On July 28 and 29, 2026, the Korea Exchange (KRX) triggered Stage 1 circuit breakers in both the KOSPI and KOSDAQ markets. According to the local financial newspaper Seoul Economic Daily, this marked the first time KOSPI had triggered a circuit breaker on two consecutive trading days, and the third time for KOSDAQ, following instances in 2008 and 2020.

Some observers have described the sharp decline as an "AI bubble collapse." But on the morning of the 29th, semiconductor giant SK hynix reported its April–June revenue and operating profit, both of which were the highest ever recorded for a quarter. Stock prices can fall even after strong earnings—if expectations already priced in were even higher, then a record profit does nothing to stop selling.

What can be reliably read from this price action is not that AI demand has disappeared. Rather, it reflects a market structure in which index concentration in two semiconductor companies has intensified, and South Korea's own financial regulators had already been wary of the amplification pathway created by single-stock leveraged products that had swelled rapidly in a short period.

AD

Halted for 20 Minutes After an 8% Drop, With a Separate 10-Minute Reopening Process

Laying out KRX's disclosures in chronological order, the first trigger occurred on KOSPI at 10:13:43 on the 28th. The index fell 8.02% from the previous day's close of 6,755.75 to 6,213.51. At 12:01:12 the same day, KOSDAQ also met the trigger condition, down 8.09% from the previous day at 702.91. The previous-day closing values cited in KRX's notice the following day, the 29th, were 6,023.66 for KOSPI and 705.85 for KOSDAQ—down 10.84% and 7.72% day-over-day, respectively.

Selling did not stop on the 29th either. KOSDAQ triggered again at 12:19:12, down 8.05% at 649.00, and KOSPI followed at 12:32:32, down 8.15% at 5,532.33, triggering Stage 1 once more. In all four instances, the market-wide halt was triggered because the index remained at least 8% below the previous day's close for one continuous minute.

The mechanism itself is not simply a "30-minute trading halt." In Stage 1, trading is suspended for 20 minutes, during which only cancellations of existing orders are accepted. This is followed by a 10-minute single-price auction period in which orders are collected and a price is determined before continuous trading resumes. Combining the 20-minute halt and the 10-minute reopening procedure totals 30 minutes, but under KRX rules, the halt itself lasts 20 minutes.

If the decline deepens, Stage 2 is triggered once the index falls 15% or more and remains at least 1% below the Stage 1 trigger level for one continuous minute, again resulting in a 20-minute halt. If the index falls 20% or more and remains at least 1% below the Stage 2 trigger level, Stage 3 is triggered, ending trading for the day.

The "sidecar" mechanism, which halts program-trading order quotes for five minutes in response to a sudden change in futures prices, is a separate system. Circuit breakers halt the entire market. Distinguishing between the two makes it easier to see that multiple safety mechanisms activated one after another on the 28th and 29th, and to understand the scope of orders each mechanism suspended.

SK hynix's Operating Profit Rose 6.57-Fold, Yet Market Selling Continued

SK hynix's April–June results, released on the 29th, showed revenue of KRW 79.3187 trillion and operating profit of KRW 60.5426 trillion. Year-over-year, revenue rose 257% and operating profit rose 557%, with operating profit expanding 6.57-fold from KRW 9.2129 trillion in the same period last year. Rising prices for high-performance memory used in AI servers, as well as DRAM and NAND, contributed to the results.

Leading indicators for the business also show little sign of demand collapse. SK hynix stated that it had completed long-term supply contract negotiations with more than 10 companies, including major customers, and that mass shipments of HBM4 began in the second quarter. The company plans to significantly ramp up production in the second half of the year. That said, the earnings figures are pre-audit. Details of contract terms have not been disclosed, so their effects on future shipment volumes or pricing cannot be confirmed.

Stock prices move not on the absolute size of profits but on the gap between results and pre-earnings expectations. If investors were anticipating even higher profits or greater shareholder returns, a headline of record profits and a selling reaction can coexist. Overlaid on this is index skew. According to data from the Financial Services Commission (FSC), the combined weight of Samsung Electronics and SK hynix in KOSPI's total market capitalization rose from 34% at the end of 2025 to 41% at the end of April 2026, 49% on May 26, and 52% on July 15.

In a market where the combined market capitalization of these two semiconductor companies exceeds half of KOSPI as a whole, revaluations of either company can easily spill over into the entire index. This circuit breaker episode does not indicate that South Korea's listed companies broadly suffered the same deterioration in performance all at once. What matters is that the rally driven by AI expectations had rendered the index highly sensitive to the price movements of a small number of stocks.

AD

Single-Stock Leveraged Products Launched in May Reach KRW 11.9 Trillion in Seven Weeks

Another key issue is the single-stock leveraged ETFs and ETNs launched in South Korea on May 27, 2026. According to the FSC, the combined market capitalization of 16 products grew from KRW 4.4 trillion at launch to KRW 11.9 trillion by July 15. Trading value also expanded, from KRW 10.4 trillion to KRW 13.0 trillion. The annualized volatility of Samsung Electronics and SK hynix reached 96% and 113%, respectively, between May 26 and July 10.

Daily double-leveraged products aim to deliver twice the "daily" percentage change of the underlying stock price, adjusting their positions every day. If rebalancing trades that realign the leverage ratio after large price swings become concentrated just before market close, this could amplify late-session price movements through supply and demand in the underlying stock and derivatives markets. The FSC explicitly raised this pathway at an industry meeting on the 28th and asked asset managers to spread out rebalancing timing. At the same time, it acknowledged that moving up the rebalancing timing could widen tracking error for these products.

Regulatory action did not begin suddenly after the crash. On July 16, the FSC temporarily suspended new listings of single-stock leverage, inverse, and covered-call products, as well as advertising and promotional events for existing products. The plan to raise the minimum deposit requirement from KRW 10 million to KRW 30 million in cash was moved up in the July 28 announcement, with implementation now set for July 31. An additional one hour of prior investor education and enhanced evaluation will be implemented early, while strengthened management of tracking-error ratios is scheduled for August 19.

If demand does not cool, the FSC has stated it will also consider investment-experience requirements and per-investor investment caps. However, these remain options under consideration, not regulations already in effect. Moreover, while the explosive growth of leveraged products and their rebalancing mechanisms is described as something that "could amplify volatility," this is not evidence that they alone caused the two-day decline. Even the regulator's own language stops short of assigning causation, expressing concern about amplification rather than confirming a cause.

Judging an "AI Bubble Collapse" Requires More Evidence of Deteriorating Real Demand

What this sharp decline shows is that if bullish expectations underpinning the AI rally were to crumble, concentration in a handful of semiconductor stocks combined with leverage could magnify the index's downside. SK hynix's record profit is evidence that AI demand genuinely exists. At the same time, whether that profit justifies the valuation already priced into the stock is a separate question.

To assess an AI bubble collapse from a fundamentals perspective, one would need to continuously track HBM shipment volumes and pricing, customer capital expenditure, cancellations of long-term contracts, inventory buildup, and order cancellations. From a market-structure perspective, the outstanding balance of single-stock leveraged products, pre-close rebalancing activity, and the index weighting of Samsung Electronics and SK hynix all warrant close monitoring.

A two-day sharp decline alone cannot prove that AI demand has ended. But the fact that buyers did not return even after record profits, and that safety mechanisms were triggered on consecutive days, reflects the possibility that the expectations and position concentration that had propped up South Korean equities were in a state capable of amplifying a downturn. What deserves attention ahead of the next earnings figures is not the size of the profit itself, but whether the real-world demand generating that profit continues to grow—and how much capital remains willing to bet on the same expectations.