Why Korean Markets Are Warning About the AI Bubble Before Wall Street

The sell-off that Seoul saw coming

The Korean financial press didn’t wait for Wall Street to admit the AI infrastructure trade was fraying. On August 16, a single Goldman Sachs report made its way through Seoul’s trading desks and into headlines months before the US news cycle caught up: hedge funds had been net-selling Goldman’s Global AI Basket for five consecutive trading days starting August 8, marking the second-largest annual outflow recorded in the bank’s metrics.

More importantly, these weren’t new short bets. They were unwinding longs — actual capital exiting positions accumulated during the AI euphoria of early 2025. The money shifted into healthcare and consumer staples. That’s not a tactical rebalancing. It’s a portfolio-level de-risking signal from institutions that typically move last, and when they move, everyone else scrambles to follow before the window closes.

The timing tells its own story. Korean outlet Maeil Business Newspaper broke the finding on August 17, while American financial press barely acknowledged it until mid-September. By then, the damage was already propagating through exchange-traded products linked to AI infrastructure exposure. The sequencing — Seoul spotting it first — wasn’t coincidence. Korea’s semiconductor ecosystem sits at the exact chokepoint of AI demand, making Korean fund managers acutely sensitive to shifts in capital expenditure cycles that American observers, insulated by language barriers and different market hours, missed entirely.

CoreWeave provided the catalyst. The company’s attempt to scale its GPU rental business through heavy debt financing collided with rising borrowing costs and investor doubt about repayment capacity. The stock fell 12% on the same day as that Goldman report. The Philadelphia Semiconductor Index dropped 11.41% that week alone — the worst performance since June’s 4.6% decline. What made the move particularly destructive was that it wasn’t just CoreWeave itself dragging down the sector. The panic spread to companies with zero direct relationship to the distressed lender, suggesting the market was pricing in systemic risk rather than idiosyncratic failure.

Why Korea matters here

The connection isn’t abstract. American hedge funds that took concentrated positions in SK Hynix and SanDisk — losses that were already public knowledge by mid-October — found themselves caught in a squeeze when Korean semiconductor stocks turned. The same funds that bet big on memory chip demand from AI data centers became the ones forced to sell them. This created a perverse dynamic where the very institutions inflating Korean valuations became the ones liquidating them, amplifying every move in both directions.

This created the exact feedback loop that destroyed the KOSPI in July 2024. When one fund liquidates to limit losses, prices fall, triggering margin calls and forced selling from others. The situational awareness fund — a leveraged AI infrastructure bet managed by former OpenAI researcher Leopold Asensio — collapsed under this mechanism last summer. It borrowed up to four times its capital, went long semiconductors and short software, and when the trades reversed, competitors identified its book and shorted it into oblivion. Citadel bought the entire portfolio at a fire sale, emerging from the carnage with assets acquired at fractions of their book value.

Korean markets are living through that memory again. The KOSPI has dropped more than 20% from around 7,000 to below 5,000 this October — a mirror of July’s carnage. And the rumor mill is already spinning: VARA, a US-based AI-focused fund holding $40.1 billion in common stock and options across 123 positions as of late June, is supposedly being liquidated. Bloomberg Energy, Navios, and CoreWeave are among its largest holdings. Jim Cramer amplified the speculation on X, writing that hedge funds might handle liquidations and “start fresh on Monday.” Whether that happens or not, the contagion narrative is already priced in, and Korean institutional investors are bracing for a second wave that targets the very same semiconductor names they watched collapse eight months prior.

The structural problem underneath

Macro headwinds are making things worse. The US midterm elections next month are pushing investors toward defensive positioning. Renewed military tension between the US and Iran threatens to spike oil prices, which would pressure the Federal Reserve into maintaining higher rates for longer. Higher rates directly increase the cost of capital for data center operators like CoreWeave — exactly the companies banking on cheap debt to fund their GPU fleets.

This isn’t just a sentiment problem. It’s a balance sheet problem. The entire AI infrastructure thesis assumes that hyperscalers can borrow at favorable terms indefinitely to fund capex that may take years to generate returns. When borrowing costs rise, the math changes. Models that assumed 5% debt service now face 8%, 9%, or higher — and those incremental points erase hundreds of millions in projected margins across multi-year construction timelines. The companies most exposed aren’t the ones building the chips; they’re the ones leasing the capacity, carrying the debt, and hoping demand catches up before the interest payments do.

BofA analyst Tal Riani offered a counterpoint: CoreWeave’s ability to deliver active power at scale remains a genuine competitive advantage, and the company has ways to manage its risk exposures. Past performance, he argued, supports the thesis. But fundamentals don’t inoculate against liquidity events. When forced sellers emerge, it doesn’t matter how good the underlying business is. Solvency and liquidity are different problems, and right now, the market is pricing in a liquidity crunch that no amount of operational excellence can prevent. CoreWeave may survive. The question is whether other highly leveraged players in the infrastructure stack will face the same fate, and whether that contagion will reach Korean semiconductor suppliers who appear healthy on paper but are caught in the crosscurrents of American hedge fund deleveraging.

Second-order effects and broader implications

The ripple effects extend well beyond semiconductor valuations. Korean battery manufacturers, which have been diversifying into AI data center power solutions, are watching their premiums compress as the broader AI infrastructure narrative fractures. Companies like LG Energy Solution and Samsung SDI, which positioned themselves as enablers of the green AI transition through battery-backed power systems, are seeing their strategic narratives questioned alongside their stock prices. If hyperscalers delay capex due to financing constraints, the entire auxiliary supply chain — cooling systems, power management, backup batteries — faces demand revisions downward.

Retail investors in Korea, who poured into AI-themed ETFs at peak valuations, are now facing margin calls on products that track exactly the same semiconductor names institutional investors are dumping. The psychological impact mirrors what happened in July: small investors, seeing their portfolios halved, sell into weakness, creating additional downward pressure that institutional buyers once again exploit. This time, however, there’s less fresh capital entering the market to absorb the selling, because the same Goldman data showing hedge fund exits is visible to every retail trader using Korean financial apps.

The Yuan-Won currency pair has also shifted meaningfully, with the won strengthening as Korean institutional investors repatriate capital from American semiconductor positions. This creates an additional headwind for Korean exporters, whose competitiveness depends on a weaker won, and adds another layer of complexity to a situation that was already straining corporate profit margins.

Who wins and who loses

The winners from this rotation are defensive sectors — healthcare, utilities, consumer staples — where money is flowing as institutions shed AI exposure. The losers are clearer: Korean semiconductor exporters, any company with high leverage in the AI infrastructure stack, and retail investors who piled into AI-themed ETFs at peak valuations.

If Q3 earnings from IBM, Tesla, and Meta — all reporting around October 21-22 Korea time — come in strong, the panic could ease. Earnings are the circuit breaker that separates a correction from a crash. But the underlying tension remains unresolved: AI capex is running ahead of revenue generation, and the debt fueling that gap is becoming more expensive by the quarter. Even positive earnings may not convince skeptical investors that the revenue will materialize on the timeline markets are pricing, especially when the cost of carrying the related debt is climbing faster than operating margins can expand.

The clearer close

Korean financial media flagged the danger weeks before US outlets. That sequencing matters. When the world’s most important memory chip suppliers are in Seoul, and the funds that bet on them are American, the next leg of this unwind won’t follow Wall Street’s timeline. It will follow whoever gets margin called first — and right now, the data suggests that person is not in New York.

The July 2024 crash wasn’t an anomaly. It was a preview. What happened then happened fast, happened violently, and happened to people who didn’t see it coming because they weren’t watching the same signals. This time, the signals are louder, the interconnections are deeper, and the players on both sides of the Pacific are reading each other’s moves in real time. The Korean market isn’t just echoing American dysfunction — it’s anticipating it, positioning for it, and preparing for the moment when American hedge funds realize their exits have already been priced into a market that never stopped moving.

The AI bubble didn’t pop because the technology failed. It popped because the financial engineering around it did. And until the balance sheets clear, until the debt gets restructured or retired, until hyperscalers prove they can generate returns that justify their borrowing, the unwinding isn’t over. Korea saw it first. The rest of the world will feel it soon enough.