South Korea’s benchmark KOSPI index crashed for the second consecutive trading day on Wednesday, July 29, triggering a market-wide circuit breaker for the first time in consecutive sessions in the exchange’s history — a milestone that erases any pretense that Tuesday’s brutal selloff was an overreaction to be bought. The two-day decline has now stripped more than 18% from the index and put July on course to be the worst calendar month in KOSPI’s recorded history, with losses exceeding 30% — erasing more than a year’s worth of gains from a rally that had made South Korea one of the world’s best-performing markets earlier in 2026.
The immediate catalyst was SK Hynix’s second-quarter 2026 earnings release, which produced the most profitable quarter in the company’s history — and still fell short of what the market required. By the time Wednesday’s circuit breaker halted trading, the index’s collapse over two sessions had destroyed more than ₩396 trillion (approximately $270 billion; exchange rates as of July 29, 2026; conversions are approximate) in combined Samsung Electronics and SK Hynix market value, according to Korea Exchange data reported by TechTimes.
But the earnings miss is only the surface of what happened this week. Beneath it, three separate structural threats arrived simultaneously — China’s CXMT completing Asia’s largest IPO of 2026, a Chinese state-backed company beginning domestic production of semiconductor lithography machines that Western export controls were designed to keep out of Beijing’s reach, and renewed scrutiny over Nvidia’s $250 billion financing guarantee to OpenAI that critics have called circular. Together, those three developments challenged the foundational investment thesis behind two years of record-breaking AI memory profits: the assumption that South Korean chipmakers hold a permanent technological moat that no competitor can breach, at any price.
SK Hynix Posted a Record Quarter and Still Missed
SK Hynix reported Q2 2026 operating profit of ₩60.54 trillion (approximately $41.8 billion), a 557% year-over-year increase, on revenue of ₩79.32 trillion (approximately $54.8 billion), which was 257% higher than the same period last year. The company’s operating margin reached 76%, an all-time record for any memory chipmaker. Cumulative first-half revenue crossed ₩100 trillion (approximately $69.1 billion) for the first time in SK Hynix’s history.
None of that was enough. According to LSEG SmartEstimate data weighted toward analysts with the strongest recent forecast track records, the consensus had projected operating profit at approximately ₩64 trillion (approximately $44.2 billion) and revenue at approximately ₩84 trillion (approximately $58.1 billion). SK Hynix missed on both counts: 6.6% below the operating profit consensus and 5.5% below the revenue consensus. The Korean-listed shares fell as much as 15.81% intraday to ₩1.305 million (approximately $902) before recovering to close at ₩1.401 million (approximately $968), down 9.61%, according to Seoul Economic Daily. The company’s Nasdaq-listed American Depositary Receipts — which had launched on July 10 in the largest foreign IPO in U.S. history, raising approximately $26.5 billion — had already fallen 8.76% on Tuesday to close at an all-time ADR low of $130.49, well below the listing price.
The reason for the miss was technical, not structural — a fact that management tried hard to communicate on the earnings call. SK Hynix’s next-generation HBM4 product, the high-bandwidth memory chip at the center of AI accelerator demand, began mass production shipments in Q2, but the ramp was slower than analysts had modeled, delaying revenue recognition into the second half of the year. Samples of the following-generation HBM4E — which stacks 12 or more layers of DRAM dies connected by hundreds of thousands of microscopic copper channels called through-silicon vias — were distributed to major customers in the first half of 2026, with volume production targeted for 2027.
Han Ji-young, a researcher at Kiwoom Securities, offered a blunt assessment of what drove the selling: "The essence of today’s plunge is that, as expectations for a rebound following the 10% drop the previous day retreated, most shareholders are locking in losses and triggering panic selling."
Management Pushed Back Hard on AI Slowdown Fears
On the earnings call, SK Hynix executives offered a firm rebuttal to market pessimism. Management said customer demand continues to outpace supply and that the company has completed long-term agreements with approximately ten major customers — typically covering five years with volume commitments, price-stabilization mechanisms, and financial deposits — which provide multi-year demand visibility that spot orders cannot reflect.
"Our major customers are still requesting more memory supply," said Song Hyeon-jong, SK Hynix’s President of Corporate Center, on the call. On the question of whether AI infrastructure investment is plateauing, management argued the opposite: that hyperscale customers are using existing infrastructure more efficiently and accelerating the push toward monetization, broadening AI adoption rather than constraining it.
The company raised its full-year capital expenditure guidance to the high end of its previously stated ₩40 trillion (approximately $27.6 billion) range, accelerating the production ramp at its M15X facility and pulling forward the mass production schedule. For the third quarter, SK Hynix projected DRAM shipments to grow approximately 10% and NAND shipments to rise by a low single-digit percentage, with pricing momentum expected to remain positive as AI server demand keeps supply tight at the high end.
The company also pointed to its balance sheet: net cash of ₩69.4 trillion (approximately $48.0 billion) following a ₩88 trillion (approximately $60.8 billion) cash position, and a debt-to-equity ratio of just 7% — the financial foundation of a company betting confidently on continued AI infrastructure spending.
What Is High-Bandwidth Memory and Why Can China Not Simply Copy It?
SK Hynix’s dominant position in the AI memory market rests on a technology called high-bandwidth memory, or HBM, that is structurally different from the commodity DRAM chips in desktop computers and data center servers. In conventional DRAM, memory chips sit side by side on a circuit board; data moves along comparatively slow electrical pathways. HBM stacks multiple DRAM dies vertically — the current HBM3E generation uses up to 12 layers — and connects them through tens of thousands of through-silicon vias, microscopic copper channels punched through each die. The result is memory bandwidth of 1.2 terabytes per second, compared to approximately 51 gigabytes per second for conventional DDR5 — a 23-fold advantage that makes HBM the only memory architecture capable of feeding the enormous computational appetite of modern AI accelerators like Nvidia’s H100 and B200.
SK Hynix holds approximately 56% of the global HBM market, a dominant position it built by pioneering the technology and maintaining a lead in manufacturing yield and packaging complexity. HBM manufacturing is orders of magnitude more complex than commodity DRAM: the yields required for stacking and bonding twelve silicon dies with sub-micron precision are beyond what CXMT, the Chinese DRAM champion, can currently produce. This distinction is what separates Tuesday and Wednesday’s selloff from a simple "China is catching up" narrative. CXMT poses a credible medium-term threat to commodity DRAM pricing; it does not today, and cannot in the near term, challenge SK Hynix’s position in the specialized HBM market that actually drives the company’s record margins.
The bear case on SK Hynix, properly stated, is not that CXMT can make HBM. It is that CXMT’s commodity DRAM expansion, funded by its $8.6 billion IPO war chest, could suppress the broader DRAM pricing environment — and that pricing pressure in conventional memory could eventually erode the premium that has kept SK Hynix’s overall margins at 76%.
Three Shocks Arrived in the Same Week
Tuesday’s collapse was not about SK Hynix’s earnings in isolation. Three separate structural developments converged in the same 48-hour window, and their combined weight is what turned a large selloff into a historic rout.
China’s CXMT completed Asia’s largest IPO of 2026. On July 27, ChangXin Memory Technologies debuted on Shanghai’s STAR Market after raising ¥57.92 billion (approximately $8.6 billion) in Asia’s largest IPO of the year. Shares surged approximately 466% on the first day of trading, briefly making CXMT the most valuable company listed on a mainland Chinese exchange with a market capitalization of approximately ¥3.3 trillion (approximately $480 billion). CXMT secured a 7.67% share of the global DRAM market per its IPO prospectus, and the capital raised is designated for wafer production expansion, DDR5 development, and R&D — the investments that, over a multi-year horizon, could meaningfully alter the commodity DRAM supply picture.
Analysts at Mirae Asset Securities noted that investor concern centered less on CXMT’s present earnings and more on what its $8.6 billion war chest could finance: a rapid expansion of manufacturing capacity that could flood commodity DRAM markets in a cycle that the Korean chipmakers’ capital spending plans had not fully priced in.
A Chinese state-backed company began domestic production of immersion DUV lithography machines. On July 27, Reuters and The Information reported that a state-backed Shanghai group — identified by Reuters as Shanghai Aishengna Electronic Technology Group and described by The Information as incorporating teams from Shanghai Yuliangsheng Technology — had begun manufacturing immersion deep ultraviolet lithography machines domestically, with the first deliveries scheduled to SMIC, Hua Hong Semiconductor, and CXMT later this year. Production targets approximately five machines in 2026 and around 20 in 2027.
Immersion DUV lithography is the most advanced chipmaking tool that Western export restrictions have allowed Chinese foundries to acquire. The machines use a 193-nanometer argon fluoride laser focused through a water medium, enabling circuit patterns in the 28-nanometer to 7-nanometer range through multi-patterning techniques. ASML, the Dutch monopoly supplier of such systems, ships approximately 130 immersion systems per year. Chinese output of five units in 2026 represents less than 4% of that scale, and independent analysts including TrendForce confirmed that the domestic systems lag ASML on yield, throughput, and reliability — advantages built over decades of ecosystem development that cannot be replicated quickly. What the development does challenge, substantively, is the decade-long strategic assumption that Western export controls could permanently cap China’s chipmaking capability below advanced nodes.
Nvidia’s $250 billion OpenAI financing guarantee reignited circular financing concerns. Over the weekend, the Wall Street Journal reported — and Bloomberg and Reuters independently confirmed — that Nvidia is in talks to provide approximately $250 billion in financing guarantees to backstop OpenAI’s lease of a 10-gigawatt data center campus in southern Ohio. The structure is straightforward in its circularity: Nvidia guarantees OpenAI’s data center debt; OpenAI fills the data center with Nvidia chips; Nvidia books the chip sales as revenue. Michael Burry, who disclosed an expanded short position against Nvidia on July 24, wrote on social media that the arrangement amounts to "around and around we go." Portfolio manager Gary Tan at Allspring Global Investments stated that "investors remain concerned about circular financing" even as Nvidia’s CEO Jensen Huang dismissed such characterizations as "preposterous."
The circular financing concern matters for Korean chipmakers because it raises the question that the entire rally rested on: is AI infrastructure demand real, independent, end-user-driven demand — or is it in part a financing loop that inflates the appearance of demand without proportionate end-user consumption?
Export Controls Face Their First Real Test
The strategic significance of the DUV lithography development extends beyond the near-term selloff in Korean chip stocks. For the past decade, the United States and its allies have operated on the theory that denying China access to advanced chipmaking equipment — primarily ASML’s extreme ultraviolet (EUV) systems, which enable chip nodes below 7 nanometers — would permanently cap Chinese semiconductor capability and preserve Western companies’ technological advantage.
Immersion DUV was the highest-capability tool that China could still legally obtain from ASML, enabling 7-nanometer-class chips through multi-patterning. China’s domestic production of such tools — even at the current limited scale and reliability — represents the first time that Beijing has demonstrated the industrial capacity to build the very category of equipment at the heart of the export-control strategy.
It does not close the EUV gap. China’s domestic EUV program remains at a prototype stage, years away from commercialization, and the EUV constraint remains the binding one for cutting-edge chips below 7 nanometers. But the DUV development shows that the equipment-denial strategy, like any constraint imposed from outside, creates strong incentives for the constrained party to build around it — and that China now has the industrial base to begin doing so in a category that the export-control architects assumed was secure.
US lawmakers have responded. The MATCH Act, introduced in April 2026 and reported out of the House Foreign Affairs Committee, would cut SMIC, Hua Hong, CXMT, and other named entities not only from new ASML sales but from servicing and technical assistance — extending restrictions to ASML’s already-installed fleet in Chinese fabs.
The Anatomy of a Two-Day Panic
Context is essential for understanding the scale of what has happened to Korean markets in July 2026. The KOSPI had more than doubled during the first half of the year — at one point crossing 9,000 for the first time in its history on June 18, driven by SK Hynix’s announcement of HBM4E sample shipments — making it one of the world’s best-performing major equity indices. That rally was built almost entirely on Samsung Electronics and SK Hynix, which together account for more than 40% of the KOSPI’s total market capitalization, creating an index that functions, in structural terms, as a leveraged bet on two companies.
On Tuesday, July 28, the KOSPI plunged 10.84% to close at 6,023.66 — its fourth-largest single-day percentage decline in its history — activating a market-wide circuit breaker, the exchange’s eighth of 2026 alone. Samsung Electronics and SK Hynix each fell more than 13%. By one accounting, the two-session rout erased approximately $270 billion in combined market value between the two chipmakers.
Wednesday’s session deepened the carnage. After an early attempted rebound, the KOSPI reversed into a broad decline and fell 5.98% to close at 5,663.24, triggering the circuit breaker for the ninth time in 2026 and the second consecutive trading session. SK Hynix’s Korean-listed shares fell 9.61% to close at approximately ₩1.401 million (approximately $968), while Samsung Electronics dropped 5.23% to approximately ₩208,500 (approximately $144), slipping below the psychologically significant ₩200,000 level at points during the session.
Foreign investors, who had been net buyers for much of 2026, turned net sellers in Wednesday’s session, offloading approximately ₩419 billion (approximately $290 million). Retail investors sold more than ₩2 trillion (approximately $1.38 billion) net. Institutional investors responded with net purchases exceeding ₩2.4 trillion (approximately $1.66 billion) but could not fully absorb the combined selling pressure.
South Korea’s Finance Minister Koo Yun-cheol appeared before a parliamentary inquiry Wednesday and apologized for the introduction of single-stock leveraged ETFs — products tied to Samsung Electronics and SK Hynix shares that amplify both gains and losses — acknowledging that they were brought to market without sufficient prudent consideration. The Korea Financial Services Commission has separately been reported to be considering tighter leverage controls as a near-term stabilization measure.
Samsung Is Next — and the Stakes Are the Same
Samsung Electronics is scheduled to release its full second-quarter 2026 results, including a business segment breakdown, on July 30, with the earnings call beginning at 10 a.m. KST (9 p.m. ET on July 29). In preliminary guidance filed July 7, Samsung projected consolidated Q2 operating profit of approximately ₩89.4 trillion (approximately $61.8 billion) on sales of approximately ₩171 trillion (approximately $118.2 billion) — figures that would represent roughly a 19-fold increase in operating profit year-over-year. If confirmed, the full results would represent Samsung’s third consecutive record quarterly profit.
The preliminary guidance already produced a familiar dynamic: Samsung’s shares fell approximately 6.9% on the day the guidance was released, as investors focused on the revenue figure, which narrowly missed consensus, rather than the record operating profit, which beat expectations. On Wednesday in Seoul, Samsung shed approximately 5.23% — compounding Tuesday’s 13%+ decline.
Given SK Hynix’s experience — record actual results meeting a market that had already moved past them — the setup for Samsung’s July 30 report is not encouraging for Korean equities in the near term.
Why This Matters Beyond Seoul
The KOSPI’s collapse is a proxy for a larger question that the entire global AI investment community is now being forced to confront: how much of the AI infrastructure spending boom is real, durable, end-user demand — and how much of it is a loop of capital that circulates among a small number of companies, reinforcing the appearance of demand without necessarily reflecting what end users are willing to pay for AI services?
SK Hynix’s own guidance — continued double-digit DRAM shipment growth, pricing strength, long-term agreements with ten major customers — points toward durable demand. The company’s 76% operating margin suggests pricing power at the high-bandwidth memory tier that commodity competition from CXMT, even at scale, cannot immediately undercut. The HBM market and the commodity DRAM market are genuinely different businesses.
But new capacity from SK Hynix, Samsung, and Micron is all entering production simultaneously, and that supply will eventually reach the income statement. For current margins to hold, AI infrastructure demand must not merely remain robust — it must grow fast enough to absorb the incremental supply. That is the bet Korean equity markets were making when the KOSPI crossed 9,000. It is a bet that is now being repriced in real time, against a backdrop where the structural assumptions that underpinned it — permanent Western chipmaking moats, unchallenged AI demand growth, linear capital deployment by hyperscalers — are being examined more skeptically than at any point in the two-year rally.
Frequently Asked Questions
Why did South Korea’s stock market trigger a circuit breaker for two days in a row?
A circuit breaker is a regulatory mechanism that automatically halts trading when a market index drops by a specified percentage — for the KOSPI, the threshold is an 8% decline, which triggers a 20-minute pause. Two consecutive circuit breaker triggers have never happened before in the KOSPI’s history. The back-to-back halts on July 28 and July 29 reflect the convergence of three structural concerns that arrived in the same 48-hour window: China’s CXMT completing Asia’s largest IPO of 2026 with capital designated for DRAM expansion, China beginning domestic production of semiconductor lithography equipment that Western export controls were designed to keep out of reach, and questions resurfacing about whether Nvidia’s $250 billion financing guarantee to OpenAI reflects sustainable AI demand or a circular financing arrangement. SK Hynix’s record Q2 earnings — which still missed analyst consensus — provided the immediate trigger, but the structural concerns are what turned a selloff into a historic rout.
What did SK Hynix actually report, and why did the stock still fall?
SK Hynix reported Q2 2026 operating profit of ₩60.54 trillion (approximately $41.8 billion), a 557% year-over-year increase, with revenue of ₩79.32 trillion (approximately $54.8 billion) and an all-time record operating margin of 76%. By any conventional measure, those are extraordinary results. The problem is that the stock’s valuation had been pricing in an even more extraordinary outcome: analyst consensus had expected approximately ₩64 trillion (approximately $44.2 billion) in operating profit and ₩84 trillion (approximately $58.1 billion) in revenue. SK Hynix missed both. The miss was driven by slower-than-expected shipment ramp-ups for its next-generation HBM4 chips, which pushed some revenue recognition into the second half of 2026. The mechanism that produced the selloff — "priced for perfection" — is a well-documented market dynamic in which a stock’s valuation incorporates the maximum plausible outcome, leaving no reward for a result that is merely historic rather than miraculous.
Does China’s domestic DUV lithography production change the long-term outlook for Korean chip stocks?
It changes the structural framing more than the immediate competitive picture. What China has accomplished — beginning domestic production of immersion deep ultraviolet lithography machines, which use a 193-nanometer laser and water medium to etch circuit patterns at the 28-nanometer to 7-nanometer range — is the first demonstration that Beijing can manufacture the core category of chipmaking tool at the center of the Western export-control strategy. Current output is approximately five machines in 2026, compared to ASML’s roughly 130 per year, and independent analysts have confirmed that Chinese systems lag ASML on yield, throughput, and reliability. The gap to extreme ultraviolet lithography, which enables chips below 7 nanometers, is also unaffected by this development. However, the long-term implication — that the permanent-moat assumption underpinning SK Hynix and Samsung’s premium valuations may need to be reassessed — is what the market is beginning to price in. The selloff is, in part, a valuation correction for a thesis that just became less certain.
Is SK Hynix’s dominance in AI memory at risk from CXMT?
In the short term, at the high-bandwidth memory tier that drives SK Hynix’s record margins, the answer is no. HBM requires vertically stacking multiple DRAM dies and connecting them through tens of thousands of microscopic copper channels — a manufacturing process of extraordinary complexity that CXMT cannot currently replicate. CXMT’s competitive threat is real in the commodity DRAM market, where the company holds approximately 7.67% global market share and its $8.6 billion IPO proceeds are specifically earmarked for capacity expansion. If CXMT floods commodity DRAM markets, it could suppress broader memory pricing and put pressure on SK Hynix’s overall average selling prices — but it does not directly threaten the high-bandwidth memory tier that Nvidia, AMD, and other AI accelerator manufacturers specifically require from SK Hynix. The more pertinent risk is a longer-term one: what happens to HBM pricing if and when AI infrastructure spending growth decelerates to a pace below what current semiconductor supply additions assume.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.