The Financial Times published a column this week asking a question that has moved from academic journals to the front pages of business sections: what happens when the AI bull market begins to consume itself? Its chosen case study was South Korea, the market that has ridden the AI boom harder than any other on earth, and whose structure makes it a laboratory for the boom’s excesses.
The numbers from Seoul are staggering by any standard. The Kospi, the country’s main index, has more than tripled since early last year, making South Korea the world’s best-performing major stock market, with a 76 percent gain in 2025 alone. The country’s market capitalization ranking climbed to fifth in the world, overtaking Britain and France. The engine of the rally is a pair of companies: Samsung Electronics and SK Hynix, the memory chip makers whose products — particularly the high-bandwidth memory that powers Nvidia’s AI accelerators — have become the most sought-after commodity in technology.
By the middle of this year, Samsung and SK Hynix together accounted for more than half the value of the entire South Korean stock market, according to exchange data. The concentration is without precedent among large markets, and it has turned the Kospi into a bet concentrated on a single industry, one product category, and ultimately one customer set: the AI data center builders of the United States.
The retail participation makes the situation more fragile. South Korea has roughly 14 million individual investors, who account for 60 to 70 percent of daily trading volume, a retail share that dwarfs every other major market. The country’s so-called ant investors, who shifted savings from deposits, real estate, and crypto into semiconductor stocks over the past two years, have become the marginal price-setter in the market. When they buy, the index rises; when they sell, it falls, and their behavior is driven less by fundamentals than by momentum.
The newest instruments have intensified the dynamic. The Korea Exchange launched single-stock daily-double products tied to Samsung and SK Hynix in late May, and retail investors poured roughly 14 trillion won, about $10 billion, into them within weeks. The products amplify daily moves by design — a 5 percent gain in the underlying stock becomes a 10 percent gain in the product — and their popularity has introduced a mechanical feedback loop into the market. Margin loan balances have climbed by billions of dollars this year as investors borrowed to buy more of the same stocks.
The FT’s argument is that this is what self-referential bull markets look like. The AI boom has generated extraordinary demand for memory chips, which has produced extraordinary earnings for Samsung and SK Hynix, which has attracted retail money, which has pushed valuations higher, which has encouraged more AI infrastructure spending and more chip orders. Each loop reinforces the others, and the equity market becomes a financing vehicle for the very capex cycle that feeds it. The risk, the column argued, is that the loop runs in reverse just as efficiently: if AI demand disappoints, or if any link in the chain — chip prices, earnings, retail sentiment, the double-exposure products — breaks, the unwind amplifies the decline.
There are reasons to take the warning seriously. SK Hynix’s earnings, while record-breaking, have come to depend on a single customer relationship with Nvidia, and any shift in that relationship, whether technological or geopolitical, would hit the Korean market’s largest components directly. Samsung’s foundry business, long its strategic problem child, remains a drain even as its memory business booms. And the retail-driven rally has shown signs of exhaustion before: Korean markets have a history of sharp reversals when debt unwinds, and regulators have already begun warning about the risks of the new products.
The comparison the FT drew was to previous episodes where retail borrowing met concentrated leadership: Japan’s market in the late 1980s, the U.S. technology market in 2000, and more recently the meme-stock and crypto episodes. In each case, the fundamental story was real — Japanese manufacturing, the internet, digital assets — and in each case the market’s internal mechanics, not the fundamentals, determined the scale of the eventual correction. Korea’s AI market, by this logic, is the same pattern with faster transmission: the bull case is genuine, and the structure around it makes the ride more dangerous than the case alone justifies.
The column stopped short of predicting a crash, and the Kospi has continued to rise in the weeks since the warning was published. But the questions it raised are now part of the mainstream conversation about AI stocks everywhere. When the largest beneficiaries of the AI boom are themselves the vehicles of retail speculation, the boom’s durability depends on the discipline of the investors riding it — the least predictable variable in the market. The Korean experiment will be watched closely by every market with a concentration problem of its own.
