To The Moon

$ASML Holding NV(ASML)$  

Mathematical Money | October 2026

Yesterday the KOSPI broke below 6,000 intraday. SK Hynix fell more than 9%, Samsung Electronics more than 8%, and a sell-side trading curb was triggered for the first time in two months.

The trigger was CXMT, a Chinese memory maker, announcing progress on a memory-cell design that packs more storage into less space. Samsung, SK Hynix and Micron are roughly 90% of the world's DRAM, so a credible Chinese entrant with a density advantage threatens all three at once.

I wrote about it that day because I own Western Digital, and I argued the selloff was being applied to companies it did not belong to.

This morning the one company that decides whether anybody can build dense memory at scale reported its results. I want to test what I said against what it published, because that is a far better use of an earnings report than repeating the headline.

What ASML actually said

Third quarter revenue of €11.7 billion and net income of €5.0 billion. Gross margin of 57.5%. On both sales and margin it was the strongest quarter in the company's history. Full-year guidance went up for the third time, to €44–45 billion.

Three things in there matter far more than the beat.

One. Memory is sold out. DRAM customers are signing long-term agreements to lock in tool time years ahead, and the company says demand runs past its capacity through 2028.

Two. SK Hynix alone is installing 20 EUV machines over the next two years, specifically for high-bandwidth memory. ASML expects to ship somewhere north of 60 EUV machines in all of 2026, to every customer on earth. One Korean memory maker is taking the equivalent of a sixth of global annual output.

Three, and this is the one nobody put next to the Korea selloff. ASML has never exported a single EUV machine to China. Not one, ever.

Put those two stories side by side

The company that frightened Korea yesterday is building memory without the machine its Korean rivals are buying twenty of.

China can buy ASML's older DUV equipment, and it does — about €2.9 billion of it in the first half of this year. But the advanced tool is export-controlled and has never crossed the border. Mainland China is around 20% of ASML's sales this year, down from 33% in 2025, and the decline is the export rules working as designed.

That does not make CXMT's announcement fake. Getting more density out of older lithography is a genuine engineering achievement and it is exactly what a company does when it cannot buy the newer machine. But it does put a ceiling on it, because the workarounds cost yield and they cost money, and they get harder at every subsequent node while the Koreans simply order another EUV machine.

So the Korea selloff priced a competitive threat as though it had no constraint. The constraint is a Dutch company's export licence, and it has not moved.

And the three incumbents are spending more, not less

If DRAM were about to be commoditised by a Chinese entrant, you would expect the incumbents to pull back. The opposite is happening in 2026:

SK Hynix, about $20.5 billion, up 17%.

Samsung, about $20 billion, up 11%.

Micron, about $13.5 billion, up 23%.

More than $54 billion of combined spending, every one of them increasing. Meanwhile DRAM cleanroom capacity is genuinely tight — only Samsung and SK Hynix can expand their lines much, and Micron's new American fab is not expected to run before 2027.

Companies facing imminent commoditisation do not raise capital spending by double digits across the board. They cut.

Why this matters for what I own

Here is the part specific to my book, and it is the opposite of what the 13 October tape implied.

I own Western Digital shares and long-dated calls. When I wrote about the memory selloff I made the honest bear case rather than dodging it: cheap flash substitutes for hard drives at the margin. If Chinese competition pushed memory and flash prices structurally down, some workloads that sit on high-capacity drives would migrate, and the boundary between the two tiers would compress. I said that was a 2029 problem rather than a 2026 one.

ASML's order book argues it is further away than that.

Memory sold out through 2028, tool time locked years ahead, cleanroom capacity constrained, and the one country that might flood the market unable to buy the machine. That is not a world where flash gets cheap. It is a world where memory stays expensive. And expensive memory is precisely what keeps hard drives cost-advantaged in the bulk storage tier, which is the whole reason Western Digital's business exists.

So the equipment maker's results have made my storage position's main long-term risk less pressing, not more. That connection is not in any of the coverage I have read today, which is mostly about whether ASML is expensive.

The check that keeps working

In that 13 October post I ran one test: if this were really a demand problem for stored data, the companies that buy drives would be falling hardest. They were not. NetApp was making new highs while the drive makers were being sold.

Measured from each company's own 52-week high, as of Monday's close:

NetApp, down 0.1% — effectively sitting on its high, and up about 27% in a month.

ASML, down 13.7%.

Micron, down 21.5%.

Seagate, down 31.7%.

SanDisk, down 36.5%.

Western Digital, down 49.0%.

Third time in eight days that the customer has gone one way and the supply chain the other. I have now stopped treating that as a coincidence.

What I am not claiming

ASML did not mention CXMT. It did not comment on the Korea selloff. Every link between those two stories in this post is mine, and someone could reasonably argue that CXMT's density gain is precisely designed to route around the EUV constraint and that I am underrating it. I also cannot see CXMT's yields, because nobody outside the company can — anybody quoting a Chinese DRAM yield figure is estimating.

And the other side of the ledger deserves saying plainly: ASML beat and the shares still fell, roughly 2%, on a valuation of about 55 times earnings. Sold out through 2028 can be entirely true and the stock still be too expensive. Those are different questions and I am only answering the first.

What I am doing

Nothing today, which is the dull and correct answer. I am not adding to Western Digital, because down 49% from its high is a fact about the tape and not a thesis, and because the long-dated calls there are already the worst-performing thing I own.

What changes is the clock, not the position. The risk I named yesterday is one I would now place further out still, and that matters more for whether I roll those calls than for anything I would do this week.

What would change my mind: an export-control change. If the rules on advanced lithography to China loosen, the ceiling I have just described stops existing, and the 13 October selloff turns out to have been early rather than wrong. That is a political decision, not a technological one, and it is the thing I will be watching rather than the share prices.

The market spent a session pricing a Chinese memory breakthrough as though nothing stood in its way. The thing standing in its way reported this morning, said it was sold out until 2028, and has never sold China the machine.

One earnings report is not a thesis. But when a story and the constraint on it turn up a day apart and nobody puts them in the same sentence, that gap is usually where the interesting part is.

#semiconductors #memory #EUV #storage #investing

Stop guessing. Start calculating.

Live to fight another day. 🤙