Inside the Chip Stock Selloff: China’s Homegrown DUV Machines, CXMT’s $480 Billion Debut and Nvidia’s Circular Financing Problem

Last updated: July 28, 2026, 4:45 p.m. Eastern Time

South Korea’s benchmark stock index lost 10.84% on Tuesday, July 28, 2026 — its worst session since the outbreak of the Middle East war in March, and the eighth time this year that the Korea Exchange has been forced to halt all trading. The Kospi closed at 6,023.66, roughly a third below the record it set five weeks earlier. Samsung Electronics fell 13.39%. SK hynix fell 14.65%. In Tokyo, Kioxia lost 18.33% and Tokyo Electron 10.96%. Taiwan’s Taiex dropped 4.65%. By the time New York opened, Micron was down close to 9% and SanDisk close to 15%.

The immediate trigger for the chip stock selloff was not an earnings miss, a guidance cut or a demand warning. It was a 900-word scoop about a machine.

On Monday, July 27, the technology news site The Information reported that a state-backed company in Shanghai had begun manufacturing immersion deep-ultraviolet lithography systems — the class of tool that prints circuit patterns onto silicon wafers, and a market the Dutch company ASML has dominated for two decades. Reuters confirmed the substance of the report the following morning and went further, naming the company: Shanghai Aishengna Electronic Technology Group, a firm with no website, no public disclosures and 7 billion yuan of registered capital, incorporated less than three years ago.

That report landed on a market already braced for trouble. Hours earlier, ChangXin Memory Technologies — China’s largest DRAM maker, universally known as CXMT — had closed its first day of trading in Shanghai up roughly 466%, making it the most valuable company listed on mainland exchanges. And on the same Monday, The Wall Street Journal reported that Nvidia was in talks to guarantee about $250 billion of financing so OpenAI could lease computing capacity from a U.S. data center project, reviving a question that has shadowed the AI trade for a year: how much of the demand for AI chips is being financed by the company selling the chips?

Three separate stories, one direction of travel. Together they did something that individually none of them could have done — they made investors question the two assumptions holding up the most crowded trade in global equities: that the memory supercycle would run for years, and that Western suppliers would remain unchallenged at the top of the semiconductor value chain.

Key Takeaways

• Main development: Global semiconductor and memory shares fell sharply on July 27–28, 2026 after The Information and then Reuters reported that a Chinese state-owned firm had begun mass-producing domestically developed immersion DUV lithography machines, technology long dominated by ASML.
• Key figure: South Korea’s Kospi closed July 28 at 6,023.66, down 10.84% (732.09 points), its lowest close since April 14 and about 34% below its record close of 9,114.55 on June 22, 2026.
• Market response: Foreign investors sold a net 4.97 trillion won (about $3.4 billion) of Korean shares on July 28, according to the Korea JoongAng Daily. The Korea Exchange triggered a marketwide circuit breaker at about 10:13 a.m. Seoul time, halting trading for 20 minutes — the eighth such halt of 2026.
• Second catalyst: Nvidia’s five-year credit default swaps posted their largest one-day widening on record on July 27, according to Bloomberg, after reports the company is working on more than $750 billion of AI infrastructure commitments, including a potential $250 billion backstop for OpenAI.
• Why it matters: Korean memory shares had roughly tripled over twelve months on an AI-driven memory shortage. The July 28 session was the point at which the market began pricing two long-term risks — Chinese import substitution and vendor-financed demand — into a sector trading on peak-cycle earnings.
• What comes next: SK hynix reports second-quarter results on July 29, the Federal Reserve announces its policy decision the same day, and Microsoft, Meta, Apple and Amazon all report capital-spending plans on July 29–30.

What actually happened, hour by hour

The sequence matters, because the market’s reaction was cumulative rather than instantaneous.

The Information published first, on Monday, July 27, reporting that an unidentified Shanghai company backed by the Chinese government had started manufacturing immersion DUV lithography tools, with plans to build roughly five units in 2026 and about 20 in 2027, and with first deliveries expected this year to Semiconductor Manufacturing International Corp (SMIC), Hua Hong Semiconductor and CXMT.

U.S. and European markets reacted the same day. ASML’s Amsterdam-listed shares fell more than 8%, according to Reuters; its U.S.-listed shares closed Monday at $1,655.26, down 5.80%. Nvidia closed at $196.51, down 4.99%, stripping roughly $250 billion from its market capitalization and handing the title of world’s most valuable public company back to Apple. AMD fell more than 5%. The headline U.S. indexes barely moved — the S&P 500 finished the session at 7,413.18, up 0.02%, and the Nasdaq Composite at 24,932.08, down 0.18% — which is itself a clue about what kind of selloff this was.

Asia opened on Tuesday with all of that in the price and more still coming. Reuters published Fanny Potkin’s exclusive at 4:19 a.m. Eastern (8:19 GMT), naming Shanghai Aishengna and adding detail about its ownership and personnel. Korean and Japanese equities were already falling by then.

The Kospi hit the marketwide circuit breaker threshold at around 10:13 a.m. Seoul time, having fallen 8.02% to 6,213.51. Under Korea Exchange rules, a decline of 8% or more sustained for one minute halts all trading for 20 minutes; a separate mechanism, the sidecar, suspends program selling for five minutes when Kospi 200 futures fall 5%. Both fired on Tuesday. Trading resumed and the index kept falling, touching an intraday low of 5,992.91 — down 11.29% — before closing at 6,023.66.

By the U.S. open, the story had turned into something more complicated than a straight risk-off day.

What immersion DUV lithography is, and why one report moved half a trillion dollars

Lithography is the step in chip manufacturing where a pattern is projected onto a silicon wafer coated in light-sensitive material. The finer the pattern a machine can print, the smaller and denser the transistors, and the more capable the resulting chip. Everything else in a fab — deposition, etch, metrology, packaging — is built around what the lithography tool can resolve.

Immersion deep-ultraviolet systems sit in the middle of that hierarchy. They use 193-nanometre light and, as the name suggests, a layer of purified water between the final lens and the wafer. Water has a higher refractive index than air, which effectively shortens the wavelength at the wafer surface and lets the machine resolve finer features than a conventional “dry” DUV tool. Immersion DUV is the workhorse of the industry: it prints the overwhelming majority of the world’s logic and memory chips, and through a technique called multiple patterning — exposing the same layer repeatedly with offset masks — it can be pushed down to nodes that would otherwise require extreme-ultraviolet equipment.

That last point is the one that matters for the market. EUV, which uses 13.5-nanometre light generated by vaporising tin droplets with a laser, is the only tool that can print the most advanced logic in a single exposure, and ASML is the only company on earth that makes it. China cannot buy EUV at all; U.S.-led export controls have prohibited those sales for years. Dutch controls have separately restricted China’s access to ASML’s most advanced immersion DUV systems.

So a domestically produced immersion DUV machine does not, by itself, let China build leading-edge logic. What it does is remove a chokepoint. With enough immersion DUV capacity and enough patience for multiple patterning, Chinese foundries and memory makers can produce a very large share of the chips the world actually consumes — including the DRAM that has been in acute shortage all year — without depending on a supply chain that a foreign government can switch off.

Reuters put the technical caveat plainly. The source who described Aishengna’s programme said the machine “is expected to require further testing and remains far from matching the Dutch firm’s competing models.” Most components are domestic, but some critical parts still come from Japan. These are prototypes graduating into low-volume production, not a finished competitive product.

Markets did not price the machine. They priced the trajectory.

Shanghai Aishengna: what is known, and what is not

Reuters was first to identify the company at the centre of the story, and the profile it assembled is notable mainly for how thin the public record is.

Shanghai Aishengna Electronic Technology Group was established in August 2023 with registered capital of 7 billion yuan, or roughly $1.0 billion at the exchange rate Reuters cited of 6.7677 yuan to the dollar. It has two shareholders, both state-owned: Shanghai Electric Holding and a subsidiary of Shanghai International Trust. It has no website. It has disclosed nothing publicly about its operations. It did not respond to Reuters’ requests for comment, and neither did its shareholders.

What gives the company substance is who it absorbed. According to Reuters’ source, Aishengna has taken in teams from Yuliangsheng — a lithography startup that began testing a DUV prototype last year, and an affiliate of the Huawei-backed equipment maker SiCarrier — and from Shanghai Micro Electronics Equipment, better known as SMEE, which has been China’s designated national lithography champion for two decades and has spent most of that time falling short. Corporate and recruitment records show Aishengna and Yuliangsheng share a Shanghai address.

Read one way, that is a consolidation story: Beijing gathering scattered lithography efforts into a single, well-capitalised, state-owned vehicle after years of fragmented and largely disappointing results. Read another way, it is an admission that none of the previous efforts worked on their own.

The available evidence supports the first reading more than the second, if only because the programme has now produced hardware that customers are apparently willing to take delivery of. But the honest position is that almost everything about Aishengna’s capability — yield, throughput, uptime, overlay accuracy, cost per wafer — is unknown outside China. Those are precisely the metrics on which ASML built its monopoly, and none of them appear in any public source.

ASML: caught between an export ban and an import substitute

No company embodies the contradiction in Western semiconductor policy more precisely than ASML, and Tuesday made that uncomfortably visible.

ASML shares are up roughly 50% in 2026 and the company has become Europe’s most valuable listed business, with some investors openly speculating about it becoming the continent’s first trillion-dollar firm. That rally is a direct function of the AI buildout: every additional gigawatt of data centre capacity ultimately requires more wafers, and every wafer passes through an ASML machine.

Then, over two sessions, roughly 10% of that value came off — more than €60 billion, according to Reuters’ Toby Sterling — on a report about five machines.

The proportionality question is worth taking seriously, and the sell-side largely thinks the reaction was overdone. ASML shipped 131 immersion DUV systems in 2025. A Chinese programme building five units this year and 20 next year is, in JPMorgan’s framing, immaterial to ASML’s medium-term revenue. “That said, this is another data point in China’s equipment self-sufficiency story, and it raises the long-term risk to ASML’s China revenue,” the bank’s analysts wrote in a note cited by Reuters.

Ipek Ozkardeskaya, an analyst at Swissquote, was blunter, describing the emergence of a genuine Chinese competitor as a potential “nightmare scenario” for ASML if it breaks the company’s hold on the DUV market.

Both views can be right on different horizons, and the market’s job on Tuesday was to decide which horizon to discount. What tipped it was the second half of the squeeze. ASML told investors this month that it expects about 20% of 2026 revenue — roughly €9 billion — to come from China, a figure that already reflects several rounds of export restrictions. The U.S. Congress is currently debating legislation that could block ASML’s remaining immersion DUV exports to China, which is exactly the segment Aishengna is targeting. If that legislation passes and the Chinese tool works, ASML loses the revenue and gains the competitor in the same motion.

Sanne van der Lugt, a researcher affiliated with the Netherlands’ Leiden Asia Centre, made the point that most cleanly captures why this story unsettled investors more than a five-unit production run should. “What this shows is that U.S. export controls successfully created a business case for Chinese lithography,” she told Reuters. “It was not what they were intended to do, but that is the outcome.”

Her second observation is the commercial argument in one sentence. A chipmaker choosing an inferior lithography tool normally accepts lower productivity and a higher cost per chip — a bad trade. Export controls change the arithmetic entirely, because the alternative is not a better machine, it is no machine. Chinese firms, van der Lugt said, “already do not really trust that they can rely on Dutch technology in the long run.” For them, “it may be this or nothing.”

There is precedent for how this ends, and it is not comforting for incumbents. ASML did not inherit the DUV market; it took it from Nikon and Canon over roughly two decades of steady, unglamorous improvement in yield and throughput. Both Japanese firms sold off sharply on Tuesday — Nikon, which still makes immersion DUV systems, fell 9.2%, and Canon fell as much as 6.3%, each posting their largest declines in more than two months. The market appears to have understood the analogy.

The chokepoints that actually matter

Investors reacting to a lithography headline benefit from knowing where lithography sits in the wider structure of the industry, because the semiconductor supply chain has several chokepoints and lithography is only the most famous of them.

Chip manufacturing splits into roughly four layers. Design tools — electronic design automation software — are supplied by a handful of mostly American firms. Chip design itself is dominated by fabless companies such as Nvidia, AMD, Qualcomm and Apple, along with integrated device manufacturers like Samsung, Intel and the memory makers. Manufacturing equipment comes from a small group of specialists: ASML in lithography, Applied Materials and Lam Research in deposition and etch, Tokyo Electron across several process steps, KLA in inspection and metrology, and Advantest and Teradyne in test. Fabrication itself is concentrated in Taiwan, South Korea, Japan, the United States and increasingly China.

Each layer contains its own bottleneck, and lithography’s prominence in the public conversation partly reflects how visually legible it is. An EUV machine costs upward of $200 million, weighs as much as an aircraft and ships in multiple cargo planes. It makes a satisfying story. But a chipmaker deprived of advanced photoresist, or high-purity process chemicals, or precision optics, or inspection tools capable of finding defects at single-digit nanometre scale, is just as stuck as one deprived of a scanner.

This matters for interpreting Tuesday’s move because the market treated a lithography development as though it resolved the entire self-sufficiency question. It does not. Several observations follow directly from the reporting itself.

Reuters’ source noted that while most components in Aishengna’s system are domestic, some critical parts still come from Japan. That single clause carries substantial weight. Lithography optics in particular are among the most demanding precision-manufactured objects in existence — ASML’s are supplied by Carl Zeiss SMT under a decades-old partnership that is itself a chokepoint. A Chinese system that depends on Japanese components for its most sensitive subsystems has substituted one foreign dependency for another, in a narrower and arguably more fragile form.

Second, immersion DUV is a middle-tier capability. It is genuinely sufficient for DRAM, for mature logic, for analogue, power and automotive chips — an enormous and profitable share of global semiconductor consumption. It is not sufficient, without heroic multiple-patterning schemes that destroy yield economics, for the leading-edge logic that trains large models. EUV remains the gate, and Reuters reported in December that China had produced an EUV prototype but remained years from production. Nothing announced this week changes that assessment.

Third, the metric that determines commercial viability is not resolution but cost per good die. That is a function of throughput, uptime, overlay accuracy and yield, compounded over years of production. A machine that prints the right feature size at half the wafer-per-hour rate of the incumbent, with worse uptime, produces chips at a materially higher cost. In an open market that is fatal. In a market shaped by export controls, it may be survivable — and that, rather than any technical parity, is the reason the report mattered.

Eight years of export controls, and the response they produced

The current situation did not arrive suddenly. It is the accumulated result of a policy that has been tightening, in stages, since the late 2010s.

The first meaningful restriction concerned EUV. The Dutch government declined to renew ASML’s export licence for EUV systems to China at the end of the 2010s, following sustained U.S. pressure. That decision alone froze Chinese access to the single tool required for leading-edge logic and set the strategic problem Beijing has been working on ever since.

Restrictions broadened substantially in October 2022, when the U.S. Commerce Department introduced sweeping controls covering advanced computing chips, semiconductor manufacturing equipment and, notably, the ability of U.S. persons to support Chinese advanced-node fabs. Because equipment requires continuous servicing by specialist engineers, the personnel provision was in some respects more disruptive than the hardware restrictions.

The Netherlands and Japan aligned their own controls in 2023, which brought ASML’s more advanced immersion DUV systems into scope alongside key tools from Tokyo Electron and other Japanese suppliers. Subsequent rounds extended coverage to additional equipment categories, high-bandwidth memory and a growing list of restricted entities.

China’s response ran on two tracks. The first was stockpiling: in the windows before each round of restrictions took effect, Chinese buyers purchased equipment at extraordinary volumes, which is why ASML’s China revenue share swung so violently over several years and why the company’s guidance for roughly 20% of 2026 revenue from China reads as a normalisation rather than a collapse. The second track was substitution, funded through successive iterations of the national semiconductor investment fund and channelled into a widening set of domestic equipment firms — of which SiCarrier, the Huawei-linked group whose affiliate Yuliangsheng has now been folded into Aishengna, is the most prominent.

The strategic critique of the policy has been consistent from the start and has little to do with whether the controls were justified. It is that a restriction which is severe enough to hurt but not severe enough to be permanent creates precisely the demand signal a domestic industry needs. Chinese chipmakers who might once have bought ASML tools indefinitely, because they were better and cheaper, now face a supply relationship that a foreign legislature can revoke. Under those conditions, a worse domestic tool has a strategic value that its specifications do not capture.

That is the argument Sanne van der Lugt of the Leiden Asia Centre made to Reuters, and it is why her formulation — that the controls “successfully created a business case for Chinese lithography” — travelled so widely this week. It is not a claim that the controls failed at their immediate objective. China’s leading-edge logic capability is demonstrably behind where it would otherwise be. It is a claim about second-order effects on a ten-to-twenty year horizon, which is exactly the horizon equity valuations attempt to discount.

The pending U.S. legislation to restrict ASML’s remaining immersion DUV exports sits directly on this fault line. Its supporters argue it closes a gap that has allowed China to build out mature-node capacity at scale. Its critics argue it would eliminate the last commercial reason for Chinese fabs to buy Western tools at all, converting a partial dependency into none. Both readings are internally coherent. The market, on July 28, appeared to weight the second.

CXMT’s debut: $8.6 billion raised, $480 billion created, and a very large question

The lithography story would have been enough on its own. It arrived alongside a listing that changed the shape of China’s equity market in a single session.

ChangXin Memory Technologies priced its Shanghai STAR Market initial public offering at 8.66 yuan a share and sold 6.688 billion shares, raising 57.92 billion yuan — about $8.6 billion. That made it the largest listing on mainland Chinese exchanges since Agricultural Bank of China went public in 2010, the largest STAR Market offering ever, and Asia’s biggest IPO of 2026. Institutional demand for the allocated tranche exceeded 500 times the shares on offer.

The stock opened Monday at 49.50 yuan. By the midday break it was at 54.65 yuan, up 531% from the offer price, valuing the company at 3.66 trillion yuan. It closed the session up roughly 466%, with a market capitalisation in the region of 3.3 trillion yuan — about $480 billion at prevailing rates. That was enough to displace Industrial and Commercial Bank of China as the most heavily weighted company on the mainland market and to put CXMT above Intel in market value.

The operating business underneath that valuation is real and growing extremely fast. CXMT is the world’s fourth-largest DRAM manufacturer. It reported 50.8 billion yuan of revenue in the first quarter of 2026, up more than 700% year over year, as the global memory shortage collided with China’s domestic AI server buildout. Of the IPO proceeds, 29.5 billion yuan is earmarked for production-line upgrades, DRAM technology development and forward-looking DRAM research.

Two things about that allocation deserve attention, because they cut against the panic.

First, the funded projects are DRAM projects. They are not a standalone high-bandwidth memory programme. CXMT has been developing HBM-related capability, but commercial HBM production requires advances in stacking, advanced packaging, testing and — critically — customer qualification, and none of those are solved by capital alone. HBM is where Samsung and SK hynix earn their margins; conventional DRAM is where they are most exposed to a low-cost entrant.

Second, CXMT’s shares are not yet eligible for trading through the Shanghai–Hong Kong Stock Connect, which means foreign investors cannot freely buy or sell them. Han Ji-young, an analyst at Kiwoom Securities, made this point directly in the Korea JoongAng Daily’s coverage: while the blockbuster IPO “may weigh on sentiment by heightening concerns over intensifying competition from Chinese memory chipmakers, the scope for actual capital outflows from Korean semiconductor stocks appears limited.”

In other words, this was not a rotation. Foreign investors did not sell Samsung to buy CXMT, because for the most part they cannot buy CXMT. What they did was reassess what a well-capitalised Chinese DRAM champion — one that may soon be buying domestically made lithography tools — does to the terminal value of the memory business.

CXMT shares fell 3% on Tuesday, incidentally, while the companies it is supposedly displacing fell four to five times as much.

Sources: TechNode’s report on the CXMT listing and Korea Economic Daily coverage of the July 28 Asian chip selloff.

Anatomy of a Korean crash

To understand why Seoul fell twice as hard as Tokyo and more than twice as hard as Taipei, it helps to understand what the Kospi had become by June.

The index started 2026 in the mid-4,000s and closed at a record 9,114.55 on June 22 — a gain of roughly 100% in under six months, and a tripling over twelve. Goldman Sachs had pegged 2026 earnings growth for the Korean market at around 300%, the strongest annual profit expansion in any Asian market since the recovery from the 1997–98 financial crisis. That was not a bubble in the sense of profits that did not exist. The profits existed. Samsung guided to a second-quarter operating profit of 89.4 trillion won, up 1,810% from a year earlier. SK hynix was expected by a consensus of 14 domestic brokerages to report 64.1 trillion won of quarterly operating profit — more than its entire 2025 total of 47.2 trillion won — on 84.1 trillion won of revenue, at an operating margin somewhere between 74% and 77%.

Those are not normal semiconductor margins. They are the margins of a company selling a commodity into a shortage it did not create and cannot immediately fix.

The problem with a market that rises 100% in six months on two stocks is that the positioning becomes as important as the fundamentals. By late June, the Kospi rally had drawn in a very large volume of borrowed money. Investor deposits at Korean brokerages — the cash retail investors keep on the sidelines — stood at 109.17 trillion won as of July 27, down more than 27 trillion won from their June 23 peak. Outstanding margin loans peaked at 38.63 trillion won on June 24 and had fallen to 32.74 trillion won, according to Korea Financial Investment Association data cited by the Korea JoongAng Daily.

Margin balances do not fall by 5.9 trillion won because investors calmly reduce risk. They fall because positions are closed, voluntarily or otherwise.

Layered on top of that was a product innovation that Korean regulators have since come to regret. In May 2026, Korean exchanges approved single-stock leveraged exchange-traded funds tracking Samsung Electronics and SK hynix at two times daily exposure. The mechanics of such a fund are unforgiving in a downtrend: when the underlying stock falls, the fund must sell to maintain its leverage ratio, regardless of price. That selling pressures the stock, which pressures margin accounts, which forces more selling. President Lee Jae-myung has publicly identified these products as a source of market instability and ordered financial authorities to accelerate safeguards; the Financial Services Commission subsequently convened Samsung, Mirae Asset, KB and other major asset managers to work through the measures.

The volatility statistics tell the story without editorialising. Through July 24, the Kospi market had triggered 41 sidecars in 2026 — more than the 26 recorded during the whole of the 2008 global financial crisis, according to the Seoul Economic Daily. Tuesday’s marketwide circuit breaker was the eighth of the year. The mechanism has existed since 2000; more than half of all activations in its history have occurred in the past seven months.

Kim Seok-hwan, an analyst at Mirae Asset Securities, gave the most useful characterisation of the session. “A new shock didn’t bring the market down. It was familiar risks finally being priced in as crowded positioning met thin liquidity,” he told the Korea JoongAng Daily. “The key question now isn’t whether the index will fall further, but whether the latest plunge marks the beginning of a deterioration in fundamentals or the final stage of an unwinding of excessive positioning.”

That is the right question, and nobody can answer it from the tape alone.

Sources: Korea JoongAng Daily for Seoul closes; Korea Economic Daily for Tokyo and Taipei closes; The Motley Fool’s July 28 midday market report for U.S. afternoon levels. U.S. prices are intraday as of approximately 12:45 p.m. Eastern Time and are not closing prices.

The memory supercycle that made the fall this far

None of this makes sense without the price data underneath it.

Through the first half of 2026, memory chips went through the sharpest price appreciation in the industry’s history. According to TrendForce, conventional DRAM contract prices rose roughly 93% to 98% quarter over quarter in the first quarter of 2026 — the largest quarterly increase ever recorded, with PC DRAM specifically rising more than 100%. The second quarter brought another 58% to 63% quarterly increase. Prices did not merely recover from a downturn; they roughly quadrupled from where they had been.

The mechanism was straightforward and, for once, not speculative. AI training and inference workloads consume enormous quantities of high-bandwidth memory, and HBM consumes far more wafer capacity per gigabit than conventional DRAM because of its stacked construction and lower yields. As Samsung, SK hynix and Micron reallocated fab capacity toward HBM and server DRAM, they starved the PC, smartphone and automotive markets of supply. A shortage in one product line was manufactured by a boom in another.

That is what produced operating margins in the mid-seventies at SK hynix, a number that would be implausible in almost any other manufacturing business. It is also, by construction, temporary. Margins that high are an invitation to add capacity, and capacity is being added — by the incumbents, and now by a Chinese competitor with $8.6 billion of fresh equity and a domestic tool supply chain taking shape behind it.

The cycle had, in fact, already begun to turn before Tuesday. TrendForce’s July survey projected that server DRAM contract prices would rise 13% to 18% quarter over quarter in the third quarter — still a substantial increase, but a marked deceleration from the first half. The firm attributed the slowdown partly to long-term agreements capping increases and partly to customer price tolerance weakening once supplier operating margins pushed above 80%. Buyers, in other words, had started to push back.

Source: TrendForce, “Long-Term Agreements Cap Price Increases; Server DRAM Contract Prices Expected to Rise 13–18% QoQ in 3Q26”.

A stock priced for 60% quarterly price increases behaves very differently when the number becomes 15%. Add a credible new supplier and the question stops being about the slope of the curve and starts being about the shape of the whole cycle.

The two companies that are, effectively, the Korean stock market

Index concentration is usually a footnote. In Korea in 2026 it is the whole story, and it explains why a report about a Shanghai machine shop translated into a 732-point move in Seoul.

Samsung Electronics

Samsung is not a memory company, which is part of what makes its 2026 results so striking. It runs a mobile handset business, a consumer electronics business, a display business and a contract chip foundry alongside its memory operation. In most years those divisions smooth each other out; memory is violently cyclical, handsets are seasonal, displays are competitive and low-margin.

In 2026 the smoothing stopped working, because memory overwhelmed everything else. Guided second-quarter operating profit of 89.4 trillion won, up 1,810% year on year, is not a company that has improved its handset margins. It is a memory business earning shortage economics on an enormous installed capacity base.

That has two consequences investors should hold simultaneously. Samsung’s diversification means it is less exposed to a memory downturn than a pure-play competitor — it has other businesses to fall back on, and a foundry operation that benefits from broader semiconductor demand. But the same diversification means the market is currently valuing Samsung almost entirely on the division whose pricing power is most likely to normalise. A 13.39% single-day decline in a company of Samsung’s scale is a statement about which part of the business investors were actually paying for.

SK hynix

SK hynix is the purer expression of the trade, and its 14.65% fall reflects that. The company has built the strongest position in high-bandwidth memory, the stacked DRAM that sits beside AI accelerators, and HBM is the highest-value, most technically demanding memory product in production. That leadership is what produced consensus expectations of a 74% to 77% operating margin for the second quarter — a figure with almost no precedent in commodity manufacturing.

It also produces two forms of concentration risk that deserve explicit statement. The first is customer concentration: HBM demand is driven by a small number of accelerator makers, with Nvidia by far the largest. The July 24 agreement formalising a long-term memory partnership between the two companies is commercially valuable and simultaneously deepens that dependence. The second is product concentration: allocating fab capacity to HBM is what created the conventional DRAM shortage that lifted prices across the board. If HBM demand moderates, that capacity returns to the conventional market, and the shortage unwinds from both directions at once.

SK Square, SK hynix’s holding company, fell 15.6% on Tuesday — more than the operating company itself, which is the customary behaviour of a holding structure when the market marks down its principal asset and widens the discount at the same time.

Micron and the American memory complex

The third global DRAM producer sits in Boise, Idaho, and it fell about 8.9% in Tuesday’s U.S. session. Micron is the only large memory manufacturer headquartered outside East Asia, which has made it a beneficiary of supply-chain diversification policy and a direct competitor in exactly the segments a Chinese entrant would target.

The wider American memory and storage complex fell harder than the accelerator names. SanDisk dropped about 14.6%, Seagate and Western Digital each around 8% to 9% by various accounts, while Nvidia itself finished the afternoon slightly higher. That dispersion is informative. The market did not sell “AI.” It sold the parts of the AI supply chain where a low-cost Asian entrant, a decelerating price cycle and commodity economics intersect.

SanDisk’s position illustrates how far these stocks had run. Down roughly 50% from its June peak, it remained the best-performing stock in the S&P 500 for 2026 with gains above 360%. A halving that leaves a stock quadrupled for the year is not a crash in any economically meaningful sense. It is the unwinding of an extraordinary move.

Circular financing: what Nvidia has actually agreed to, and what it has not

The second pillar of Tuesday’s selloff had nothing to do with China.

On Monday, The Wall Street Journal reported that Nvidia was in talks to guarantee roughly $250 billion of financing to allow OpenAI to lease computing capacity from a U.S. data centre project. Bloomberg reported separately that Nvidia is working on a broader package of AI deals worth more than $750 billion in total, and that the company had already announced more than $540 billion of such arrangements in 2026 before the OpenAI discussions.

Some of that total is already public and documented. On July 24, Nvidia and SK Group announced what the companies described as a “$500-billion-plus comprehensive partnership,” formalised through letters of intent. Under it, SK Telecom would build a two-gigawatt AI factory using Nvidia’s DSX platform and Vera Rubin accelerated computing, powered by SK hynix HBM4, with the first facility planned to come online in 2027. SK hynix separately entered a long-term AI memory partnership under which Nvidia secures supply of next-generation memory and the two co-develop future HBM.

“South Korea has all the ingredients to become a global AI powerhouse,” Nvidia founder and chief executive Jensen Huang said in the announcement. SK Group chairman Chey Tae-won framed the deal as helping Korea “transcend its role as a leading adopter of AI and become a global hub that drives AI innovation.”

It is worth being precise about what a letter of intent is. It is not a purchase order, a signed contract or recognised revenue. It is a statement of intent to negotiate definitive agreements. Headline numbers attached to letters of intent should be read as ceilings on ambition, not as bookings.

The concern that has attached itself to these arrangements is what market participants call circular financing, or less charitably, round-tripping. In the structure investors worry about, a supplier takes equity stakes in, or provides debt guarantees to, its own customers; the customers use that capital to buy the supplier’s products; and the supplier books the resulting sales as demand. Revenue is real in an accounting sense but the underlying end-demand may be weaker than the reported numbers imply, because the supplier funded it.

Nvidia has publicly rejected the characterisation that it uses circular financing schemes. Prominent short sellers have publicly disagreed. Both the International Monetary Fund and the Bank for International Settlements have flagged AI-related circular financing as a systemic downside risk in their published assessments — which is a meaningful signal, given how rarely either institution names a specific financing pattern in a specific industry.

What is not in dispute is that the credit market repriced Nvidia’s risk on Monday. Bloomberg reported that the cost of protecting Nvidia’s debt against default for five years rose by as much as roughly 0.14 percentage point, to as high as about 0.82 percentage point a year — the biggest single-day increase since the swaps began actively trading in November. In plain terms, an investor who wanted five-year default protection on Nvidia paid roughly $82,000 a year per $10 million of exposure, against roughly $68,000 implied by the level before the move.

Nvidia’s credit spreads remain modest in absolute terms. A company with its cash generation is not a credit risk in any conventional sense. But CDS markets price the direction of obligations as much as the level, and $750 billion of prospective commitments — guarantees, backstops, equity investments, offtake arrangements — is a different balance-sheet profile than the one Nvidia had eighteen months ago. The swaps market noticed before the equity market did.

The capital-spending arithmetic behind the anxiety

Neither the lithography report nor the Nvidia financing story would have produced a 10% move in Seoul if investors were confident about the demand side. They are not, and the reason is a set of numbers that grew faster than anyone’s revenue forecasts.

On July 22, Alphabet raised its 2026 capital expenditure guidance to a range of $195 billion to $205 billion, up from $180 billion to $190 billion. Google Cloud revenue in the second quarter reached $24.8 billion, up 82% year over year, and management said demand for AI infrastructure continued to exceed available compute. Alphabet shares nonetheless fell about 5% after the announcement, which tells you how the market is currently weighing revenue growth against the cash required to produce it.

Alphabet is not an outlier. Microsoft has set calendar-2026 capital spending at roughly $190 billion. Amazon has guided to approximately $200 billion companywide. Estimates for combined 2026 AI-related capital expenditure across Alphabet, Microsoft, Amazon and Meta cluster between $600 billion and $725 billion.

Those figures do two things simultaneously. They guarantee near-term demand for memory, accelerators and lithography — which is the bull case for every stock that fell on Tuesday. And they create an enormous depreciation burden that will run through income statements for the next five to seven years, which is the bear case for the customers doing the spending.

The tension resolves one of two ways. Either AI revenue scales fast enough to absorb the depreciation, in which case current capital spending is understated and chip demand runs hotter for longer. Or it does not, in which case the capital expenditure line gets cut, and the memory and accelerator businesses discover that a very large share of their record backlog was funded by companies that no longer want it.

Investors have no way to settle that argument in July 2026. What they can do is reduce exposure to the assets most levered to the outcome, which is what happened in Seoul.

Wall Street’s split tape

The most revealing thing about Tuesday in New York was not how much technology fell. It was what rose.

In afternoon trading, the Dow Jones Industrial Average was up about 1.3%, gaining roughly 520 points at one stage and trading near 52,870. The S&P 500 was up about 0.35% at around 7,439. The Nasdaq Composite was flat at approximately 24,931, having been down 1.3% in the morning session. The Invesco S&P 500 Equal Weight ETF was up about 1.1%, comfortably outperforming the capitalisation-weighted index — a clean signal that the drag was concentrated in the largest technology names rather than spread across the market.

Sherwin-Williams rose about 8.9% after beating earnings expectations. Caterpillar, which has become a de facto AI infrastructure play through its exposure to data centre construction, fell about 4.2% and was the Dow’s largest single drag. Paint outperformed silicon by twelve percentage points.

The Nasdaq-100 was trading roughly 10% below its record high, putting it at the threshold of a technical correction. The Philadelphia Semiconductor Index fell for a fourth consecutive session, its longest losing streak of 2026.

Individual moves inside the chip complex were severe. Micron fell about 8.9% to $820.13. SanDisk fell about 14.6% to $1,092.01 — a stock now roughly 50% below its June peak while still, remarkably, the best-performing member of the S&P 500 for the year to date with gains above 360%. AMD fell about 7.2%. Intel fell about 4.9%. Nvidia itself, having taken the brunt of Monday’s damage, edged higher by about 0.7% to $197.90.

Meanwhile Microsoft rose about 2.3% and Alphabet about 2.5% ahead of their results. That divergence — hyperscalers up, their suppliers sharply down — is not a market losing faith in AI. It is a market rethinking who captures the value and who absorbs the capital intensity. Whether that distinction survives contact with Wednesday’s capital expenditure guidance is a separate question.

Japan’s quiet, concentrated exposure

Tokyo’s decline was smaller than Seoul’s in index terms — the Nikkei 225 fell about 4% to a two-year low, closing below 62,000 — but the damage inside the chip complex was in some cases worse.

Kioxia, the NAND flash maker spun out of Toshiba, fell 18.33% to ¥44,550, the largest single-stock decline of the session in any major market. Tokyo Electron dropped 10.96%, Advantest 10.11%, Lasertec 14.4% and Renesas 10.6%.

Each of those falls has a distinct logic. Kioxia is a pure NAND play, and NAND economics are even more commoditised than DRAM. Tokyo Electron sells process equipment into exactly the Chinese fabs that a domestic tool ecosystem would eventually serve, and has already lived through Japanese export restrictions on its most advanced products. Advantest makes the test equipment that HBM requires in disproportionate quantities, which makes it a high-beta expression of the HBM cycle. Lasertec makes mask inspection systems for EUV lithography — the most direct listed proxy for the assumption that EUV remains the industry’s uncontested frontier.

The two most interesting reactions were the smallest. Nikon fell 9.2% and Canon as much as 6.3%, each their steepest fall in more than two months. Nikon still manufactures immersion DUV systems; Canon supplies more mature lithography equipment and has been developing nanoimprint technology as an alternative path.

On a superficial reading, a new competitor in immersion DUV is bad news for Nikon, which is why the shares fell. On a longer reading, the two Japanese firms are the living evidence for both sides of the argument. They are what happens to lithography incumbents who lose a technology transition — and they are also proof that a determined national champion with a government behind it can spend two decades and still not close the gap. Japanese optics were, and remain, world-class. It was not enough.

The macro backdrop nobody could ignore

All of this took place during one of the more genuinely uncertain Federal Reserve weeks in recent memory.

The Federal Open Market Committee began a two-day meeting on Tuesday, July 28, with a decision due Wednesday, July 29. The federal funds target range stands at 3.50% to 3.75%, unchanged since the last cut in December 2025. A hold would be the fifth consecutive meeting without a change.

What made the meeting unusual is the direction of the risk. Bill Adams, chief U.S. economist at Fifth Third Commercial Bank, expected the Committee to hold, and noted that markets would focus on its assessment of core inflation “since the Fed ended forward guidance after Kevin Warsh became chair.” Adams described a genuinely mixed inflation picture: encouraging signs from home prices and rents and the fading effects of the 2025 tariff increases, set against “rebounding energy prices as disruptions to Middle East and Russian exports resurface; new tariffs; AI-related pressure on electronics prices; and labor supply bottlenecks pushing up prices of services like home health care and nursing care.”

That third item deserves emphasis, because it closes a loop back to the main story. Memory prices roughly doubling in a quarter is not only a corporate earnings event. It is an input cost shock that flows into consumer electronics prices, and it is now showing up in central bank commentary.

By Tuesday morning, markets were pricing roughly a 38% probability that the Fed would raise rates by 25 basis points on July 29, up from about 16% two weeks earlier, according to Korea Economic Daily’s account of the pricing. A basis point is one hundredth of a percentage point, so a 25-basis-point move would lift the target range to 3.75%–4.00%. That is not a base case, but it is a materially different distribution than the market carried into July, and it contributed to the risk reduction across Asian equities.

Energy moved the other way. Oil retreated sharply after the United States paused military strikes on Iran and Tehran suspended retaliatory operations, opening space for diplomacy. Iranian foreign minister Seyyed Abbas Araqchi held separate calls with his Saudi and Omani counterparts on navigation through the Strait of Hormuz. Brent crude fell about 3.9% to $84.91 a barrel and West Texas Intermediate about 3.3% to $79.87, according to CNBC — a substantial retreat from the roughly $100 level reached the previous week. Supply conditions also improved as crude loadings resumed at the Caspian Pipeline Consortium terminal on Russia’s Black Sea coast.

Currency and rates markets were comparatively calm. The dollar index rose 0.08% to 101.621. The yen traded at 163.78 per dollar, hovering just above its weakest level in four decades, with the Bank of Japan widely expected to hold rates this week. The 10-year U.S. Treasury yield fell about four basis points to 4.64% on Monday as the Middle East risk premium unwound. Gold futures slipped about 1.2% to around $4,087 an ounce and silver fell about 2.2% to $57.45.

President Donald Trump, speaking aboard Air Force One on Monday, characterised the Iran talks in transactional terms: “There’s a good chance that something could happen, and if it does, good. If it doesn’t, we go back to doing what we were doing.” Markets took the first half of that sentence and largely ignored the second.

Timeline: how the week unfolded

1. Wednesday, July 22: Alphabet raises 2026 capital expenditure guidance to $195–205 billion from $180–190 billion; shares fall about 5% despite Google Cloud revenue rising 82% to $24.8 billion.
2. Friday, July 24: Nvidia and SK Group announce a $500-billion-plus partnership covering a two-gigawatt SK Telecom AI factory and a long-term SK hynix memory agreement, formalised through letters of intent. Separately, the peso hits a record low of 61.847 to the dollar. The United States pauses military strikes on Iran.
3. Monday, July 27: CXMT debuts on the Shanghai STAR Market, closing up roughly 466%. The Information reports China has begun manufacturing immersion DUV lithography tools. The Wall Street Journal reports Nvidia is in talks to guarantee about $250 billion of OpenAI data centre financing. Nvidia closes down 4.99% at $196.51; ASML’s Amsterdam shares fall more than 8%; Nvidia’s five-year CDS post their largest one-day widening on record. Philippine President Ferdinand Marcos Jr. delivers his State of the Nation Address, calling for new tax measures.
4. Tuesday, July 28, morning in Asia: The Kospi triggers a sell-side sidecar, then a marketwide circuit breaker at about 10:13 a.m. Seoul time after falling 8.02%; trading halts for 20 minutes. The index closes down 10.84% at 6,023.66. Japan’s Nikkei 225 falls about 4% to a two-year low; Taiwan’s Taiex loses 4.65%.
5. Tuesday, July 28, 4:19 a.m. Eastern: Reuters identifies Shanghai Aishengna Electronic Technology Group as the company behind China’s immersion DUV programme.
6. Tuesday, July 28, Manila: BSP Governor Eli Remolona says there is only a “small chance” of a 50-basis-point rate increase. BDO Unibank settles a record 132 billion peso ASEAN Sustainability Bond issue. The PSEi closes up 0.54% at 6,314.90 as oil retreats.
7. Tuesday, July 28, New York: The Dow rises about 1.3% while the Nasdaq Composite finishes roughly flat; the Nasdaq-100 trades about 10% below its record. The FOMC begins its two-day meeting.
8. Wednesday, July 29 (scheduled): SK hynix reports second-quarter results. The Federal Reserve announces its policy decision. Microsoft and Meta Platforms report earnings.
9. Thursday, July 30 (scheduled): Apple and Amazon report earnings.

The other half of the Asia trade: why Manila went up

Regional coverage of Tuesday’s session tended to compress “Asia” into a single falling line. The reality was more differentiated, and the Philippine market is a useful illustration of why.

The Philippine Stock Exchange index closed at 6,314.90 on July 28, up 33.89 points or 0.54%, returning above the 6,300 level. It did so on the same day the Kospi fell 10.84%. The reason is structural: the PSEi has almost no semiconductor manufacturing weight and considerable exposure to conglomerates, banks, property and utilities, all of which benefit when energy costs fall. Retreating oil was worth more to Manila than a Chinese lithography breakthrough cost it.

That divergence sits alongside a set of domestic developments that matter more to Philippine investors than anything happening in Shanghai.

The central bank is leaning hawkish, but not aggressively

Speaking to reporters on the sidelines of a Bangko Sentral ng Pilipinas event on Tuesday, Governor Eli M. Remolona Jr. addressed whether the central bank might move more aggressively on rates — specifically, whether a 50-basis-point increase was likely. His answer, in Filipino, was “May chance naman pero baka maliit na chance”: there is a chance, but it is probably a small one.

The framing is worth noting because it inverts the direction most emerging-market central banks were moving in a year ago. The BSP is discussing the size of an increase, not whether to cut.

Remolona identified two live inflation pressures. The first is fiscal. President Marcos used his July 27 State of the Nation Address to call on Congress to pass several tax measures, including a higher threshold for income tax exemptions for low- and middle-income earners and tax relief for micro, small and medium enterprises. Asked about the inflation consequences, Remolona said the BSP was still estimating them but expected an effect — “the major impact will be in 2027, with a smaller one in 2028.”

The second is the currency. The peso fell to a record low of 61.847 to the dollar on Friday, July 24, down 9.7 centavos and breaking the previous record of 61.75 set the day before, as the Middle East conflict pushed oil higher. A weaker peso raises the local cost of imported energy and goods.

Remolona’s response to the currency question was more interesting than the headline. He called the record low “a misleading number” on the grounds that it isolates one bilateral rate while the dollar was strengthening broadly. And he defended minimal intervention with a candid piece of central banking arithmetic: “When it’s a strong dollar, we limit intervention to just maintain orderly markets. Because if we intervene against a strong dollar, we’re just helping the rest of the world get their dollars.” Spending reserves to defend a level the dollar itself is setting, in his framing, achieves nothing except depleting reserves.

He also reaffirmed the BSP’s projection that the Philippine economy will recover in the second half of the year. Readers should treat that as an official forecast rather than an established outcome; the BSP is both the forecaster and the institution whose policy determines part of the result.

BDO’s record bond sale

BDO Unibank settled its sixth peso-denominated ASEAN Sustainability Bond issue on July 28, raising 132 billion pesos — a record for the bank. The offer was launched with a base size of 5 billion pesos and closed early, on July 10, after being more than 26 times oversubscribed. The bonds carry a tenor of one and a half years and a coupon of 6.26% per annum. Proceeds are intended to finance and refinance eligible assets under the bank’s Sustainable Finance Framework, support lending and diversify funding sources.

Two features stand out. The first is the sheer scale of the oversubscription, which says something about domestic peso liquidity and the appetite of Philippine retail and institutional investors for short-dated bank paper at a 6.26% coupon in an environment where the central bank is contemplating further increases. The second is the tenor: eighteen months is short for a sustainability bond, and short-dated funding is what a bank raises when it expects rates to move and does not want to lock in long duration.

Petron’s Bohol fuel farm

Petron Corp. was selected by Aboitiz InfraCapital Bohol Airport Corp. as top-ranked partner, following a competitive request-for-proposals process, to design, build, operate and maintain a 340 million peso Jet A-1 fuel farm at Bohol-Panglao International Airport. The facility will comprise three 477-kilolitre tanks and is scheduled for completion in the third quarter of 2027. Once operational, it will provide at least five days of on-site fuel storage, replacing the airport’s existing arrangement. Commissioning is to comply with current Joint Inspection Group standards for aviation fuel safety and quality.

The airport currently consumes roughly 102 kilolitres of aviation fuel a day — about 43 kilolitres for international flights and 59 for domestic operations. At that run rate, five days of storage is approximately 510 kilolitres against 1,431 kilolitres of tank capacity, giving meaningful headroom for traffic growth.

In absolute terms this is a small contract for a company of Petron’s size, and it will not move the stock. Its significance is directional: it is an infrastructure award tied to Philippine tourism throughput and to the privatised airport operating model that Aboitiz InfraCapital has been building out. Those are slower, less dramatic stories than a 10% index decline, and over a decade they are likely to matter more to Philippine investors.

This was not a one-day event

Framing July 28 as a sudden shock obscures a drawdown that had been running for five weeks and had already produced most of its damage before the lithography story appeared.

The Kospi peaked at 9,114.55 on June 22. By July 16 it had fallen 6.37% in a single session, already more than a quarter below the high. On July 22, Alphabet’s raised capital expenditure guidance sent its shares down about 5% and put AI spending discipline back at the centre of the conversation. On July 23, the Nasdaq fell more than 2% as technology capital expenditure fears combined with a Middle East escalation that lifted oil toward $100. On July 24, both the Nasdaq and the Russell 2000 closed lower on the same theme. The Philadelphia Semiconductor Index entered Tuesday already in its longest losing streak of the year.

By the time the China reports landed, the Kospi had round-tripped roughly 34% from its record in 25 trading sessions. Margin balances had fallen 5.89 trillion won from their June 24 peak. Retail cash reserves had fallen more than 27 trillion won. President Lee Jae-myung had already publicly identified single-stock leveraged ETFs as a source of instability and instructed regulators to act; the Financial Services Commission had already convened the major asset managers.

In other words, the deleveraging was well advanced. What the lithography and financing stories supplied was not the cause of the decline but a reason for the last, fastest leg of it — the point at which a positioning unwind acquired a narrative and accelerated.

That distinction has practical consequences. If Tuesday was primarily a fundamental repricing, the appropriate response is to lower long-term earnings assumptions for the memory complex. If it was primarily the final phase of a leverage unwind wearing a fundamental costume, the appropriate response is to watch margin balances and fund flows rather than headlines. The evidence currently supports a mixture, weighted toward the second — which is precisely why Kiwoom’s Han Ji-young could describe valuations as reaching trough levels while volatility peaked, and why Mirae Asset’s Kim Seok-hwan framed the open question as whether this marks deterioration in fundamentals or the end of an unwind.

What a 34% index decline does to Korea beyond the market

Equity drawdowns in Korea have unusually direct social and political transmission, because retail participation is exceptionally high and, in 2026, exceptionally leveraged.

Tuesday’s flow data captures the dynamic. Foreign investors sold a net 4.97 trillion won. Retail investors bought a net 4.33 trillion won — absorbing roughly 87% of the foreign selling. Domestic institutions took another 630.1 billion won. A market that falls 10.84% while households buy nearly $3 billion of it is transferring risk from professional to individual balance sheets in real time.

The leverage attached to that participation is what turned a correction into repeated circuit breakers. Two-times single-stock leveraged ETFs on Samsung and SK hynix, approved in May 2026 and heavily used by retail investors, sell mechanically into declines to maintain their exposure ratio. Margin accounts liquidate when collateral values fall. Each process feeds the other, which is why the Korea Exchange has now triggered eight marketwide halts in seven months and 41 sidecars against 26 in all of 2008.

The policy response has so far been targeted rather than sweeping. President Lee ordered financial authorities to prepare additional measures on single-stock leveraged ETFs; the Financial Services Commission convened Samsung Asset Management, Mirae Asset and KB Asset Management among others to accelerate existing plans. That is a product-level intervention aimed at a specific amplification mechanism, not a market stabilisation programme. Whether it proves sufficient depends largely on whether the memory cycle stabilises.

There is also a broader macroeconomic dimension that will take longer to show up. Korea’s export performance, corporate tax receipts, capital expenditure plans and currency have all been shaped this year by two companies earning shortage margins. Samsung’s guided quarterly operating profit of 89.4 trillion won and SK hynix’s expected 64.1 trillion won are national-accounts-scale numbers. A normalisation in memory pricing that the market is now anticipating would flow through Korean growth, fiscal revenue and investment with a lag of several quarters. That is not visible in a single session’s price action, and it is the slower risk worth monitoring.

The case that the chip stock selloff was an overreaction

The strongest argument for the bulls rests on three specific numbers rather than on sentiment.

Five machines against 131. ASML shipped 131 immersion DUV systems in 2025. China’s programme, as reported, aims for roughly five units in 2026 and 20 in 2027. Even if every one of those tools works perfectly and displaces an ASML sale that would otherwise have happened — which is not the case, because ASML is already barred from selling its best immersion tools into China — the volume impact through 2027 is marginal. JPMorgan’s assessment that medium-term damage is limited is the mainstream sell-side view and it is difficult to argue with on arithmetic alone.

Capability is not competitiveness. The Reuters source explicitly described Aishengna’s machine as requiring further testing and remaining far from ASML’s equivalent models. Building a lithography system that produces an image is a different problem from building one that produces the same image, on the same overlay budget, 200 times an hour, for years, with predictable service intervals. ASML spent two decades winning that argument against Nikon and Canon, both of which were formidable optics companies with far more experience than any Chinese entrant. There is no reason to assume the learning curve is shorter this time simply because the state is paying for it.

CXMT cannot yet compete where the money is. The IPO proceeds are directed at DRAM production lines and DRAM research, not at a standalone HBM programme. High-bandwidth memory requires through-silicon vias, precision stacking, advanced packaging, extensive testing and — the hardest part — qualification by customers whose accelerators will fail if the memory does. SK hynix’s margins come from HBM. Conventional DRAM is a commodity where Chinese entry compresses prices, but it is not where the incumbents’ profit pool is concentrated.

To those, add a fourth argument about the mechanics of Tuesday itself. Kim Seok-hwan’s observation that “a new shock didn’t bring the market down” points at forced selling rather than fundamental reassessment. Leveraged ETFs rebalancing, margin accounts liquidating and foreign investors reducing a crowded position all produce price movement that has no informational content about the underlying businesses. Han Ji-young of Kiwoom Securities argued that Korean valuations appear to have reached trough levels while price and fund-flow volatility appear to be peaking — a technical, not fundamental, read.

Finally, the earnings are still coming in at records. Samsung guided to 89.4 trillion won of second-quarter operating profit. SK hynix was expected to report 64.1 trillion won. Those results describe the quarter that just ended, and they were not revised on Tuesday.

The case that the selloff was rational

The bear argument does not dispute any of that. It disputes what it means.

The trade was priced for permanence. A market does not rise 100% in six months on the expectation of a normal cycle. Goldman Sachs’ estimate of roughly 300% earnings growth for the Korean market in 2026 is not a run rate; it is a one-off adjustment to a shortage. If second-quarter operating margins of 75% are the peak — and margins that high have never been sustained in memory — then every forward multiple applied to them is wrong in the same direction.

The deceleration already started. TrendForce’s third-quarter projection of 13% to 18% quarterly server DRAM increases, against 58% to 63% in the second quarter and roughly 95% in the first, describes a cycle rolling over. The stated reason — customers resisting once supplier operating margins exceed 80% — is the classic mechanism by which memory upcycles end. Buyers do not stop buying; they stop accepting the price.

Import substitution is a terminal-value story, not a next-quarter story. This is the point that arithmetic about five machines misses. Equity valuations discount decades. If the market’s prior was that China would remain structurally dependent on Western lithography, and Tuesday’s reports shifted that prior even modestly, the correct response is a large adjustment to the far end of the cash flow stream. That looks like an overreaction when measured against 2027 revenue and looks entirely rational when measured against 2035.

Export controls have a documented backfire mechanism. Van der Lugt’s observation is the crux: restrictions created the business case they were designed to prevent. Chinese chipmakers do not need Aishengna’s machine to be as good as ASML’s. They need it to exist, because the alternative is a supply chain that a foreign legislature can sever. The U.S. Congress is currently debating exactly such a severance for the remaining immersion DUV segment. A rational Chinese buyer, facing that risk, will accept a worse tool and fund its improvement.

The financing structure is genuinely opaque. Nvidia’s CDS did not widen by a record amount because traders read a news headline. They widened because $750 billion of prospective guarantees, backstops and equity commitments changes the profile of a company’s obligations in ways that are hard to model from public disclosure. When the IMF and BIS both flag a financing pattern as a systemic risk, dismissing it as sentiment is not analysis.

And the capital spending is discretionary. The $600 billion to $725 billion of estimated 2026 hyperscaler capital expenditure is a decision, not a contract. Alphabet raised its guidance and its stock fell 5% — a market signalling that it would prefer less. Four of the largest spenders report within 48 hours of the selloff. If any of them signals moderation, the memory shortage that produced 75% operating margins resolves faster than the incumbents’ capacity plans assume.

What the market commentary argued, and what the reporting will actually support

Broadcast and wire coverage on Tuesday converged on a handful of claims. Some are well supported, some are shorthand, and one or two are doing more work than the underlying reporting justifies. It is worth separating them.

“China has broken ASML’s monopoly.” Not supported. ASML’s monopoly on EUV is untouched, and its DUV position is unchallenged outside China. What is supported is that China has begun producing an immersion DUV tool of unknown capability for a domestic market it can no longer fully serve through imports.

“A Chinese semiconductor breakthrough caused the crash.” Partially supported, and incomplete. The report was the proximate trigger. The magnitude of the Korean move was determined by positioning, leverage and forced selling — Mirae Asset’s Kim Seok-hwan said as much explicitly, and the flow data supports him. Attributing a 10.84% index decline to a single news item ignores a 100% six-month rally and a 5.9 trillion won reduction in margin balances that had been under way for weeks.

“CXMT will take share from Samsung and SK hynix.” Premature. CXMT is the fourth-largest DRAM maker with rapidly growing revenue and $8.6 billion of new capital, which makes it a credible competitor in conventional DRAM. It is not currently a competitor in HBM, and the IPO proceeds are not allocated to a standalone HBM programme. Share gains in commodity DRAM and share gains in AI memory are different propositions with different margin consequences.

“Nvidia is inflating its own demand.” Not established. Nvidia has publicly denied using circular financing schemes. What is established: the company has announced more than $540 billion of related deals in 2026, a further roughly $250 billion arrangement is reported to be under discussion, and its credit default swaps widened by a record daily amount when the reporting emerged. Those are facts about scale and market reaction, not about accounting impropriety, and the distinction should be preserved.

“The AI trade is over.” Contradicted by the same day’s tape. Microsoft rose about 2.3% and Alphabet about 2.5% on July 28 while their suppliers fell sharply. Investors were repricing the distribution of returns within the AI supply chain, not exiting the theme.

A note on single-source reporting in this sector

Both of the week’s China stories rest on unnamed sources. The Information cited two people familiar with the matter; Reuters cited one person who declined to be identified because of the sensitivity of the subject. Every company named in either report — Aishengna, its two shareholders, SMEE, Yuliangsheng, SMIC, Hua Hong and CXMT — declined to comment or did not respond.

This is not a criticism of either outlet. Reporting on Chinese state-directed technology programmes is nearly impossible on the record, and both organisations have strong track records on exactly this beat. Reuters’ December report on a Chinese EUV prototype came from the same kind of sourcing.

But readers repricing portfolios on the basis of this reporting should be clear about what they are relying on. The core factual claim — that production has begun — is corroborated across two independent outlets with different sources, which is meaningful. The identity of the company rests on one Reuters source, supported by verifiable corporate registration and recruitment records that Reuters checked independently. The performance of the machine rests on the same single source’s characterisation. And the production targets of five units this year and 20 next rest on The Information’s two sources alone.

Those are three different confidence levels attached to three different claims, and the market on Tuesday priced them as though they carried one.

Three historical comparisons, and where each one breaks

Analogies are being deployed heavily this week. Most of them are imperfect in specific, instructive ways.

Nikon and Canon, 1998–2010

This is the most directly relevant precedent, and it cuts against ASML rather than for it. In the late 1990s, Nikon and Canon dominated lithography; ASML was the challenger. ASML won not through a single technical leap but through sustained improvement in yield, throughput, uptime and service — the unglamorous operational metrics that determine cost per wafer. The lesson is that lithography leadership is contestable over a decade and that early-generation inferiority is not disqualifying.

Where it breaks: ASML’s ascent was funded by customers who chose it on merit in an open market. Aishengna’s is funded by the state and driven by customers who may have no alternative. That accelerates adoption but weakens the feedback loop that made ASML good. A monopoly supplier to captive domestic customers has less pressure to improve than a challenger fighting for every socket.

The 2000 technology peak

The Kospi has drawn this comparison repeatedly. The index roughly tripled over twelve months, driven by two stocks, on an infrastructure buildout whose end-demand economics were unproven. Foreign investors sold 4.97 trillion won in a single day. Retail investors bought 4.33 trillion won of it — a pattern that has recurred at market turning points for a century.

Where it breaks: the 2000 comparison assumes the earnings were not real. In Korea’s case, they demonstrably are. Samsung’s 89.4 trillion won of guided quarterly operating profit is cash, not a projection about eyeballs. The question is not whether the profits exist but whether they persist, which is a cyclical question rather than a solvency one.

The 2018 and 2022 memory downturns

Memory has crashed twice in the last decade after supply caught up with demand. Both times, the peak in share prices preceded the peak in reported earnings by two to three quarters, because the market discounted the price decline before it showed up in the income statement. If that pattern holds, the June 22 peak in the Kospi would place the earnings peak somewhere in late 2026.

Where it breaks: previous memory cycles were driven by PC and smartphone demand, which is replacement-driven and reasonably predictable. AI infrastructure demand has no established replacement cycle and is being set by a handful of buyers whose capital budgets are strategic rather than demand-derived. The cycle could be shorter than 2018’s or considerably longer. Nobody has a template.

Risks worth tracking from here

• Memory price deceleration. The third-quarter projection of 13–18% quarterly increases for server DRAM is the single most important forward number for Samsung, SK hynix, Micron and CXMT. A move toward flat or negative quarterly pricing would change the earnings trajectory materially.
• Hyperscaler capital expenditure guidance. Microsoft, Meta, Apple and Amazon all report on July 29–30. Any signal of moderation in AI infrastructure spending transmits directly to memory and equipment demand.
• U.S. legislation on DUV exports. Congress is debating restrictions on ASML’s remaining immersion DUV sales to China. Passage would remove roughly the segment generating a meaningful share of ASML’s approximately €9 billion of expected 2026 China revenue while strengthening the case for Chinese substitutes.
• Qualification of Chinese lithography tools. The decisive evidence will be whether SMIC, Hua Hong or CXMT publicly qualify Aishengna’s machines for production use, and at what nodes. That is verifiable in a way that a single anonymous source is not.
• Korean leverage unwind. Margin loans have already fallen from 38.63 trillion won to 32.74 trillion won. Continued forced liquidation, amplified by two-times single-stock leveraged ETFs, can produce price moves disconnected from fundamentals in either direction.
• Nvidia’s financing commitments. If the reported OpenAI backstop is formalised, the scale and structure of the guarantee — and how it is accounted for — will determine whether the CDS widening was a one-day repricing or the start of a trend.
• Federal Reserve policy. Market pricing has shifted toward the possibility of a rate increase. A hawkish surprise on July 29 would compress valuations across long-duration assets, of which AI infrastructure equities are among the longest.
• Middle East de-escalation reversing. Oil’s retreat from roughly $100 to the mid-$80s is contingent on talks that President Trump himself described conditionally. Shipping risk through the Strait of Hormuz and disruption in the Red Sea remain live.
• Currency stress in emerging Asia. The peso’s record low and the yen near a four-decade trough both reflect dollar strength. Further dollar appreciation raises imported inflation across the region and constrains central banks that would otherwise support growth.
• Concentration risk in Korean indices. Two stocks drove the Kospi’s rise and its fall. Index-level exposure to Korea is, in practice, a leveraged position on the global memory cycle.

What happens next

The calendar over the following 48 hours is unusually dense, and each item speaks directly to one of the questions raised on Tuesday.

SK hynix second-quarter results, July 29. These will be the first hard data on whether the memory boom is still accelerating. Consensus among 14 domestic brokerages points to 84.1 trillion won of revenue and 64.1 trillion won of operating profit, at a margin somewhere between 74% and 77%. The results themselves are backward-looking; what matters is management’s commentary on third- and fourth-quarter pricing, HBM allocation and the status of the long-term agreement structure with customers. A company earning a 75% operating margin has limited scope to surprise positively on the quarter and considerable scope to disappoint on the outlook.

The FOMC decision, July 29. Consensus expects the target range to hold at 3.50%–3.75%. Market pricing carries a roughly 38% probability of a 25-basis-point increase. Under Chair Kevin Warsh the Committee has discontinued forward guidance, so the statement language on core inflation will carry disproportionate weight.

Microsoft and Meta Platforms earnings, July 29; Apple and Amazon, July 30. Three of the four are among the largest AI infrastructure spenders on earth. Their capital expenditure guidance is, functionally, the demand forecast for the entire semiconductor complex that sold off this week.

Beyond the immediate week, the questions are slower. Whether Aishengna’s tools are accepted by customers. Whether Congress restricts ASML’s remaining China exports. Whether CXMT converts $8.6 billion of equity into competitive DRAM capacity, and on what timeline. Whether the reported Nvidia–OpenAI arrangement becomes a signed agreement or quietly disappears. None of those will be settled by the end of the quarter.

The terms that did the heavy lifting this week

Several pieces of specialist vocabulary carried most of the analytical weight in Tuesday’s coverage. Brief definitions, because the distinctions between them are where the argument actually lives.

• DUV and EUV. Deep ultraviolet lithography uses 193-nanometre light; extreme ultraviolet uses 13.5 nanometres. EUV is required for leading-edge logic in a single exposure and is made only by ASML. China is barred from purchasing it and, per Reuters’ December reporting, has produced a prototype but remains years from production.
• Immersion. A layer of purified water between the final lens and the wafer, which raises the effective resolution of a DUV system. This is the specific capability China is reported to have begun producing.
• Multiple patterning. Exposing the same wafer layer several times with offset masks to achieve features finer than a single exposure allows. It works, and it multiplies process steps, cost and defect risk.
• DRAM and HBM. Dynamic random-access memory is the commodity working memory in servers, PCs and phones. High-bandwidth memory is DRAM stacked vertically with through-silicon vias and packaged beside an accelerator. HBM commands far higher prices and margins; conventional DRAM is where a low-cost entrant does damage.
• Contract price versus spot price. Memory makers sell most volume on negotiated contracts, often quarterly. TrendForce’s quoted percentages refer to contract prices, which lag and smooth spot movements. A deceleration in contract price increases signals a cycle turning several months before it appears in reported revenue.
• Circuit breaker and sidecar. A Korean circuit breaker halts all trading for 20 minutes after an 8% index decline sustained for one minute. A sidecar suspends program selling for five minutes after Kospi 200 futures fall 5%. The first is a market-wide pause; the second targets automated order flow.
• Credit default swap. A contract that pays out if an issuer defaults. Its price, quoted as an annual percentage of the notional amount protected, is a market estimate of credit risk. Nvidia’s five-year CDS rising to roughly 0.82 percentage point means about $82,000 a year to protect $10 million of exposure.
• Basis point and percentage point. A basis point is one hundredth of a percentage point. A 25-basis-point Federal Reserve move changes the target range by 0.25 percentage points. These are arithmetic differences between rates, distinct from percentage changes in a price.
• Circular financing. An arrangement in which a supplier funds, guarantees or takes equity in customers who then purchase its products. Legal and common at small scale; contentious at the scale currently reported in AI infrastructure because it complicates the interpretation of demand.
• Letter of intent. A non-binding statement of intent to negotiate definitive agreements. Headline values attached to letters of intent — such as the $500-billion-plus Nvidia–SK figure — are not contracted revenue.

Frequently asked questions

Why did chip stocks fall on July 28, 2026?

Three catalysts converged. The Information reported on July 27, and Reuters confirmed on July 28, that a Chinese state-owned company had begun manufacturing immersion DUV lithography machines, a technology ASML has dominated. China’s largest DRAM maker, CXMT, closed its Shanghai debut up roughly 466%, raising concerns about a better-capitalised Chinese memory competitor. And The Wall Street Journal reported that Nvidia was in talks to guarantee about $250 billion of financing for OpenAI, reviving worries about circular financing in the AI supply chain.

How far did the Kospi fall, and why so much more than other markets?

The Kospi closed down 10.84% at 6,023.66 on July 28, its worst day since March 4. It fell harder than Tokyo or Taipei because Korean index weight is heavily concentrated in Samsung Electronics and SK hynix, because the index had roughly doubled in 2026 on the memory cycle, and because a large volume of retail margin and leveraged-ETF exposure amplified the decline through forced selling.

What is immersion DUV lithography, in plain terms?

It is a machine that prints circuit patterns onto silicon wafers using 193-nanometre ultraviolet light, with a layer of purified water between the lens and the wafer to sharpen the image. It is the workhorse tool of the semiconductor industry, used for the majority of the world’s chips, and can be pushed to more advanced nodes through repeated exposures.

Does this mean China has caught up with ASML?

No. The reported programme targets roughly five machines in 2026 and about 20 in 2027, against 131 immersion DUV systems ASML shipped in 2025. Reuters’ source said the Chinese tool requires further testing and remains far from matching ASML’s competing models, and some critical components still come from Japan. China also cannot make or buy the more advanced EUV systems required for leading-edge logic.

What is circular financing, and has Nvidia admitted to it?

Circular financing describes an arrangement in which a supplier invests in or guarantees financing for its own customers, who then use that capital to buy the supplier’s products. Nvidia has publicly stated that it does not use circular financing schemes; several prominent short sellers have publicly disagreed. The IMF and the Bank for International Settlements have both flagged the pattern as a systemic risk in the AI sector.

Why did Nvidia’s credit default swaps matter more than its share price?

Nvidia shares fell 4.99% on July 27, which is a large but not extraordinary move for the stock. The five-year CDS widened by the most on record on the same day, to as high as roughly 0.82 percentage point. Credit markets price the scale and structure of obligations rather than earnings sentiment, so the CDS move was read as a signal about the balance-sheet implications of more than $750 billion of prospective commitments.

How big was CXMT’s IPO?

CXMT raised 57.92 billion yuan, about $8.6 billion, selling 6.688 billion shares at 8.66 yuan each. It was the largest listing on a mainland Chinese exchange since Agricultural Bank of China in 2010 and the largest ever on Shanghai’s STAR Market. First-day gains of roughly 466% put its market value near 3.3 trillion yuan, approximately $480 billion.

Can foreign investors buy CXMT shares?

Not readily. As of the listing, CXMT was not eligible for trading through the Shanghai–Hong Kong Stock Connect, which means most offshore investors cannot freely buy or sell the shares. Kiwoom Securities analyst Han Ji-young cited this as a reason to expect limited actual capital outflows from Korean semiconductor stocks into CXMT, whatever the effect on sentiment.

What is a circuit breaker on the Korea Exchange?

A marketwide circuit breaker halts all trading for 20 minutes when the Kospi falls 8% or more from the previous close and holds that level for one minute. A separate mechanism, the sidecar, suspends program sell orders for five minutes when Kospi 200 futures fall at least 5%. Both were triggered on July 28. It was the eighth circuit breaker of 2026; the Kospi market has also triggered 41 sidecars this year, more than during the whole of the 2008 financial crisis.

Did Samsung and SK hynix report bad earnings?

No. Samsung guided to a record second-quarter operating profit of 89.4 trillion won, up 1,810% from a year earlier. SK hynix, which reports on July 29, was expected by a consensus of 14 Korean brokerages to post 64.1 trillion won of quarterly operating profit — more than its entire 2025 total. The selloff reflected concerns about future pricing and competition, not reported results.

What happened in the Philippine market on the same day?

The PSEi rose 0.54% to 6,314.90 as falling oil prices supported sentiment. BSP Governor Eli Remolona said there was only a “small chance” of a 50-basis-point rate increase. BDO Unibank settled a record 132 billion peso ASEAN Sustainability Bond issue at a 6.26% coupon over an 18-month tenor. Petron was awarded a 340 million peso Jet A-1 fuel farm project at Bohol-Panglao International Airport.

What should investors watch next?

SK hynix’s results and outlook on July 29, the Federal Reserve’s decision the same day, and capital expenditure guidance from Microsoft and Meta on July 29 and Apple and Amazon on July 30. Over a longer horizon, the decisive evidence on the China question will be whether SMIC, Hua Hong or CXMT publicly qualify domestically built lithography tools for production.

Final assessment

The most common description of July 28 — that a Chinese technology breakthrough crashed the global chip trade — is not quite right, and the imprecision matters.

What was reported is a state-owned company in Shanghai producing a small number of lithography tools that, by the account of the only source describing them, still require testing and remain well behind ASML’s equivalents. That is a real development. It is not a competitive threat to ASML’s 2026 or 2027 revenue, and the sell-side consensus on that point is correct.

What actually moved markets was the removal of an assumption. For a decade, the investment case for the Western semiconductor complex rested partly on the belief that export controls would keep China structurally dependent — that whatever Beijing spent, the chokepoints would hold. Tuesday’s reporting did not disprove that belief. It made it contestable. And when an assumption that supports the far end of a discounted cash flow model becomes contestable, the correct repricing is large even if the near-term numbers do not change at all.

The Korean market’s reaction was amplified by conditions that had nothing to do with China. The Kospi had roughly doubled in six months on two stocks. Margin balances had reached records. Two-times single-stock leveraged ETFs on Samsung and SK hynix, approved in May, mechanically sell into declines. Foreign investors held a crowded position and reduced it violently — 4.97 trillion won in a session. Mirae Asset’s Kim Seok-hwan is almost certainly right that no new shock was required; familiar risks met thin liquidity and forced selling did the rest.

The circular financing question is the one that deserves the most continued attention, and it is the least resolved. Nvidia says it does not engage in the practice. The IMF and BIS have both identified the pattern as a systemic risk. More than $540 billion of related deals have been announced in 2026, with a further $250 billion arrangement reported to be under discussion. The credit market repriced Nvidia’s default risk by a record daily amount on the news. Those facts do not establish that anything improper is occurring — vendor financing is a legitimate and common commercial tool. They do establish that the scale is unprecedented and the disclosure is thin, and that combination is what markets tend to punish.

What changed on July 28 is the burden of proof. Until this week, the AI infrastructure trade enjoyed the benefit of the doubt: demand was assumed durable, Western technological leadership assumed secure, and financing structures assumed benign. All three assumptions now require evidence rather than assertion.

The evidence arrives quickly. SK hynix’s outlook lands on July 29, the Fed the same day, and three of the four largest AI spenders on earth report within 48 hours. A market that has fallen 34% from its June peak has already priced a great deal of pessimism into Korean memory. Whether it has priced enough — or too much — depends almost entirely on numbers that will be public before the week is out.

Sources

• Reuters — “Exclusive: China starts production of home-grown immersion DUV chipmaking tools,” by Fanny Potkin, July 28, 2026
• Reuters — “China’s chip tool push shows ASML caught in US-China squeeze,” by Toby Sterling, July 28, 2026
• Korea JoongAng Daily — “Kospi sinks over 10 percent as China chip fears hit,” July 28, 2026
• Korea Economic Daily — “Asian Chip Stocks Suffer Worst Day as AI Financing Fears, China Shock Hit Sector,” July 28, 2026
• TechNode — “CXMT becomes China’s most valuable A-share company after $8.6 billion IPO,” July 27, 2026
• Yahoo Finance — “Nvidia drops nearly 5%, leading chip stocks lower amid renewed worries of circular financing,” by Ines Ferré, July 27, 2026
• Nvidia newsroom — “SK Group and NVIDIA Expand Strategic Partnership Across AI Factories and Next-Generation Memory,” July 24, 2026
• TrendForce — “Long-Term Agreements Cap Price Increases; Server DRAM Contract Prices Expected to Rise 13–18% QoQ in 3Q26,” July 9, 2026
• TrendForce — “Rapid Contract Price Surge Drives 1Q26 DRAM Industry Up 81% QoQ,” June 1, 2026
• The Korea Times — “SK hynix expected to post record $43.7 bil. in Q2 operating profit,” July 26, 2026
• TheStreet — “Stock Market Today (July 28, 2026): Nasdaq futures fall on chip sell-off,” including Fifth Third economist Bill Adams on the FOMC
• The Motley Fool — “Dow Jumps 520 Points While Nasdaq Struggles Through Chip Stock Panic,” July 28, 2026
• Yahoo Finance — “Stock market today: S&P 500, Nasdaq futures slide as chip sell-off deepens,” July 28, 2026
• CNBC — “U.S. crude oil falls below $80 as Iran discusses Strait of Hormuz with Saudi Arabia and Oman,” July 28, 2026
• BusinessWorld — “BSP chief sees ‘small chance’ of aggressive tightening amid renewed volatility,” by Katherine K. Chan, July 28, 2026
• The Philippine Star — “PSEi returns to 6,300 level as Mideast tensions ease,” July 28, 2026
• GMA News Online — “BDO gains record P132-B from ASEAN Sustainability Bonds issue”
• Philippine News Agency — “Aboitiz taps Petron for construction of fuel farm at Bohol airport”
• Seoul Economic Daily — “KOSPI Triggers 41 Sidecars in Six Months, More Than During Global Financial Crisis,” July 25, 2026
• Seoul Economic Daily — “KOSPI Triggers Circuit Breaker, 8th This Year,” July 28, 2026
• Korea JoongAng Daily — “Lee orders swift steps on Samsung, SK hynix leveraged ETF risks”
• CNBC — “1 hyperscaler megacap down, 3 to go. Alphabet raises the stakes on AI spending,” July 24, 2026
• Goldman Sachs — “Korea’s Stock Market Is Forecast to Set Fresh Highs”

This article is provided for general informational purposes and does not constitute financial, investment, tax, or legal advice.

Affiliate disclosure: Businessfinance.news may earn compensation from qualifying actions completed through selected links on this website, at no additional cost to the reader. Affiliate relationships do not influence our editorial reporting, analysis, or conclusions.