SK Hynix closed up nearly 30% on Friday, July 31 — its best single day on record. Samsung closed up nearly 27%, also its highest single-day gain ever. The KOSPI reportedly ripped 15.25% to 6,446.63, per an unverified reply post on X whose reliability is unknown.
None of that happened in Seoul. It happened in Redmond and Seattle, the night before.
Amazon jumped more than 9% after hours on a cloud beat. Microsoft had already popped 16% Thursday on faster-than-expected Azure growth, per CNBC (though another CNBC report cited 15%). By the time Korean markets opened, the trade was already priced — Seoul just executed it.
Last week, this letter flagged how selective chip sentiment had become — how SK Hynix's own results could drag the sector down even as the AI story held. This week the same stock swung the other way, a 25-to-30% rip within days of the selloff, per CNBC. Same names, opposite sign, and almost none of it driven by anything the companies themselves reported.
That's the story. Not the rally — the mechanism behind it.
Go back eighteen days. On July 13, SK Hynix plunged 15% and the KOSPI shed more than 8% in a single session in what E8 describes as a chip-driven crash. The Nasdaq 100 followed with a 1.88% drop. Micron fell 4%, SanDisk 12%, Intel 6%.
Now the mirror image: US hyperscaler earnings beat, and the KOSPI moonshots 15%. Both moves originated in the same place, and it wasn't Korea.
Here's what welds them together. Samsung and SK Hynix together are more than half the KOSPI. When they move, the national index moves — there's no offsetting weight to absorb it.
And what moves them is no longer domestic. Data-center demand made up roughly 40% of global DRAM demand last year. That figure now exceeds half, and it's still climbing.
So follow the chain. Half the KOSPI is two memory makers. More than half of what those two makers sell now goes into data centers. And those data centers are being built by four American hyperscalers whose capex budgets are set in US boardrooms.
The result is a number that should worry anyone holding Korean equities as a diversifier: the 60-day correlation between the KOSPI and the Nasdaq 100 has climbed to about 0.50, per Rayliant Global Advisors — the highest since 2021.
A national stock index is now trading as a leveraged proxy for four American capex lines. That's not diversification. That's the same bet, with a currency conversion and a time-zone lag.
A national stock index is now trading as a leveraged proxy for four American capex lines. That's not diversification. That's the same bet, with a currency conversion and a time-zone lag.
The lag is the tell. Samsung and SK Hynix trade hours before Wall Street opens, which turns them into an early read on how US investors will react to AI news. Korea has become the pre-market for the American AI trade.
Now here's the part everyone is cheering that they shouldn't be.
The bullish catalyst on Thursday was Samsung's own warning that the memory crunch could last into 2028. Micron rallied 18% on it, SanDisk 26%. Lam Research popped 18%, its best day since 1999, on AI-driven equipment demand.
A shortage running to 2028 sounds like pricing power, and for the sellers it is. UBS initiated SK Hynix's ADR at Buy with a $204 target, arguing DRAM supply may not keep pace with demand until 2028.
The UBS math is genuinely striking. Analyst Nicolas Godoy figures the current price reflects a long-term ROE of about 17.7% — less than half the 40.2% average he projects for 2027 through 2031. He forecasts 36% DRAM demand growth next year, up from 22% this year.
But look at what makes that number durable. UBS praised SK Hynix's roughly ten long-term supply contracts — including deals with large US cloud providers — with HBM negotiations already underway for 2027 and beyond.
That's the trap dressed as a feature. Those contracts lock in stable revenue, yes. They also lock SK Hynix's revenue base directly to the same US hyperscalers whose capex sets the whole cycle.
A multi-year shortage doesn't loosen the KOSPI-Nasdaq correlation. It welds it shut. The longer the shortage runs and the more long-term supply gets contracted to American cloud buyers, the more completely Korea's index becomes a derivative of US infrastructure spending — with a fixed maturity of roughly 2028.
The asymmetry a reply under Bloomberg's post nailed it best: "A shortage is a rally for the two firms selling it, a bill for everyone buying. Same headline, opposite sign depending on your side."
That's the whole thing in one line. For SK Hynix and Samsung, the shortage is margin. For every hyperscaler buying DRAM to fill a data center, it's a rising cost line. And for anyone holding the KOSPI, it's exposure to both sides of a single trade with no hedge in the index.
Because here's what concentration actually removes: the shock absorber.
In a diversified index, if one cyclical theme rolls over, other sectors cushion the fall. In a KOSPI where half the weight is two memory makers levered to four capex budgets, there is nothing to cushion anything. When the theme moves, the whole index moves — 8% down on July 13, 15% up on July 31.
Some analysts have already warned that a slowdown in hyperscaler capex would hit Korea harder than most markets, precisely because half the KOSPI now rests on one cyclical theme. Korean memory stocks also carry more volatility than US peers, and leveraged ETF flows amplify the swings on both ends.
Notice what the market rewarded Thursday and Friday. Ortus Advisors' Andrew Jackson pointed out that Azure beat while Microsoft kept capital spending "in check," and that a "'spend at all costs' mentality has been punished by the market."
Sit with that. Arguably, the very thing that would end the shortage — hyperscalers pulling back on capex — is what the market appears to reward, given that per CNBC a "spend at all costs" mentality has been punished. The good news for cloud discipline is the bad news for Korea's index. Those are the same event.
The demand signal is real for now. Google Cloud grew 63%, Azure about 40%, AWS 28% on a $364 billion backlog, and the top four hyperscalers' capex is projected to grow 81% in 2026, per ainvest. As long as that holds, the shortage holds and the KOSPI rips.
But the mechanism cuts both ways, and it's the same mechanism. There's no version of this where Korea decouples on the upside and insulates on the downside. The 0.50 correlation isn't a mood — it's structural, built out of index weights, demand mix, and supply contracts.
Watch Samsung's guidance, which typically lands about two weeks ahead of major US semiconductor results. It's the next read on how tightly these two markets are trading. And watch CXMT — China's Changxin, which surged 466% on its Shanghai debut to become China's most valuable listed company. New supply is exactly what breaks a shortage before 2028.
The 30% day was real. So was the 15% day two weeks earlier, in the other direction.
Both came from the same place. Until 2028 — or until hyperscaler capex blinks — Korea doesn't own its own market. It's arguably renting Nasdaq's, at a leverage ratio nobody underwrote on purpose.