South Korea's KOSPI crashed nearly 10% on June 23, triggering a circuit breaker for the fourth time this year, after Broadcom's Q3 AI chip guidance of $16 billion missed Wall Street's $17.2 billion estimate by 7% , a rare stumble that revealed just how fragile the concentration underpinning the AI hardware supercycle has become.
The numbers from Broadcom's second-quarter fiscal 2026 results, reported June 3, were not actually bad. Revenue came in at $22.19 billion, up 48% year-on-year. AI semiconductor revenue hit $10.8 billion, a 143% increase from the prior year. Free cash flow reached $10.26 billion. CEO Hock Tan disclosed that AI orders had exceeded $30 billion, triple the revenue shipped, extending visibility well into 2028. By almost any conventional measure, Broadcom delivered. The problem was that it didn't deliver enough. When the company guided Q3 AI chip revenue to $16 billion and declined to raise its full-year AI semiconductor forecast, a market that had already priced in perfection across a 63% rally since late March decided it had heard enough. AVGO fell roughly 14% on June 4.
Seoul felt that in ways no other city on Earth quite would. The KOSPI's brutal June 23 session , down 9.99%, its fifth-largest single-day fall on record , was not solely Broadcom's fault. Three things converged at once. MSCI announced, again, that South Korea would not be added to its Developed Markets watchlist, removing the passive-inflow catalyst that foreign investors had been pricing in. Regulators simultaneously raised concerns about newly approved leveraged single-stock ETFs tied to Samsung and SK Hynix, products that had only cleared approval the prior month. And the Federal Reserve's hawkish June 17 meeting had already unnerved retail investors who had borrowed heavily to chase chip stocks. When the selling started, it fed itself. Foreign investors dumped roughly 5.79 trillion won, around $3.8 billion, in a single session.
Samsung Electronics fell about 12% on the day. SK Hynix dropped roughly 12.5%. Together, the two companies account for nearly half the KOSPI's total market weight , somewhere between 40% and 48% depending on the session , and had contributed an estimated 70% of the index's 2026 gains before Monday. That is not a diversified equity market. It is a memory chip referendum wrapped in a national index.
Here's the thing about Broadcom's guidance: the $16 billion Q3 AI chip figure still represents more than 200% year-on-year growth. The miss was against the Street consensus, not against any fundamental deterioration in demand. Hyperscaler capex, from Google, Meta, Microsoft, and others, remains aggressive. Broadcom's own order backlog stretching through 2028 is not the posture of a company watching customers pull back. What the miss actually exposed is something more mechanical: the AI trade had become a guidance arbitrage game. Beating earnings was no longer sufficient. You had to beat the whisper number and raise the ceiling. Broadcom did neither.
That matters for Korea in a very specific way. The KOSPI's dependence on Samsung and SK Hynix made it one of the purest expressions of the AI memory supercycle thesis anywhere in public markets. When the memory supercycle was the consensus, that concentration looked like focus. When a single earnings call in San Jose introduced doubt, it looked like fragility. As CNBC's analysis made clear after the June 23 session, the MSCI decision compounded the damage because the $30 billion in passive inflows analysts had estimated would follow a developed-market upgrade evaporated on the same day.
SK Hynix had in fact just overtaken Samsung Electronics as the KOSPI's largest company by market capitalization for the first time in the index's history , a sign of how thoroughly the memory chip trade had reshuffled Korean capital markets. That the same reshuffling made the index more vulnerable, not less, to a single data point out of Broadcom is the real story of June 23.
Whether the broader AI infrastructure trade is genuinely softening or whether this was an oversold reaction to a whisper miss is not yet resolved. Broadcom's $30 billion-plus order backlog and 200%-plus year-on-year AI revenue growth are not the numbers of a stalled cycle. But institutional investors watching the KOSPI drop 10% in a day on one guidance line now have a very concrete reason to think harder about how much concentration risk they're carrying in AI infrastructure names, and what happens when the next quarter comes in at $17 billion instead of $18 billion.
The MSCI barrier gives the structural problem a different dimension. Korea cannot attract the stable, passive, index-weight-driven inflows that developed-market status would bring because offshore won trading remains prohibited and market access infrastructure hasn't met the threshold. That means the KOSPI is more dependent on momentum-driven, leverage-amplified retail and foreign speculative flows than a market of its sophistication should be. Fix the won trading restrictions, clear the MSCI hurdle, and a repeat of June 23 becomes less likely. Leave them in place, and the next Broadcom earnings call is always a potential circuit breaker.
Also read: GCash parent Mynt files for a $1.5 billion IPO that would reset what Southeast Asian fintech is worth • Solana's price crash to multi-year lows tells only half the story of what the network is actually doing • X Money rolls out to more verified users and its 6% savings rate is just the opening move