Korea's Best Export Month Ever Met Its Worst Stock Month Since 1997
South Korean exports rose 62.8% year over year in July to $98.89 billion, with semiconductor exports at roughly $41 billion — the second consecutive month above $40 billion, a level unprecedented for any major exporting economy. Semiconductor exports jumped 179% and computer sales surged 404%. Over the same month, the Kospi suffered its worst monthly decline since the 1997 Asian financial crisis. Both facts are true. Only one of them can be the right read on the next twelve months.
The numbers, side by side
The fundamentals:
- July exports: $98.89 billion, up 62.8% year over year
- Semiconductor exports: approximately $41 billion, second straight month above $40 billion
- Semiconductor export growth: 179%
- Computer exports: up 404% on AI infrastructure investment by U.S. technology firms
- June semiconductor exports: $44.8 billion, up 199.5% year over year, 43.8% of total exports
The price action:
- Kospi intraday record: 9,385.59 on June 19
- Close on July 29: 5,663.24
- Single-session decline July 28: 10.84%, to 6,023.66
- Samsung Electronics and SK Hynix: down roughly 13.4% and 14% on the worst sessions
- Market volatility: highest level on record for the year
A 40% index drawdown alongside triple-digit export growth is not a normal cycle. It is the market pricing a threat to future margins that current revenue does not yet reflect.
What the market is actually afraid of
The selloff was not about demand. It was about three structural threats that arrived within weeks of each other:
China's CXMT. The Chinese memory maker completed one of Asia's largest IPOs of 2026, with capital designated for DRAM expansion. Reports of a debut surge above 400% on its Shanghai STAR listing crystallized the fear that Chinese commodity DRAM capacity is about to scale.
Domestic Chinese lithography. A state-backed Chinese group began domestic production of immersion lithography equipment. If China closes the equipment gap, the export-control moat protecting Korean and Taiwanese leadership erodes over years rather than decades.
Circular financing scrutiny. Renewed attention on Nvidia's financing arrangements with OpenAI raised questions about how much of the AI demand signal is genuine end demand versus vendor-financed capacity.
There is a fourth, quieter risk. Samsung and SK Hynix plants in China depend on equipment, components and software governed by U.S. export controls. The end-user access validations both firms held were revoked in September 2025 and replaced with annual licences in 2026. Annual renewal is a materially more fragile arrangement than open validation, and it is a recurring political risk with a fixed calendar date.
Why the bear case may be aimed at the wrong tier
The CXMT threat is real, but independent analysis suggests it is aimed at the commodity segment, not the high-margin one.
High-bandwidth memory requires through-silicon via precision and die-stacking yields that analysts have confirmed CXMT does not currently possess. As of mid-2026, roughly 2% of CXMT's wafer capacity targets HBM, and at the HBM3E generation — already one behind what Samsung and SK Hynix are shipping. Samsung has since unveiled HBM4.
| Memory tier | Who competes | Margin profile | CXMT threat |
|---|---|---|---|
| HBM (AI accelerators) | Samsung, SK Hynix, Micron | Highest | Minimal near-term |
| DDR5 advanced | Korean/US incumbents | High | Emerging |
| DDR4 commodity | Broad, including CXMT, YMTC | Cyclical | Direct and immediate |
| NAND | Multiple | Mixed | Growing |
Nomura's analysts have argued Samsung and SK Hynix should be reframed as structural AI infrastructure assets rather than traditional memory cyclicals, publishing target prices well above prevailing levels. Whether that reframing survives contact with a supply cycle is the open question.
The genuine risk on the other side is timing, not technology. Samsung, SK Hynix and Micron are all accelerating capacity expansion into 2027. TrendForce projects a DRAM supply-demand gap of roughly 1% to 2% in 2026, widening further in 2027. A widening gap sounds bullish — until the capacity being built to close it arrives all at once.
What Canadian investors should take from this
Direct Korean equity exposure is uncommon in Canadian portfolios, but indirect exposure is not. Emerging-market and all-world ETFs carry meaningful Korean and Taiwanese semiconductor weight, and Taiwan's memory names have been running in the opposite direction — five Taiwanese memory stocks hit limit-up on August 3, with combined market value of four bellwethers rising roughly NT$220.8 billion in one day, even as Samsung and SK Hynix fell about 9% on profit-taking.
That is a rotation within the same theme, across borders, invisible at the index level. If you hold a broad EM fund and think you own "Asia semiconductors," you own a specific mix of winners and losers in an active reshuffle.
Check what you actually hold with the Quorum AI scanner, and track the names through My Watchlist.
Bottom line
Korea is running the single best export performance of any major economy on record while its stock market prices a competitive collapse. Those two things resolve within eighteen months, and the resolution runs through DRAM supply in 2027, not through this quarter's shipments.
Frequently asked questions
How much did South Korean exports grow in July 2026?
Exports rose 62.8% year over year to $98.89 billion. Semiconductor exports jumped 179% to roughly $41 billion, the second consecutive month above $40 billion, and computer exports rose 404%.
Why did the Kospi crash if exports are at records?
The selloff reflected structural competitive threats rather than weak demand: China's CXMT scaling DRAM capacity, Chinese domestic lithography production, scrutiny of AI circular financing, and fragile annual export-control licences for Korean fabs in China.
Can Chinese memory makers compete with Samsung and SK Hynix?
In commodity DRAM, increasingly yes. In high-bandwidth memory — the high-margin tier serving AI accelerators — analysts assess that CXMT currently lacks the required through-silicon via precision and die-stacking yields, with roughly 2% of wafer capacity targeting HBM at a generation behind the leaders.
Primary sources
Disclaimer: Educational content only. Not investment advice. Company names appear as illustrations, not recommendations. Figures reflect data available as of August 5, 2026. Written by Elizabeta Dimoska. See our editorial standards.
Elizabeta Dimoska
Founder and writer of RiskStock. Self-directed investor covering ETFs, long-term investing, tax-advantaged accounts (TFSA, RRSP, Roth IRA, 401(k)), retirement, macro, and markets — in plain English, with every claim tied to a primary source. Not a licensed financial advisor; RiskStock is educational. See our editorial standards.
More from Elizabeta Dimoska →
Comments