ASML, TSMC and the China Chip Selloff of October 2026: Which Moats CXMT Can Reach

Over the weekend of October 10, 2026, CXMT's president told an industry event in Shanghai that the Chinese memory maker had made key progress on a 4F² DRAM cell, the denser layout Samsung and SK hynix are also chasing, and that it would ship a server memory product built on it by the end of the year. Nobody outside CXMT has verified it. On Monday Micron and SK hynix's US shares fell about 4% each. On Tuesday SK hynix fell about 9% in Seoul and Samsung about 8%, by TIKR's count, and a Saxo strategist called the CXMT news a clear trigger for the selloff in memory stocks.

ASML went down with them. It was off more than 1% in Amsterdam on Tuesday, alongside BE Semiconductor and ASM International, a day before it reported anything. On Wednesday it reported the best quarter in its history and fell about 2% in New York, which Investing.com put down to valuation, on a day when hot inflation data and the CoreWeave bond scare hit the whole tech sector. By then the semiconductor index was down more than 10% in five days.

The case for selling everything is worth stating properly, because it isn't silly. China is pouring capital into every layer of the chip stack, it's been closing gaps faster than export controls were meant to allow, and every layer it closes takes a customer away from someone Western. The headlines on Tuesday framed it that way, as China competition weighing on chips, not on memory. If CXMT can make denser DRAM, the argument goes, Chinese toolmakers and foundries can't be far behind, and the premium on every Western chip name assumes they are.

The table also shows the market already drew a line, at least in size. Memory fell 4 to 9% on the claim, and ASML fell 1 to 2% on days that had other news in them too. So my disagreement isn't that the market treated ASML like SK hynix. It's that it sold ASML on the CXMT news at all, as if a Chinese company getting better at DRAM were bad for the company that sells the scanners DRAM is made on. ASML's own quarter doesn't read that way.

I've taken the three memory makers CXMT competes with and the two chokepoint names China buys from. The matrix shows the pillars that decide whether money and time are enough to catch up. Scale economics is the one a state-funded entrant can buy outright. Proprietary data and system of record take decades of production to build. Transaction embedding is what keeps a customer once a rival exists.

The two groups don't look alike. None of the memory makers has a system of record, because a customer buys DRAM from all three at once, and their proprietary data reads intact rather than strong. Scale is the pillar CXMT can attack, and it's the one the IPO money is for. CXMT raised about $8.6 billion in its Shanghai listing to build fabs, Hana Securities expects its DRAM wafer capacity to pass 500,000 a month by 2028, and Counterpoint has its revenue share at 8% in the first quarter of 2026, more than double a year earlier. A 4F² cell that works would add a density step on top, and density is the other half of cost per bit. Micron is the exception in the group. The framework's only strong pillar there is transaction embedding, from the take-or-pay contracts, which is a cushion CXMT can't take away by building fabs.

ASML and TSMC read differently. Both are strong on proprietary data, system of record and transaction embedding. None of that is something a new entrant can buy with an IPO. It's decades of machine performance data at ASML, and at TSMC the process recipes and the design kit every leading-edge customer has designed against. CXMT's new cell doesn't route around either. It has to be patterned on someone's lithography, and the reporting on CXMT's expansion says that lithography is ASML's.

The scorecard isn't built to sort on this, and valuation moves these names more than any one week's news. What I take from the screen is narrower. A working Chinese DRAM cell would be a pricing problem for the first group. For the second it would mostly be more demand for tools and, through Chinese chip designers, for wafers.

• China can't buy EUV at all, so CXMT has to build whatever cell it makes on immersion DUV, with multiple exposures where it needs finer features. The FT reported in September that CXMT and YMTC have stockpiled enough ASML scanners for three years of planned expansion, so the scanners its next fabs run on are ASML's.
• Scanners keep needing ASML after they're installed. Tech Times, relaying analysis from Model Diplomat and the American Enterprise Institute, says DUV scanners need calibration, replacement optics and firmware from ASML about every six months. That's why the MATCH Act's proposed licence requirement for servicing tools at CXMT, YMTC and SMIC is the lever export hawks keep coming back to.
• CXMT is a memory company. Nothing in its announcement touches the leading-edge logic TSMC makes for Apple, NVIDIA and AMD. Chinese chip designers are TSMC customers where the rules allow, and TSMC's competition in China is SMIC, a separate company with a separate story.

TSMC reported its third quarter on October 15, the morning after ASML, and its geographic split makes the same point more bluntly. China, meaning customers headquartered there, was 5% of TSMC's revenue in the quarter, down from 8% a year earlier. North America was 79%. If the CXMT news were going to reach TSMC, it would have to come through that 5%, and CXMT isn't a foundry customer in the first place.

The margin is the line a bear would pick on. It slipped about a point from the second quarter while revenue jumped, and the CFO put that down to the steep 2nm ramp and overseas fabs. Nothing in the print ties it to China. The quarter was 68% high-performance computing, 79% advanced nodes and 79% North American customers, and TSMC guided the fourth quarter up again and said demand is still far above what it can supply. That's a description of the AI build, not of a company exposed to a Chinese memory maker. China didn't come up on the call at all.

There are real ways China reaches ASML, and the first is domestic lithography. According to Korea JoongAng Daily, Aishengna, a Chinese toolmaker, plans to ship about five DUV lithography tools this year and 20 in 2027. That's tiny next to the 121 systems ASML sold in the third quarter alone, but it's aimed at exactly the part of ASML's business China still buys. Every Chinese-made scanner that works at volume is one ASML doesn't sell or service there, and China is still around a fifth of ASML's business this year.

The second is the stockpile itself. If CXMT and YMTC already own three years of ASML scanners, China's memory build-out doesn't need new ASML orders for a while, and by the time it does, domestic tools may be good enough for the older layers. Seen that way, CXMT's progress isn't good for ASML's future China revenue at all. The servicing lever is real, but the MATCH Act isn't expected to be enacted this Congress, and either outcome hurts ASML. If it passes, ASML loses the service revenue. If it doesn't, the stockpile keeps running.

The third is technical, and I can't settle it. A 4F² cell is a vertical structure, and part of the reason the industry is moving to it is that shrinking the old layout with lithography alone has got hard. If that's right, the new cell may lean more on etch and deposition than on lithography, and those are the tool categories where Chinese makers like Naura and AMEC are furthest along. In that case a Chinese DRAM breakthrough would be good for Chinese toolmakers, not for ASML.

The fourth applies to TSMC. SMIC has been making 7nm-class chips for Huawei on multi-patterned DUV, the same workaround CXMT is using, so China does compete with TSMC, just not at the leading edge and not through CXMT.

The fifth is second-order and I think it's the strongest. Stock Titan's reading of ASML's presentation has South Korea at 31% of third-quarter system sales, and that's Samsung and SK hynix building memory capacity. If CXMT really does take server DRAM share and push prices down, Korean memory capex is the first thing to get cut, and that's ASML's order book. A memory price war doesn't need a single Chinese scanner to reach ASML. It only needs Samsung and SK hynix to buy fewer of ASML's.

I'd concede the first two as real, slow risks to ASML's China revenue. The matrix doesn't show them because the framework treats ASML's China exposure as a sales headwind rather than a moat question, and 20 Chinese scanners in 2027 is still a forecast, not a shipment. I can't rule out the third, and if it's right it weakens the mechanism above without touching ASML outside China. The fifth is where I disagree on timing more than direction. ASML says its planned supply won't fully meet demand in 2027 or 2028 and some customers want to order for 2030, so a Korean capex cut would shorten a long queue before it emptied a factory. That's a cyclical risk the valuation should carry, and it's different from a Chinese competitor reaching the moat.

For the memory makers, I think the market's reaction is roughly the right shape, even if the size is a guess. CXMT is going after the pillar it can buy, it has the capital to do it, and its share has already more than doubled in a year. The framework rates scale intact rather than strong at Micron and Samsung, and SK hynix's strong rating rests on its HBM lead, which a server DRAM product doesn't touch yet. Micron's take-or-pay contracts, which the Micron article argues are what sorts the memory makers in a downturn, matter more after this week.

For ASML and TSMC, the CXMT day put a memory company's risk on companies that sell to memory and chip companies. The China risk they actually carry is ASML's shrinking and eventually substitutable China DUV business, SMIC at TSMC's mature nodes, and a Korean capex cycle if memory prices crack. I'd watch all three, and none of them is a rival getting to the process step a customer can't skip, which the TSMC article argues is where the moat sits. Whether the selloff made either name cheap enough is a valuation question, and the scorecard above carries the live answer.

When I read the next China chip headline, the first thing I'll check is whether the Chinese company in it sells to the same customers as the stock I'm looking at or buys from it. This week that question put Samsung, SK hynix and Micron on one side and ASML and TSMC on the other, and I don't think the selloff made that distinction.