Korea’s AI Chip Trade Broke Its Own Market: Inside the KOSPI’s Record Crash

The rise and fall of South Korea's AI memory supercycle.

The index that beat every market on earth for five months just gave back the better part of that gain in a matter of weeks, and the reason has less to do with sentiment than with a leverage structure nobody stress-tested for a memory-chip supercycle unwinding into a rate hike and an oil shock at the same time.

By Dorian

June was a record-setting month almost by the day.

The KOSPI hit a fresh intraday high of 8,933.62 on June 2, closed above 9,000 for the first time on June 18 at 9,063.84, and by most accounts touched its cycle peak near 9,385 on June 19. Sources disagree on the exact top — one data desk puts it closer to 9,114 — but the range is not in question: this was the highest the index had ever traded, up somewhere between 80 and 120 percent since January depending on which starting print you use.

On June 22, by most reporting, SK Hynix passed Samsung Electronics in market capitalization for the first time in Korean market history. Individual investors kept buying through it; foreign investors had already started selling into the strength days earlier.

[TABLE — Korea Cross-Asset Dashboard, July 29 2026]

I. The Misread

The consensus read on the reversal that started June 23 was profit-taking after a euphoric run.

That read survived about three weeks.

It could not survive July 28, when SK Hynix reported record second-quarter revenue and operating profit and the stock still fell nearly 5 percent, dragging the index down 10.84 percent to 6,023.66 in a single session. Record earnings, and the stock got sold anyway. That is not a valuation reset. That is a market discovering that its own leverage was the actual position, and the AI-demand thesis was just the collateral backing it.

II. The Mechanism

South Korea’s retail base trades on margin more aggressively than almost any developed market, and 2026 added a second layer:

leveraged ETFs tracking the semiconductor complex, sized for a market that had not had a real drawdown since the AI trade began.

When SK Hynix and Samsung — which together dominate KOSPI weighting — started printing single-day declines in the high single digits, those products did not just track the fall, they forced it.

Two circuit breakers and three sidecar halts hit inside the single week of June 22-23 alone, three of only eleven circuit-breaker events in the KOSPI’s entire trading history landing inside that one month. Seven had been triggered in total by July 13, a figure two independent trackers agree on even where they disagree on the closing print that day — one has the index at 6,806.93 that session, another at 6,880.98, roughly 70 points apart. Selling begat margin calls, margin calls begat forced selling, and a market with 954 constituents behaved, for a week, like two stocks and their derivatives.

[CHART 1 — KOSPI Composite, January–July 2026: Peak to Circuit Breaker]

Layer onto that a Bank of Korea that had spent three and a half years on hold.

On July 16, the BOK raised its base rate 25 basis points to 2.75 percent, the first hike since January 2023, after June inflation came in at 3.2 percent against a 2 percent target.

The move was well-flagged in advance. What wasn’t fully priced was the timing: a hawkish central bank meeting landing three days before the market’s worst single session of the year, pulling liquidity out of a system that was already deleveraging on its own.

III. The Pressure Point

The won is where the mechanism shows up cleanest. It touched a seventeen-year low near 1,560 against the dollar on June 5, weakened by the same capital that had piled into Korean equities on margin now needing dollars to cover.

By July, two forces reversed that slide almost as fast as it happened.

SK Hynix’s U.S. ADR listing — reportedly raising on the order of $26.5 billion, a figure from a single data desk and not yet confirmed elsewhere — triggered a wave of dollar conversion, alongside a broader push from Seoul for exporters to repatriate earnings.

That pulled the won back to the 1,459–1,485 range by late month, its strongest level since early May.

Second-quarter GDP reportedly grew 0.6 percent against a Bank of Korea forecast of 0.2 percent, driven by semiconductor exports that kept printing even as the equity market trading on those exports fell apart.

[CHART 2 — USD/KRW, June–July 2026: 17-Year Low to Recovery]

This is the split that matters going into August. The real economy — exports, GDP, the current account — is not confirming the crash. The equity market’s own plumbing is.

IV. Market Expression

The rate story compounds it. Korea’s policy rate now sits roughly 100 basis points below the Fed funds rate, the narrowest gap in three and a half years — constructive for the won on a carry basis, but it also removes the cheap-money backdrop that helped inflate KOSPI multiples in the first half. Meanwhile Brent crude spiked above $100 around July 22 on U.S.-Iran tensions, then eased into the July 28-29 session — oil is now doing double duty as an inflation input for a central bank already hiking, and a risk-sentiment input for a chip sector that trades like a leveraged bet on global growth.

[CHART 3 — Bank of Korea Policy Rate vs. Fed Funds, 2023–2026: The Narrowing Gap]

Realized volatility on the KOSPI has topped 60 percent this year — by one measure close to double Japan’s Nikkei and above Bitcoin’s. Separately, the exchange’s own first-half data shows average intraday volatility at 3.30 percent, the second-highest half-year reading on record, behind only the first half of 1998.

Two different volatility metrics, same conclusion: this is a market that traded like a momentum asset on the way up and is trading like one on the way down.

V. Invalidation

The bull case is not dead.

A Goldman target of 12,000 has been circulating in Seoul brokerage notes — reported secondhand, not traced to a primary source, so treat it as sentiment rather than fact — but if taken at face value it implies close to 100 percent upside from the July 28 low.

That call only works if HBM demand growth decelerates rather than reverses, and if the deleveraging in leveraged products completes without a second forced-selling wave.

Watch three things:

whether SK Hynix and Samsung stabilize on their next prints without another double-digit single-day drop,

whether the won holds the 1,450-1,490 band without a fresh break toward 1,550, and

whether the BOK signals a pause after this hike rather than a cycle. A close back above 7,200 on volume,

without a fresh circuit breaker on the way there, says the deleveraging is over and the earnings story is back in control. Failure to hold 6,000 on a closing basis says the unwind isn’t finished.

VI. Tactical View

The trade here is not “Korea is broken” or “Korea is cheap.” It is that the equity market and the underlying export economy have temporarily decoupled, and that decoupling is a leverage story, not a demand story.

Positioning for a reconnection — long the won, selective in memory names post-capitulation, cautious on leveraged ETF exposure until volume normalizes — fits the mechanism better than a binary call on the index level.