

🏛️ Who Controls SK hynix, and What Do Minority Holders Get?

SK Square controls it with only 20.07% — an unusually thin grip for a Korean chaebol affiliate — and the shareholder return arrives almost entirely as buy-and-cancel, not as dividends.

The ownership table, as of 31 December 2025, is not what a reader primed on Korean governance would expect.

Source: annual report shareholder sections, rcpNo 20260317000635, as of 31 Dec 2025, against 728,002,365 shares then outstanding. These rows do not sum to 100% and are not meant to — the small-shareholder line is a separate DART statistic that overlaps in part with the institutions listed above it, so the column adds to 101.54%. The Capital Research and BlackRock figures are drawn by the company from those investors’ own large-holding reports and may lag. Foreign ownership stood at 50.6% on 28 August 2026, from a market-data vendor rather than a filing, so treat it as approximate. The three named institutions together hold 17.94% — close to the controlling shareholder’s own stake.

Two structural points a foreign investor should register. First, the largest shareholder holds barely a fifth. SK Square itself is a holding company whose main asset is this stake, which is why it trades at a persistent discount to the value of what it owns — we unpacked that mechanism in our SK Square NAV analysis, and it is a textbook instance of the wider Korea Discount. Second, 1,186,328 holders — 99.99% of the register by headcount — each own less than 1% of the company, and together hold 63.34%. The filing defines this line precisely as “발행주식 총수의 100분의 1에 미달하는 주식을 보유한 주주”, holders of less than one-hundredth of shares outstanding, counted by shareholder number on the register. It does not say how many are individuals, or Koreans: with foreign ownership around 50.6%, a large part of that block will be custodial and institutional. What it does establish is that no single holder outside the named five reaches 1%.

So how much is actually returned to shareholders?

Almost all of it now comes through cancellation: a $29.1bn buy-and-cancel programme resolved on 19 August 2026, against FY2025 dividends of $1.52bn.

The dividend is small. For FY2025 the company declared ₩3,000 per share, $1.52bn in aggregate, on a payout ratio of 4.90% — a yield of 0.18% at the current share price (Source: DART dividend disclosure, FY2025). That total combines a year-end payment with the quarterly ones; the most recent quarterly declaration, on 7 August 2026, was ₩375 per share, $199m in total, on a record date of 31 August 2026 (Source: DART rcpNo 20260807800589). That same filing closed with a commitment to finalise and announce additional shareholder-return measures during the third quarter.

On 19 August 2026 that announcement arrived: a resolution to buy back and cancel 24,070,000 shares for ₩40,004,340,000,000 — $29.1bn — purchased on-market between 20 August and 19 November 2026 through SK Securities (Sources: DART rcpNo 20260819800340 and rcpNo 20260819000254). That is 19.1× the company’s entire FY2025 dividend. If you judge this company’s shareholder return by its dividend yield you will conclude it returns almost nothing, and you will be wrong by more than an order of magnitude. We have used the full-year declared dividend as the denominator, not the quarterly run-rate; there is no disclosure indicating the year-end payment will be discontinued.

Three details in those filings that do not travel into English coverage:

• The share count is derived from the price, not fixed. The filing states that 24,070,000 was calculated by dividing the won amount by the previous day’s close of ₩1,662,000, and that the actual number will change with the share price during the three-month acquisition window and be restated by amended filing (rcpNo 20260819800340). Our own arithmetic on the 28 August close, not a disclosed figure: the same money would retire about 24.2 million shares. A weaker share price retires more stock. The announced 3.30% of shares outstanding is a floor, not a ceiling.
• There is a legal ceiling, and the company used 44.7% of it. The treasury-acquisition filing shows the calculation: a distributable-profit base of ₩91,554,763,368,271 ($66.6bn) less dividends and adjustments gives an acquisition limit of ₩89,448,011,686,863 ($65.0bn). The $29.1bn programme consumes 44.7% of that (Source: DART rcpNo 20260819000254, the treasury-share acquisition resolution — the ceiling table is in that filing, not in the separate cancellation resolution).
• The unrealised gains cannot fund it. Inside that ceiling calculation is a deduction of ₩12,239,000,231,484 — $8.9bn — described as “미실현이익”, unrealised gains, which Korean company law excludes from distributable profit. The mark-to-market income that made the headline earnings look enormous is, by statute, not available to pay for buybacks or dividends. The line item itself — that specific deduction, inside that specific ceiling table — is the part we have not seen quoted in English. The broader link between this company’s earnings quality and its distributable capacity we made ourselves in the August analysis, which covered the March capital-reserve reduction that created the headroom in the first place.

One absence is still worth recording. Korea’s Value-Up programme has a dedicated voluntary disclosure vehicle, 기업가치제고계획. SK hynix filed one on 27 November 2024, and an implementation update on 27 November 2025. The base plan’s third pillar is written as 「재무관리 방향성 및 신규 주주환원 정책 제시」. The 2025 update reports a lower debt ratio and ₩4tn of net cash under that pillar, and says nothing about shareholder return. The August cancellation is a large capital return by any measure; it was not filed under the Value-Up vehicle, and the distinction matters for anyone screening on that basis.

📌 What the Half-Year Report Settled — and What It Did Not

The half-year report confirms that a non-cash financial-instrument valuation gain accounts for 82.9% of first-half non-operating income, and confirms the KIOXIA holding sits inside long-term investment assets — but it does not itself name KIOXIA as the source of the gain.

In August we published an analysis arguing that most of the quarter’s record profit sat below operating profit, inferred that a KIOXIA revaluation was the most likely source, and named the specific document that would confirm or kill the inference: the half-year report’s investment-asset notes. That report was filed on 14 August. Here is the scorecard, because a falsification condition you never go back and check is not a falsification condition.

• Confirmed: the cash-flow statement backs out ₩63,178,974m — $45.9bn — of “금융상품평가이익”, financial-instrument valuation gains, as a non-cash item. First-half non-operating income (pre-tax less operating) was $55.4bn. So 82.9% of it is a valuation mark, not cash.
• Confirmed: long-term investment assets rose from $10.6bn to $62.0bn over six months, and the notes state that the KIOXIA investment is held inside that line, indirectly through BCPE Pangea Cayman2 Limited.
• Confirmed, and it matters: the notes disclose that SK hynix’s direct and indirect holding in KIOXIA is capped for a period, that its right to appoint directors is restricted during that period — restricted, the filing says, not absent — and that it cannot exercise significant influence over KIOXIA’s management. That is precisely why the stake is a fair-valued financial asset running through the income statement rather than an equity-method investment.
• Answered, and less reassuringly than we expected: we asked whether the stake is marked at KIOXIA’s listed price or at a fund’s own fair value. At 30 June the $62.0bn of long-term investment assets is 96.2% Level 2, 3.8% Level 3, and nothing at all in Level 1 — so no part of it is carried at a straight market quote. But the period-end split understates the judgement involved. The notes disclose that “당반기에 일부 장기투자자산의 평가방법이 변경되어 수준 3에서 수준 2로 이동” — “during the half-year the valuation method for some long-term investment assets was changed and they moved from Level 3 to Level 2, and the group recognises a change of level at the end of the reporting period in which the event or change in circumstances causing the transfer occurred.” On the filing’s own timing rule, then, the valuation gain was struck while the assets were still Level 3 — the most judgemental tier — and the move to Level 2 was recognised only at the period end. The Level 3 movement table sets it out, and it reconciles exactly:

Source: DART half-year report Level 3 movement note, rcpNo 20260814003509. Won: opening ₩14,547,099m, purchases ₩2,111,992m, disposals ₩10,655,502m, valuation ₩62,943,900m, FX ₩397,942m, reclassification ₩66,077,210m, closing ₩3,268,221m. The rows reconcile to the closing balance exactly. Two things a reader should take from it: the gain was recognised at Level 3, and $7.7bn of the position was disposed of during the half — which is the realisation we flagged as an open question in August.

• Not settled: the report does not attribute the valuation gain to KIOXIA by name. That step remains our inference. It is a well-supported one — the size, the timing and the disclosed location of the holding all fit — but it is an inference, and we are not going to upgrade it because it is convenient.

The other open item from that article has also closed. It noted that the company had committed to announcing additional shareholder-return measures during the third quarter, and called that “a falsifiable thing to watch”. On 19 August the $29.1bn cancellation was announced. Watched, and it happened.

🌏 Is Foreign Money Buying or Selling SK hynix?

Selling, in every window we can measure — but the pace is not the same across windows, and the sixty-session picture is the one that carries the story.

Source: our own investor-flow database, built from Korea Investment & Securities data, cross-checked against Korea Exchange settled closes. Method: each day’s net share count is multiplied by that day’s settled close and the daily figures are summed, then converted at ₩1,375.67/$. These are approximations of value traded, not execution prices. The price column is measured from the first session’s close inside each window, not from the close before it — and for the twenty-session row that choice decides the sign. Measured from the 30 July close of ₩1,322,000 the same window is +25.0%, because 31 July is inside it. The five- and sixty-session rows do not flip on the same test (−4.5% and −30.0% respectively from their prior closes).

Note the shape, and note how fragile the twenty-day number is. Foreigners sold more over five sessions than over twenty, which is only possible because there was heavy buying inside the longer window — and almost all of it sits on one day. On 31 July 2026 the shares rose 29.95% and foreign investors bought +$2.75bn in a single session, as much as the whole 12–18 August stretch combined (+$2.79bn). 31 July is the first day of the twenty-session window. Drop that one boundary day and the twenty-day figure goes from −$1.73bn to −$4.49bn, 2.6 times larger. The sign is the same in every window we ran, so the direction is not an artefact; the magnitude very much is, and any twenty-day flow number for this stock should be read with its start date in hand. If you want the mechanics of why these flows move Korean prices at all, we cover it in how foreign and institutional flows move Korean stocks, and we publish a running weekly read in the Foreign Flow Watch series.

The sixty-session row is the one to sit with: foreigners net sold $21.57bn while Korean retail net bought $16.78bn, into a 28.1% decline. That is the register from the table above changing hands in real time.

🏷️ Which SK hynix Ticker Are You Actually Buying?

Only 000660 on the KOSPI is the operating company; SKHY on Nasdaq is an ADR worth one-tenth of a common share and has traded at a persistent premium; and 402340 is SK Square, the holding company that owns 20.07% of SK hynix — a different security with a NAV discount and other assets attached.

The ADR is not a free substitute for the home share: it has consistently traded above parity, and the won/dollar rate sits between the two prices, so SKHY can move when the Seoul line has not. We work through both effects in the ADR decoupling piece and set out the tickers, the conversion arithmetic and the look-alike symbols in the guide to Korean chip prices in dollars. If you would rather buy the Seoul line directly, our guide to buying Korean stocks as a foreign investor and the full list of Korean ADRs cover the practicalities and the tax.

📚 Lingo Check

🗂️ Our SK hynix Research Timeline

This article is the reference layer. Everything below is analysis that sits on top of it, in the order we published it.

🎯 Why It Matters for K-Export Stars

Because the numbers that decide how you value this company are filed in Korean, in a system English-language coverage does not read.

This site exists because the most important disclosures about Korean companies are written in Korean, filed in a Korean system, and never translated. SK hynix is the cleanest demonstration of that we have found.

The single most-wanted number about this company — how much of it is DRAM and how much is NAND — is absent from the annual report, which shows one undifferentiated line, and present in an interim segment note that exists only in Korean. The customer-concentration collapse from 27.31% to 13.35% is in that same note. The December 2026 expiry of a Chinese regulatory obligation to help a competitor enter a market is in a contingent-liability paragraph. The fact that mark-to-market gains are legally barred from funding buybacks is a line item inside a buyback filing’s ceiling calculation.

None of these are secret. They are simply in a language and a system that English-language coverage does not reach into. That gap is not a trading edge — it is a comprehension edge, and it is what we are here to close.

Disclosure of interest. The author holds a small long position in SK hynix (000660) — four shares — as part of a personal Korean equity portfolio, and also holds Samsung Electronics (005930), which appears in the peer table above. No position is held in Micron, KIOXIA or SK Square. This article was researched from primary filings and no compensation of any kind was received in connection with it.

Sources and method. Annual report FY2025: DART rcpNo 20260317000635. Half-year report H1 2026: rcpNo 20260814003509. Facility investments: 20260225801974, 20260722800829, 20260807800602, 20260807800613. Long-term investment plan: 20260629800981. Buyback and cancellation: 20260819800340 and 20260819000254. Quarterly dividend: 20260807800589. Prices, share counts and market capitalisation from the Korea Exchange, settled closes only. Investor flows from our own database built on Korea Investment & Securities data. Micron figures are public market data as of 28 August 2026 and are not primary sources for us. All dollar conversions use ₩1,375.67/$ as of 28 August 2026. Figures are point-in-time as dated; prices and multiples go stale quickly.

We log corrections rather than editing silently — see our corrections policy.

This company is one of ten in our K-Semiconductor Filing Map — a supply-chain map where every cell is marked by what the filing itself supports, and each one links to the receipt number it came from.

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 drawn from public filings and market data as dated, and may be superseded. Investing in Korean equities involves risk of loss, including currency risk. Do your own research and consult a licensed adviser before making any investment decision.