
South Korea, a G-20 economy, has had its stock market index halted 7 times this year, sailing past records set in 2008 and 1997 during the Asian market crash. For stretches of this year, it has been more volatile than Bitcoin. On the 22nd of June the index closed at an all-time high of 9,114; the very next session it fell nearly 10%. President Lee, a former trader, ran his campaign on a number: drag the KOSPI to 5,000, about double where it then sat. Pledging a stock-index target is the fiscal cousin of promising better weather, and the pledge was greeted as such. Then the market obliged, and kept obliging, and kept obliging, cresting near 9,000 in June, nearly double the promise, before the floor gave out.
The companies at the center of the frenzy are among the most profitable industrial firms ever. Samsung Electronics and SK Hynix make the high-bandwidth memory that every AI server on the planet is starving for, and their earnings are no rumor: Samsung’s operating profit multiplied several times over year-on-year, and SK Hynix’s roughly quintupled. They are the arms dealers of the AI war, and their order books are very real, which makes what happened next far more interesting than a garden-variety bubble.
The Machine That Sells at the Close
Out of a population of about 51 million, South Korea has something like 14 million retail investors (nearly a third of the entire country holds a brokerage account) and on a typical day they are roughly half of all trading volume, against something nearer a fifth in the United States. Foreign funds run about a third of the volume, and Korea’s own pension funds and asset managers, the people who invest for a living, a mere eighteen percent. So when foreign money began sprinting for the exits this year, yanking some $95 billion out over six months, it was this retail army -- the “ants,” in the local slang -- who stood underneath with their arms open and soaked up roughly $80 billion on the way down. For two years, being the permanent bid beneath the best-performing market on earth was a glorious trade, and the ants had the reason right: memory chips really had become the chokepoint of the decade. But greed turns a winning hand into a broke player, and as Munger once said, “there are three ways to go broke: liquor, ladies, and leverage.”
In late spring, Korean brokerages loosed a swarm of single-stock leveraged ETFs; funds that vow twice the daily move of Samsung or SK Hynix, shares that already swing 8 or 10% in an ordinary session. To keep that two-to-one ratio, the fund must rebalance every afternoon: buy more after the stock rises, sell more after it falls, mechanically, at the close, with no opinion about whether that is wise. The computer just fires. On a two-stock index during a chip selloff it becomes a doom loop from hell: a fund forced to dump shares into a market already sinking, which sinks it further, which forces the next sale. On rough days, per Bloomberg, this rebalancing swelled to as much as two-thirds of all trading in SK Hynix, a company worth north of a trillion dollars; one July session forced $5 billion of selling at the bell purely to rebalance. No one chose to sell it. The afternoon simply arrived, and the machine did what it does.
The KOSPI, in other words, had become a kind of K-pop act: two idols carrying the entire show, a devoted and heavily borrowed fandom, a performance staged for an audience overseas. And leverage did what leverage does: compress time for better or for worse. A leveraged fund decays because it resets daily, a choppy round trip that leaves the underlying stock flat still leaves the fund lower, and a sharp drop is punished twice. When SK Hynix fell about 18%, its two-times ETF fell closer to 48. More than 1.2 million leveraged retail accounts got margin calls, roughly 1 in every 30 working-age adults in the country, and several hundred thousand were liquidated outright, much of it at the close, some so abruptly that the sale failed to cover the loan and the holder awoke owing the broker money.

The following infamous experiment is an easy way to visualize what leverage fatigue really is. The investor Victor Haghani gave finance-literate volunteers a coin they were told, in writing, was rigged to land heads 60% of the time, and let them bet it freely. A biased coin you can wager repeatedly is close to free money -- yet most of them lost, and many got cleaned out inside half an hour, because they bet too large and an ordinary streak of tails arrived before the edge could compound. Korea is that experiment at national scale: a weighted coin, the AI boom, and a country that staked too much on each flip. Leverage never actually improves the underlying asset you own; it only sets a timer on how long you are allowed to be right before an ordinary dip clears you out.
Seoul Pays, Silicon Valley Collects
Korea runs a record current-account economic surplus, selling the physical tools the whole AI gold rush runs on, so by the textbook its currency ought to be among the strongest on earth. Instead the won is among Asia’s weakest, lately brushing levels last seen amid the rubble of 1998.
The economist Brad Setser has a name for where the money went: DRAMdollars, a nod to the petrodollar era, when oil states recycled crude revenues into American assets instead of bringing them home. Samsung and SK Hynix now do the same with memory, keeping much of their dollar haul offshore to build data centers and expand abroad. And Korea’s retail army, when it is not detonating leveraged funds at home, has become one of the world’s most eager buyers of American tech stocks, including, with a novelist’s sense of irony, Nvidia, the very customer its own factories supply. Every US share bought in Seoul means won sold for dollars first, so the country exports its own savings and imports downward pressure on its exchange rate, which is part of why the central bank was cornered into raising rates into a falling market last week.
On top of all this pressure, Seoul refuses to allow a freely convertible and offshore-traded won because the scar of the 1997 IMF bailout still runs policy. So MSCI keeps filing the literal best-performing major market on the planet under “emerging,” on the technicality that a London fund cannot trade the won around the clock. A country can win the industrial layer of a revolution and still be treated as a junior partner in its own boom.
But the numbers truly holding the AI market high in the sky are nowhere near Seoul. The entire Korean market is basically leveraged wager on the capital-expenditure budgets of about four American companies -- Amazon, Google, Meta, Microsoft -- whose combined AI spending is on track for well over half a trillion dollars this year. I say the entire Korean market because SK and Samsung make up about 52% of the entire index. And here we were in America frightened over NVIDIA making up 7% of the S&P. This month the market began to truly doubt those spending forecasts. Taiwan’s TSMC, as near the engine room of AI as any firm gets, posted record profit and lifted its outlook; yet enthusiasm cooled, because it also admitted that staying ahead would cost far more in capex and press on its margins. Investors stared at a company minting money and, for the first time in three years, fretted about the price of staying in the race. Every single blue-chip has had pretty unreal earnings this year, but as soon as the market scrolls upon those CapEx figures, they jump.

Buried in that flinch is the question you’ve heard all too well by now; what does all this spending actually return? Should AI congeal into something closer to electricity -- models largely interchangeable, priced toward zero by competition -- the value will be immense, and most of it will pool around the people using the tools while the firms that spent the fortune capture a sliver. In that world the hyperscalers push a data center or two back a quarter, and issue no press release about it. The first tremor of that decision would be felt in Suwon and Icheon long before it reached Silicon Valley, in an index two stocks wide, lashed to billions in automated leverage that runs precisely as hard in reverse. Every step up the mountain was a decision millions of people made for themselves. A fall, if it comes, is never put to a vote.
See ya, folks. And stay curious!
*Credit to the fantastic breakdowns on the Patrick Boyle YouTube Channel for giving me inspiration for this Substack.*
—J&E