Korea's Leverage Crash: SK Hynix, the KOSPI, and What Actually Broke

I wrote the first version of this piece after four days of a story that looked done. It was not close to done. What started as a record Nasdaq listing turned, over the following four weeks, into South Korea's single worst month for equities on record (worse, by some measures, than any single month of the 1997 Asian Financial Crisis or 2008), with 1.2 million retail accounts hit by margin calls, a president's disapproval rating cracking 50% for the first time, and a market still whipsawing violently as of this week. This is the full version: the real timeline, the mechanics of how a brand-new financial product turned a normal correction into a crisis, and a check on the headline number itself. Every outlet covering this cites some version of a large SK Hynix drawdown (30%, 35%, or worse). None of them are fabricating anything; the underlying reporting on the leverage and the margin calls is solid. But a headline reaches for the biggest number you can still justify, not necessarily the most accurate one. Once I checked SK Hynix's own Nasdaq-listed stock against an honest starting point, rather than the one that produces the biggest number, the real figure turned out to be smaller, and more interesting, than what has been reported. It is also not the only place this story does not quite line up with how it has been told: almost every account of this crash, including my own first draft, starts the clock on 14 July. The index itself says otherwise.

Timeline

19 June — the KOSPI closes at an all-time high of 9,385.59.

23 June — four days later, the index falls 10% in a single session. SK Hynix and Samsung both drop more than 12%, and Korea's exchange halts trading for 20 minutes. This is a real, single-day crash, and it barely shows up in any retrospective on this story, including the earlier version of this piece.

2 July — Meta announces plans to monetize its excess AI computing capacity, and the market reads that as a sign the AI buildout might need fewer chips than everyone assumed. The KOSPI falls below 8,000, more than 14% under its June peak, over a week before SK Hynix's ADR even starts trading. Whatever people mean by "the crash," the index had already been falling for three weeks by 14 July.

10 July — SK Hynix completes its Nasdaq ADR debut, raising $26.5bn, the largest US listing ever by a foreign company (bigger than Alibaba's $25bn in 2014). 177.9m ADRs price at $149; the stock closes its first session at $168.01.

13-14 July — Seoul-listed SK Hynix common shares fall 15.37% on the 14th, the KOSPI drops over 9% intraday and trips trading halts. This is the point everyone, including my own earlier write-up, treated as "the crash." It was not. It was the opening act. (More on what actually happened to the ADR that same day below, since it is the most interesting single data point in this story.)

Through mid-July — the real mechanism driving this turns out to be a specific, dateable regulatory choice, not an inevitable market outcome. Korea had previously prohibited single-stock leveraged ETFs. Funds had to hold at least 10 stocks, with any single name capped at 30%. On 28 April 2026, regulators amended that rule, raising the single-stock concentration limit to 100%. The first 2x-daily single-stock products tied to Samsung and SK Hynix launched a month later, on 27 May. Korean retail investors bought a net ₩14tn of them versus roughly ₩2tn from foreign investors. Outstanding leveraged bets peaked at ₩29.2tn (~$19.7bn) in early July. By 13 July, more than 1.2 million retail leveraged accounts had triggered margin calls; 320,000-360,000 were forcibly liquidated, totaling ₩2.3tn in forced selling in about two and a half months.

27-28 July — two things land almost simultaneously. First, reports that a Chinese state-backed firm (Shanghai Aishengna) had begun mass-producing homegrown immersion DUV lithography tools (domestically-made chipmaking machines that reduce China's dependence on Dutch supplier ASML). Second, SK Hynix reports Q2 results: revenue of ₩79.32tn, up 257% year-on-year, with a 76% operating margin. It was a genuinely record quarter that still missed consensus (~₩84tn revenue expected), largely because some HBM4 shipment recognition slipped into Q3. Bank of Korea also raised rates into this same window.

July, in full — the KOSPI posts its worst month on record, though "worst" depends on how you measure it, and most coverage does not say which way it is measuring. Using the index's own closing prices: from the 9,385.59 peak on 19 June to the closing low of 5,593.56 on 30 July, that is a 40.4% decline, a number I calculated directly from the daily closes rather than taking anyone's word for it. Some outlets report smaller figures, closer to 22-23%, which lines up with a standard calendar-month comparison (July's own start-of-month level to its end-of-month close) rather than peak-to-trough, and that number looks smaller mainly because the historic 31 July rebound landed inside the same calendar month and pulled the month-end close back up. Both numbers are real; they are just answering different questions. The "worse than 1997 and 2008" claims usually use the bigger, peak-to-trough figure, and compare it to 1997's and 2008's calendar-month declines (27% and 23%), which is a comparison worth knowing the terms of before repeating it. Seven circuit breakers triggered across the month; one two-day stretch alone wiped out ₩864.5tn, enough to force an emergency government meeting on the evening of 29 July.

29 July — the trough. SK Hynix's ADR bottoms at $126.79.

31 July — the single most violent reversal in this entire story, and one worth knowing on its own. The KOSPI closed up 17.91%, the largest single-day percentage gain and point gain in the index's 70-year history, beating the prior record of 11.95% set during the 2008 financial crisis. Samsung Electronics surged 28%; SK Hynix's Seoul-listed shares soared 30%. It came three trading days after the index had fallen 17.2% in that same short window. A market that can move nearly 18% in either direction inside a single week is not a market that has repriced calmly to a new equilibrium. It is a market still being pushed around by exactly the same forced-flow mechanics (this time unwinding short positions and margin-call-driven selling pressure, in reverse) that caused the crash in the first place.

Late July into August — the government responds: the Financial Services Commission suspends new single-stock leveraged ETF listings, bans their marketing, and triples the minimum deposit for new leveraged investors to ₩30m (~$20,000). It is also weighing a cap limiting any investor's leveraged exposure to 20% of their total portfolio. None of it stops President Lee Jae-myung's disapproval rating from breaking 50% for the first time in his term. Korea's own National Pension Service, drawn into the stabilization effort, has been publicly criticized as "an amplifier, not a stabilizer."

This week — the market is still nowhere near settled, and the 6 August selloff has a specific, telling trigger of its own. The KOSPI rose 3.76% to 6,598 on 5 August on hopes of a US-Iran deal, then fell 4.58% to 6,296.38 on 6 August as Samsung closed down 6.3% and SK Hynix fell nearly 9%. The actual proximate cause: US storage chipmakers SanDisk and Western Digital both beat Q4 revenue and earnings estimates the same week, but gave forward guidance that fell short of the sky-high bar their own stock prices had set (SanDisk was up 469% year-to-date heading into the print, Western Digital up 202%). Western Digital fell 14-16%, SanDisk 8-11%, and the disappointment dragged SK Hynix and Micron down with it in sympathy. That is the third time in this story the exact same pattern shows up: a company beats real, substantial estimates, and gets sold off anyway because the guidance did not clear a bar that had already priced in perfection. SK Hynix's own 28 July print was the first instance; this is the second. This is not a market that has found a floor and stopped moving. It is still trading like a market with real fear in it, and the fear is specifically about valuation and guidance, not about underlying results.

The full timeline, the real daily price data behind the chart below (with live formulas, not just static numbers), the leverage/margin-call figures, and the bull/bear case side by side are all in the data workbook (.xlsx), if you want to check my math or build on it yourself.

Two listings

This is worth surfacing on its own, because it is the reason the next section's number is different from the one you have been reading elsewhere. On 14 July, the day every headline cites as "the crash," SK Hynix's Nasdaq ADR did not fall. It closed at $193.92, up from $152.35 the day before, a 27% single-day gain, on the same day the Seoul-listed common shares fell 15.37%.

Two listings of the same company, same day, moving in opposite directions by double-digit percentages. The mechanical reason is straightforward once you see it. Korea's leveraged ETFs are Korean-won products, listed in Seoul, tied to the Seoul-listed shares. They have no direct connection to a Nasdaq-listed ADR that had been trading for all of four days. The new, dollar-denominated ADR was still absorbing genuine first-week demand (helped by the earlier reported 7x oversubscription) even as Seoul-side holders were already getting hit by the leveraged-ETF unwind. The two listings did not actually converge into a shared, coherent price until the China/earnings shock a full two weeks later dragged both down together.

That split matters for more than trivia. 14 July is the date most coverage, including my own earlier draft, treats as SK Hynix's pre-crash peak. It was not a normal price level for the ADR at all. It was a temporary, IPO-driven spike sitting on top of the stock's actual trend, and measuring "the crash" from that spike inflates the number.

What the price data actually shows

Every article on this crash cites some version of a large SK Hynix drawdown. That is a reasonable thing to report, since the leverage, the margin calls, and the forced liquidations behind it are all real. But "SK Hynix's ADR crashed 35%" is the kind of number you get when you measure from that inflated 14 July print, the biggest one defensibly available. It is not the most accurate description of what happened to the stock.

Index SK Hynix's ADR and Micron (a comparable, US-listed, unlevered memory-chip peer riding the same AI/HBM demand cycle, with zero exposure to Korea's leveraged-ETF market) to 100 on 10 July, SKHY's actual first trading day. There is no earlier price history for this specific security, so the window cannot start any earlier than that. Run both through the 29 July trough and SK Hynix's ADR fell 24.53%. Micron fell 24.54%. Statistically the same stock.

The "35% vs 25%" gap that has been cited everywhere, including in my own first draft of this piece, only appears if you start counting from 14 July. Strip out that one temporary IPO-week pop and there is no incremental Korea-specific damage left in the ADR's own price at all. It moved with the same global correction that hit an unlevered peer, down to the decimal point.

What is real, and actually Korea-specific, is everything that happened on the other listing: the 1.2 million margin calls, the 320,000+ forced liquidations, the political fallout. All of it ran through the Seoul-listed shares and Korea's leveraged-ETF plumbing, which is mechanically separate from the Nasdaq ADR. The leverage crisis is real. It just is not what shows up in the ADR's own price chart once you measure it from an honest starting point.

The China shock

The Shanghai Aishengna DUV news is real and worth taking seriously as a multi-year signal, but it is currently much smaller than the market's one-day reaction implied. The Chinese firm is expected to deliver roughly five immersion DUV units in 2026, rising to about 20 more in 2027. ASML alone expects to ship around 130 immersion systems in 2026. The domestic Chinese tools reportedly still lag ASML's technology and depend on some non-domestic components. This is a real, credible first step toward Chinese self-sufficiency in a category it has never mass-produced before, and a genuine long-term reason for Korean chipmakers to keep investing in staying ahead. But it changes essentially nothing about 2026 or 2027 memory-chip supply and demand. The market's violent reaction to it looks much more like an already-fragile, over-levered market treating a symbolic headline as a reason to sell than a rational repricing of near-term competitive risk.

Comparing to historical cases

It is tempting to reach for Korea's last great financial crisis, and some of the July numbers genuinely did exceed single-month declines from that era. But the mechanism is close to the opposite. 1997 was an external funding crisis. Foreign banks stopped rolling over loans to Korean banks and companies, who found they could not secure new financing at any price. 2026 is a domestic leverage crisis, manufactured almost entirely at home: a financial product regulators approved in late May, bought overwhelmingly by Korean retail investors rather than foreign capital, unwinding through margin calls rather than a sovereign funding freeze. Foreign investors did pull real money out in July (widely reported around $13bn), but as a consequence of the crash, not its original cause. It is a different disease, and I think a much shorter one, because a leverage-driven crash burns out once the forced sellers are gone, and the 1.2 million margin calls and 320,000+ liquidations by mid-July did most of that flushing already. A funding crisis does not resolve nearly that fast.

The 2018 memory-chip downturn is the other obvious comparison, and it is also a poor match for what just happened, but for a reason worth noting, because I think it points at the actual forward risk. 2018 was a supply-side glut: hyperscalers had over-ordered during the first big cloud buildout, and DRAM prices then fell roughly 60% over the following four quarters as that excess capacity actually arrived. Nothing about July 2026 resembles that. Real demand data from this same window all point at demand that is still real and still accelerating, not a 2018-style glut arriving into a saturated market. Amazon, Microsoft, Alphabet and Meta are guiding to a combined $725bn of 2026 AI capex, up 77% year-on-year, with Evercore and Bank of America now modeling over $1tn combined for 2027. S&P Global's July PMI showed the fastest tech-equipment output growth since May 2021.

But new fab capacity from Samsung, SK Hynix, Micron and Kioxia is not due to reach volume production until late 2027 or 2028. Every one of them is expanding capacity right now, at the same time, into a demand curve that even the bulls describe as a "supercycle," a word that basically admits growth this fast is not permanent, even while everyone is still riding it. That is not this crash. It is a real, specific, later risk. If 2027-2028 capacity additions land just as hyperscaler capex growth naturally decelerates off this year's exceptional base, you get something that could look a lot more like 2018 than anything that happened in July 2026.

Where the valuation sits now

Everything above is about what already happened. Here is where the stock actually sits today, because that is the part most retrospectives skip.

SK Hynix's Seoul-listed shares closed at ₩1,718,000 on 6 August. At that price, the stock trades around 5.3-5.9x forward earnings, well below its own 5-year average (6.5x mean, 11.2x median, 2021-2025) and a long way down from the 17.8x multiple it commanded at its own peak in March 2026, right before the crash. The 37-analyst consensus tracked by Investing.com puts a 12-month price target of ₩3,189,340 on the stock, about 86% above where it closed on 6 August.

An 86% gap between price and consensus target, on a stock this widely covered, is a real number worth sitting with. It is not automatically a "the stock is cheap" signal though, and this is exactly where the cyclicality point from the section above actually matters. Memory chips are the textbook case of a business where a low forward P/E can mean the opposite of cheap: the market already expects this year's extraordinary earnings not to repeat, so a multiple that looks low against this year's earnings can look completely normal, or even expensive, against a normalized, mid-cycle number. A 5x multiple on a peak-cycle year and a 5x multiple on a sustainable year are not the same stock, even when the P/E prints the same.

I am not building a full discounted cash flow model on top of this, on purpose. A DCF for a business this cyclical needs multi-year revenue and margin assumptions that would mostly be guesses dressed up as precision, especially heading into the 2027-2028 capacity question raised above. What the real numbers here do support is narrower and more useful: the multiple has compressed hard, the consensus gap is real and large, and whether that gap is the market being wrong or the market correctly discounting for reversion depends entirely on which side of the 2027-2028 capacity question you land on.

The bull case

SK Hynix just posted 257% revenue growth and a 76% operating margin, and "missed" only because consensus had run even further ahead of an already-extraordinary number. That is a sign of a market pricing in near-perfection, not a business in trouble. J.P. Morgan's own read, published the same week as the fresh 6 August selloff, was that the sell-off "had not derailed" the AI investment cycle, and "we do not see any fundamental indicators that signal meaningful weakness in the next 6-12 months." Real hyperscaler capex guidance backs that up directly. On this view, July was a leverage-and-positioning accident laid on top of a genuinely intact fundamental story, and the current multiple compression is a buying opportunity for anyone willing to sit through the volatility.

The bear case

Michael Burry's own framing, from a 4 August note, a day before the market's next leg down: "I continue to believe it is possible we are near a major top, and possibly a 1987-type fall." The bear case is not really about SK Hynix's own numbers. It is about crowding and leverage across the entire AI trade, of which Korea's retail leveraged-ETF mania was one particularly visible, particularly Korean symptom, not a uniquely Korean disease. If the same positioning fragility exists in less visible forms elsewhere in the AI trade (margin debt, options-driven flows, concentrated index weights), then a Korea-style unwind could recur anywhere sentiment cracks next, and the 2027-2028 capacity-glut risk above sits waiting underneath all of it regardless of how the next twelve months of sentiment go.

My view

I do not think demand actually broke here, and I do not really think this was a "SK Hynix crashed" story either. It was two stories wearing one company's name. The ADR's own price move was, once you check it against an honest starting point, statistically identical to Micron's, a stock with zero Korean leverage anywhere near it. Whatever Korea's leverage crisis actually broke, it did not show up in that number. What it did break is real. Korea built a genuinely fragile piece of market plumbing (a brand-new, unseasoned 2x retail product) directly underneath the most crowded trade in the world, on the Seoul-listed side specifically, and that combination was always going to break violently the first time sentiment wobbled even slightly, which a slightly-worse-than-priced-for-perfection quarter and a symbolic Chinese headline were more than enough to trigger. That leverage-driven part (the margin calls, the forced liquidations) has mostly already worked through the system by mid-July, which is why I would expect the extreme, product-specific volatility to fade faster than a real funding crisis (1997) or a real supply glut (2018) would.

What I do not think fades: the multiple. Even once the forced selling is fully done, I would expect SK Hynix and Samsung to trade at a structurally lower multiple on their HBM earnings than they did in June, simply because the market has now watched exactly how fragile Korean retail positioning can be, and will keep some of that risk priced in going forward. The risk I would actually be watching two years out is not anything from this July. It is whether 2027-2028 capacity arrives right as hyperscaler capex growth normalizes off this year's number, because that is the setup that has ended every memory cycle before this one.

Sources

• SK Hynix Nasdaq debut, $26.5bn ADR listing — Bloomberg
• SK Hynix Nasdaq debut sets record — Yahoo Finance
• Capital exodus hits KOSPI after SK Hynix's Nasdaq debut — Korea JoongAng Daily
• Give me my money back: Korean traders' leveraged bets unravel — CNBC
• Minister apologizes as leveraged ETF investors nurse losses — CNBC
• 1.2 million accounts hit margin calls — BigGo Finance
• South Korea's KOSPI plummets 23% in July, record circuit breakers — KuCoin
• KOSPI crash officially worse than 1997 and 2008 — Yahoo Finance
• China begins mass-producing homegrown DUV lithography tools — Tom's Hardware
• China's chip breakthrough comes with big caveats — CNBC
• SK hynix 2Q26 results — SK hynix Newsroom
• SK Hynix Q2 revenue jumps 257% YoY — BigGo Finance
• Chip stocks shed $1tn in selloff — CNBC
• Asian tech stocks drop, SK Hynix plunges 10% — CNBC
• Korea holds emergency meeting as ₩864tn leaves market — Yahoo Finance
• Kospi rout puts Lee government on alert — Bloombit
• NPS becomes an amplifier, not a stabilizer — Korea JoongAng Daily
• KOSPI falls to 6,200 level as chip giants plunge — Seoul Economic Daily
• Hyperscalers plan $725bn 2026 AI capex — AI Weekly
• Michael Burry bets against rally, warns of 1987-type fall — CNBC
• J.P. Morgan: Asia tech sell-off has not derailed AI investment cycle — CNBC
• South Korea's market crash cost retail investors $39bn, incl. the 28 April rule change — ProfG Media
• 120,000 Korean retail investors face margin calls — KuCoin
• KOSPI surged 17.9% in a single day, largest one-day gain in history — TradingKey
• KOSPI hit all-time high of 9,385.59 on 19 June, down 27% by 13 July — IndMoney
• US storage stocks fall as SanDisk, Western Digital guidance disappoints — TradingKey
• Kospi falls 4.6% as Samsung, SK hynix lead tech selloff on AI spending fears — Korea JoongAng Daily
• Kospi Index Slides 4.6% With Samsung, SK Hynix Falling on Chip Concerns, 23 June — Bloomberg
• Meta's AI pivot triggers global chip sell-off, sending Kospi below 8,000, 2 July — KED Global
• KOSPI daily historical closing prices — Investing.com
• SK Hynix consensus estimates, 12-month price target — Investing.com
• SK Hynix forward P/E ratio — GuruFocus
• SK Hynix 5-year P/E history (mean, median, peak) — Wisesheets