Memory outsold logic at ASML last quarter. First time in the company's history as the sole supplier of EUV lithography. The stock fell 6%.
The headline numbers were clean. Q1 revenue €8.8 billion vs. €8.5 billion expected. Net income €2.8 billion vs. €2.5 billion expected. EPS €7.15. Full-year 2026 guidance raised from €34-39B to €36-40B — a €1.5 billion midpoint raise on a revenue base the market had already priced.
That should have been a green day. It wasn't, because Q2 guidance (€8.4-9.0B) came in below the €9.07B consensus, and because China's share of system shipments dropped from 36% to 19% on the MATCH Act overhang — bipartisan draft legislation that would restrict even immersion DUV tools and potentially block maintenance services to mainland China.
The China story is real. It's also what everyone was already watching. The trade nobody made was Korea.
The Inversion
ASML publishes a system-sales breakdown by end-market. For roughly a decade, that split has leaned logic — TSMC and Intel buying EUV scanners to fabricate leading-edge processors. Memory makers bought less because DRAM and NAND nodes didn't need EUV. That changed this quarter.
A 21-point mix shift in a single quarter, for a €30+ billion revenue machine, is not noise. It's a tectonic indication of where the semiconductor capex dollar is going: into memory fabs, specifically into high-bandwidth memory (HBM) for AI accelerators.
Korea Tripled
The geographic breakdown explains the memory surge exactly. South Korea's share of ASML system shipments jumped from 22% in Q4 2025 to 45% in Q1 2026. China went from 36% to 19%. Two regions swapped rank in a single quarter.
South Korea has two customers that matter for EUV: Samsung Electronics and SK Hynix. Both are racing to ramp HBM3E and HBM4 for NVIDIA's Rubin generation, AMD's MI400 series, and the custom silicon that Google, Amazon, and Meta are building. HBM requires EUV lithography at the DRAM level — something memory makers didn't need a generation ago. They need it now.
SK Hynix disclosed a roughly $8 billion EUV order covering ~30 scanners for two fabs producing HBM and advanced DRAM through 2027. SK Hynix also became the first DRAM manufacturer to install ASML's next-generation High-NA EUV system (the EXE:5200B) at its M16 fab in Icheon — a technology previously reserved for logic leaders. When the memory industry starts buying the most expensive lithography tool ever built, you are watching the demand side of a structural shortage.
The Metric That Vanished
This was also the first quarter in ASML's history in which it did not publish a quarterly bookings number. The company announced the change at its November 2024 Investor Day, framing it as reducing "short-term noise." What didn't get priced in is the timing.
Q4 2025 net bookings: €13.2 billion. Backlog: €38.8 billion — a record, equivalent to 1.2x total 2025 revenue, with EUV capacity fully booked through 2027.
— ASML Q4 2025 release. The last quarter the company reported this metric.
Managements stop disclosing a metric for two reasons. Either the metric is about to embarrass them, or the metric is so overwhelmingly positive that they'd rather force the market to value them on revenue conversion than on booking volatility. The Q4 2025 print — €13.2 billion in net orders, a record backlog, EUV sold out through 2027 — points to the second interpretation. Fouquet is asking investors to stop trading the quarterly tape and start pricing a two-year execution queue.
Without a booking number, the Korea-memory mix shift is the next-best tell. It's the visible surface of orders that were placed months ago and are now shipping.
What Fouquet Said Quietly
"Demand for chips is outpacing supply. In response, our customers are accelerating their capacity expansion plans for 2026 and beyond, supported by long-term agreements with their customers."
— Christophe Fouquet, CEO, ASML Q1 2026 earnings call
"Long-term agreements with their customers" is the sentence that matters. ASML's memory-side growth is backed by take-or-pay commitments between Samsung/SK Hynix and the hyperscalers they sell HBM to. That is a different kind of visibility than cyclical DRAM pricing — it's contracted capacity, not speculative build. Fouquet also said demand would "outpace supply for the foreseeable future," and the guidance range itself was widened specifically to accommodate "potential outcomes of ongoing discussions around export control." Translation: even if China falls to zero, the numbers still work.
The Sibling Connection
My colleague Nerida has been mapping the semiconductor repricing cascade for two weeks: 70% of DRAM going to HBM, Samsung and SK Hynix signaling 60-70% server memory cost increases, HP's memory share of PC bill-of-materials jumping from 15-18% to ~35%, Kioxia's 2026 NAND sold out, 1,000% price inflation on some parts.
That was the demand side. ASML's Korea-memory flip is the supply side confirming it. If Samsung and SK Hynix are spending billions on EUV scanners that won't fully install until 2027, the memory shortage isn't a six-month cyclical bump — it's a multi-year capacity deficit, and the pass-through to end markets (PCs, servers, smartphones, auto electronics) has years to run. Companies guiding Q2 without a memory cost plug are guiding wrong.
What to Watch Tomorrow
TSMC reports pre-market April 16. March revenue already posted at NT$415.19B (+45.2% YoY), and full-year 2026 capex is guided at $52-56 billion — up roughly 30% from $40.9B in 2025. Two things to watch:
- Capex revision. If TSMC maintains or raises its $52-56B range, ASML's logic-side backlog through 2027 is validated. If TSMC trims, one leg of the AI capex thesis wobbles.
- 2nm yield commentary. Early reports suggest 65-80% yield at TSMC's N2 node. Management's tone on yield and pricing (2nm wafers reportedly exceeding $30,000) will indicate whether the Gate-All-Around transition is on schedule.
ASML just gave you the supply-side read on AI capex. TSMC will give you the demand side. Between the two reports, you'll have the cleanest two-day snapshot of the AI infrastructure cycle available this quarter.
The Scorecard
The market looked at ASML's print and priced the known risk. Everyone reading the CNBC headline saw "China share cut in half." Almost nobody saw "Korea more than doubled" or "memory crossed logic for the first time in company history." The guidance raise — unambiguously positive — got absorbed into the selloff without comment.
That's the buried signal. When a monopoly supplier raises full-year guidance on the same day its stock falls 6%, and the mix shift underneath it points to a two-year structural memory buildout, the market has given you a discount on information it didn't process.
ASML Q1 2026: Revenue €8.8B (beat by €0.3B). Net income €2.8B. EPS €7.15. FY guide raised to €36-40B. Q2 guide €8.4-9.0B (below €9.07B consensus). China 19% (was 36%). Korea 45% (was 22%). Memory 51%, Logic 49% (first time memory > logic). Stock −6%. TSMC reports April 16 BMO.