The Korean stock index falls for the second day in a row. AI-related stocks plummet. But it is too early to talk about the bubble bursting.
We have finally reached the first real one stress test for what has become the most important stock market rally in recent years, namely that of semiconductors and, more generally, AI.
The collapse of the Korean index
-10.84% on Tuesday and a further -5.98% today. These are the terrifying results of the last two days of trading on the index KOSPI South Korean. Extremely important for AI and semiconductorsbecause this is where the shares of Samsung And SK Hynixwhich together make up 70% of the global chip market.
So have we moved from a phase of “compelling storytelling” to “testing the fundamentals” for AI? The final verdict will come tonight with the results Microsoft And Halfbut Korean data already indicates that the market appears to be punishing the disproportion between explosive capital expenditure (capex) and still insufficient returns.
What happened in two days
The collapse began on Tuesday, with a global sell-off in semiconductors that also affected the Taiwanese and Japanese stock markets: KOSPI -10.84%, Samsung -13.39%, SK Hynix -14.65%, Taiex -4.65%, Nikkei -4.
The pressure continued on Wednesday, with the KOSPI opening 1.09% higher before collapsing to 5,262.77 (-12.63%), to close at 5,663.24 (-5.98%).
SK Hynix collapsed despite announcing record numbers: operating profit of 41.20 billion dollars (+557% YoY) and revenues of 54.55 billion dollars (+257% YoY), but both below estimates 44.01 billion And 57.77 billion dollars.
The context is broader
Broadening the look, the KOSPI has collapsed by 34% in 25 days from the peak reached on June 22nd. Overseas investors dumped $137 billion in Asian stocks in H1 2026, with Korea and Taiwan at the epicenter.
SK Hynix shares plummeted by more than 13% in two daysdespite record-breaking second-quarter 2026 results, is a textbook case of how financial markets price future expectations and not past results.
The market is discounting some precise factors that indicate a potential phase change in the semiconductor cycle for artificial intelligence.
The first factor concerns that 5% decrease in profits compared to consensus estimates. Despite exceptional growth of 557% year-on-year, analysts had incorporated projections that assumed an indefinite acceleration in demand for AI chips.
That 5% less indicates that the growth curve is flattening. Comparison with previous years becomes increasingly difficult and growth of 557%, while extraordinary, represents a slowdown compared to the triple-digit rates of previous quarters.
The second factor is the slowdown of HBM price growththe high bandwidth memory used in AI. According to analysts from Goldman Sachs and TrendForce, HBM prices are highly likely to see their first decline or double-digit decline in 2026.
After years of insufficient supply, increased production capacity and the entry of new players are loosening the market. This gives bargaining power back to major customers like NVIDIA.
The first crack in the AI consensus
The collapse of KOSPI represents a psychological turning point for the global financial market, the first real sign of a breakdown in collective consensus on artificial intelligence.
Until yesterday, the market passively bought the narrative according to which spending on AI infrastructure would be infinite and always profitable. Today, the violent Korean sell-off shows that this unshakable faith is fading.
South Korea acts as the weakest and most sensitive point in the entire global ecosystem. On the one hand, the country is hyper-exposed to semiconductorswhich account for more than twenty percent of its exports, making its economy a direct gauge of the health of the sector.
Eyes on Wall Street
The real epicenter, however, is not in Seoul but in the United Statesand the final verdict on the fate of the tech rally could come between today and tomorrow with the corporate results of Microsoft and Meta.
Many answers could come, primarily to the why despite billion-dollar investments, they fail to generate proportional net profits.
Second, why should investors continue to finance ever-increasing capital expenditure if returns do not materialize. The outcome of these responses will determine the overall direction.
Doubts exist, just think that, according to what emerged in the main American technology publications (Tom’s Hardware, TechRadar), up to 75% of chips sold have not yet entered operation in data centers or operates at drastically low levels of efficiency. Most of the chips sold, in other words, would be unused.
