South Korea equity market suffered one of its worst sessions in years on Tuesday, and the trigger was not a central bank, an earnings miss or a currency crisis. It was a report that Chinese factories have begun mass production of the machines used to manufacture semiconductors.
The Kospi index fell 10.8% to close at 6,023.66, its weakest finish since April. Trading was halted more than once as circuit breakers tripped on the speed of the decline. Samsung Electronics lost 13.4%. SK Hynix fell 14.7%, and its US-listed shares closed at $143, below the $149 price at which they were first sold to the public.
The selling did not stay in Seoul. It moved through Japanese chip names, then into the American session, where semiconductor stocks fell for a fourth consecutive day even as the broader US market advanced.
What made this different from the ordinary volatility that has surrounded artificial intelligence stocks all year is the nature of the news. This was not a change in demand. It was a change in who is able to supply.
What was actually reported
The specific trigger was a report that Chinese manufacturers have started mass production of deep ultraviolet lithography tools, known as DUV. Lithography machines are the equipment that prints circuit patterns onto silicon wafers. They are the single most difficult product in the semiconductor supply chain to build, and for two decades the market for the most advanced versions has been effectively a monopoly held by one Dutch company.
Alongside that report, the Chinese memory manufacturer CXMT listed on the Shanghai exchange and rose 466% on its debut before giving back about 4% in the following session. A domestic memory champion arriving at an enormous valuation, in the same week as a domestic tool-making breakthrough, is a combination the market read as a single message.
Why lithography is the whole argument
To understand why a machine caused a 10% move in a national index, it helps to separate two technologies that are often blurred together.
EUV, extreme ultraviolet lithography, is required for the most advanced logic chips. It remains the harder problem, and access to it is tightly controlled. DUV is the older, less exotic technology. It is also the workhorse. A very large share of the world semiconductor output, including most memory chips and most mature-node logic, is manufactured on DUV tools.
Export restrictions imposed since 2022 were built on a specific assumption: that denying China access to advanced lithography would cap the sophistication and volume of what Chinese fabs could produce, and that the gap would take many years to close because the engineering is genuinely hard. Restricting DUV sales was a central pillar of that policy precisely because DUV is where the volume is.
If domestic DUV production is real and scaling, that pillar weakens. Not immediately, and not completely. But the market does not price the end state. It prices the direction, and the direction changed on Tuesday.
Why memory took the worst of it
Samsung and SK Hynix are not primarily logic companies. They dominate memory, and specifically the high-bandwidth memory that sits alongside processors in artificial intelligence systems. That business has been extraordinarily profitable through this cycle for one reason: supply has been scarce relative to demand.
Memory is a commodity with a long and unforgiving history. When capacity is tight, margins are enormous. When capacity arrives, prices fall faster than anyone forecasts, because the product is fungible and the marginal cost of running an existing fab is low. The industry has been through this cycle repeatedly since the 1980s, and it has bankrupted large companies each time.
The entire bull case for Korean memory rested on the belief that meaningful new supply could not appear quickly. A Chinese memory maker arriving at a large valuation, in a week when Chinese tool production was reported, attacks that belief directly.
This also explains why the reaction was so violent relative to the size of the news. These stocks had risen a very long way. Positioning was heavy. The report did not need to be conclusive to force selling; it only needed to make the core assumption debatable.
The history this rhymes with
Export controls on strategic technology have a consistent record, and it is not the record policymakers usually expect.
Restrictions typically work in the short run and accelerate substitution in the long run. They raise costs and delay timelines for the restricted party, which is a real effect. But they also convert a commercial procurement decision into a national priority, and they remove the competitive pressure that normally makes domestic champions uneconomic. A country that could previously buy a tool more cheaply than it could build one is suddenly willing to fund the building regardless of cost.
The pattern appeared in aerospace, in nuclear technology, in satellite components and in machine tools during earlier decades. In each case the restricted party eventually developed a domestic alternative that was initially inferior, expensive and heavily subsidised, and in several cases eventually competitive.
The open question in semiconductors was never whether this would happen. It was how long it would take, and whether the incumbent could stay far enough ahead that the domestic alternative arrived into a market that had already moved on.
What this means for the AI build-out
There are two readings, and they point in opposite directions for different companies.
- For the memory manufacturers, cheaper and more plentiful supply is unambiguously negative. Their recent earnings power came from scarcity.
- For the companies buying memory by the container load to fill data centres, cheaper components are helpful. Component cost inflation has been cited as a reason capital budgets keep rising.
- For equipment makers, domestic Chinese tooling removes a large future customer and eventually creates a competitor in lower-end segments.
- For the policy framework itself, the episode raises a harder question about whether restriction has bought time or merely bought a competitor.
The second point deserves emphasis, because it complicates the simple narrative that this is bad news for artificial intelligence. Our analysis of Big Tech capital spending noted that rising component costs, particularly in high-bandwidth memory, were breaking the assumption that unit costs fall over time. Abundant memory supply would repair that assumption. It would be painful for the sellers and useful for the buyers.
What remains unverified
A degree of caution is warranted, and the size of the market reaction is not itself evidence.
Mass production of a tool is not the same as mass production of a competitive tool. Lithography systems are judged on throughput, yield, uptime and precision, and a machine that works in a demonstration is a long way from a machine that a fab will run continuously against tight tolerances. None of those operating characteristics were established by the report that moved the market.
Nor does DUV capability address the most advanced logic nodes, which still require EUV. The immediate competitive threat is concentrated in memory and mature-node manufacturing, which is where the Korean producers happen to be most exposed. That is why the selling was so specific.
Outlook
The Kospi decline is best understood as the repricing of a single assumption rather than a verdict on artificial intelligence demand. Investors had been paying for scarcity. They were reminded that scarcity in semiconductors is a temporary condition, historically the most temporary condition in the industry.
What to watch from here is prosaic. Memory contract pricing over the coming quarters will show whether supply is genuinely arriving. Capital expenditure guidance from the Korean producers will show whether they intend to defend share or protect margin. And equipment order books will show whether Chinese demand for imported tools is falling because domestic alternatives exist, or merely because inventories are full.
For now, one report was enough to erase more than a tenth of a national stock market in a session. That tells you how much of the AI trade was resting on the belief that no one else could build the machines.
About the data: Index and share price moves are closing values for 28 July 2026, including the Kospi close of 6,023.66 and the declines in Samsung Electronics and SK Hynix. The SK Hynix US-listed closing price and its original offering price are as reported by the exchange. The reported start of Chinese mass production of DUV lithography tools, and the CXMT listing performance in Shanghai, are described here as market-moving reports; the operating performance of the tools concerned has not been independently established. Export control history is drawn from published policy measures since 2022.
Reader note
This article is general information, not personalized financial advice. Read our Financial Disclaimer.