ASML — initiation. Paid by both sides of the memory war.

1. This morning’s print, and why it settles the setup

ASML reported second-quarter results at 07:00 CEST today. The quarter beat on both lines, and the beat came from the right place: Installed Base Management — the recurring service-and-upgrade revenue on the machines already bolted to fab floors — came in hotter than planned, which is a quality beat (high-margin, repeatable) rather than a shipment-timing fluke.

But the quarter was never the story. With bookings retired as a disclosure, the tell on a ~42× cyclical is the guide — and the guide is where the print got loud:

• Q3 2026: net sales of €11.0–12.0B at 55–57% gross margin. Off a €9.3B quarter, that is an 18–29% sequential step-up — guided, not hoped.
• Full-year 2026 raised to €43–45B at 54–56% gross margin — from €36–40B in April, which was itself a raise from €34–39B. Three raises in six months, this one worth +16% at the midpoint.
• Capacity is being committed, not just promised: CEO Christophe Fouquet announced a ~30% expansion of low-NA EUV capacity for 2027 (on roughly 65 units this year) and said the company is investigating another ~30% for 2028, with similar moves in DUV. You do not spend that kind of capital into a demand peak you can see.

One caution before anyone extrapolates the morning: the market saw most of this coming. The ADR rose 2.9% Tuesday into the print on breathless previews, options implied a ±8.4% move, and the actual reaction on the Amsterdam line was about +1%. A beat-and-raise that moves the stock one percent is a beat-and-raise that was priced. That doesn’t diminish what it did for the thesis — it retired the guide-cut scenario that was the whole near-term bear case — but it does mean this morning hands you confirmation, not a discount.

2. How we got here — the entry was the work

This initiation began as a refusal. In late June, with ASML at all-time highs around $1,971 and every preview calling it a no-brainer, we wrote an internal one-pager that concluded: wonderful business, wrong entry — the premier name to own on the next dip, with a target buy zone of roughly $1,550–1,700, because a 2.3-beta cyclical at ~42× earnings guarantees you an air-pocket eventually.

The air-pocket took four sessions to arrive. The July 1–2 memory crash — the Korean leveraged-ETF unwind we covered in the Micron unwind note — took ASML down 11% in two days, from the $1,989 record to ~$1,769, on no change in ASML’s own file. We bought the starter on July 8 at $1,752.87, inside the zone named before the fact. Same monopoly, 11% cheaper, and the discipline — wait for the beta to do its thing — is the entire reason the position starts green instead of underwater.

We flagged the position and the coming print in the July 6 bounce note and again yesterday, where we said ASML would settle the demand question this morning. It did. This note is the promised initiation, published with the answer in hand rather than the day before it.

3. What you own — the only monopoly in the AI trade

ASML supplies 100% of the world’s EUV lithography — the machines that draw the smallest features on advanced chips — and roughly 94% of lithography overall. It is the only maker of High-NA EUV, the ~$350M-plus next generation that sub-2nm nodes will require. Nikon and Canon exited the race a decade ago; recreating the position would mean recreating Zeiss’s optics, Cymer’s light source, and a thirty-year, ~$10B web of thousands of single-source suppliers. And the moat deepens on its own: mixing lithography brands inside a fab creates overlay and yield penalties, so every installed machine locks the customer in a little further.

Two structural details matter for our book specifically. First, memory is now 51% of ASML’s system sales — the DRAM and HBM makers are adopting EUV at accelerating layer counts, which puts ASML directly upstream of the HBM build-out we own through Micron and the memory basket. Second, China is 19% of system sales and falling (from 33% in 2025), concentrated in older DUV tools; EUV has never shipped to China. The most-cited geopolitical risk on this name is a ban on something that has never happened.

4. Paid by both sides of the memory war

Regular readers know the tension in our book. We are long the memory complex — and we have cut our Micron target three times, to ~$1,100, because the open question isn’t whether AI demand is real (Micron just committed $250B of capex to it) but what Micron keeps after the price war: pricing power, HBM4 share, margin through the down-leg.

ASML is how we stay long the build-out without underwriting that answer. SK hynix’s ~$28B raise is earmarked for fabs and the EUV tools inside them; Samsung’s record profits fund the same shopping list; Micron’s $250B does too. If Micron wins HBM4, ASML sells it the machines. If SK hynix wins instead, ASML sells it the machines. The toolmaker monetizes the capex, one layer insulated from the price war itself — which is why this morning’s print matters beyond ASML: a company that sees every fab’s order book just raised its own outlook by 16% and named Memory demand explicitly alongside Logic. The demand leg of the memory debate got its confirmation. The what-does-Micron-keep leg stays open, exactly where we left it yesterday.

The next data point comes fast: TSMC reports tomorrow morning, and the hyperscaler capex prints land July 22–31. ASML just set the bar for what “the build-out is intact” sounds like.

5. The numbers, in one table

Sit with that last row for a second. The 2030 revenue model ASML laid out for investors starts at €44B. The 2026 guide issued this morning is €43–45B. The bull case’s destination just became the current year’s consensus — which is either a reason to re-rate the out-years or a reason to ask what’s left for them. We think mostly the former (High-NA barely contributes yet), but that question is the honest core of the valuation debate below.

6. Valuation — the raise de-rates it through the numerator

Before the print, the fair criticism of ASML at ~$1,775 was: ~42× forward earnings for a mid-teens grower is rich, even for a monopoly. We wrote it ourselves in June. The print undercuts the premise, not the arithmetic: a year guided to €43–45B against 2025’s base isn’t a mid-teens story anymore, and a gross margin walking from 51–53% to 54–56% compounds the earnings effect. Consensus EPS has to revise up; on the raised numbers the forward multiple compresses while the price stands still. That is the polite way markets de-rate a quality name — through the numerator, not the price.

What we will not do is call it cheap. It isn’t. You are paying a full, fair, premium price for the single most un-disruptable franchise in technology at the moment its demand curve steepened. That is a defensible thing to pay for at starter size, which is what we own. It is not a back-up-the-truck price, which is why the add-zone below matters more than the rating word.

7. Bull / base / bear — reweighted off the print

Pre-print, we carried a 25% bear case built on a guidance stumble. That scenario is off the table for this quarter — you cannot cut a guide you just raised 16% — so the residual bear case is macro, not company-specific: a semis-wide “peak capex” de-rate that takes everything in the complex down together. Twelve-to-eighteen-month ranges from ~$1,775, probabilities ours, all figures estimates:

Probability-weighted: roughly $2,000, +13% from here — on top of a starter already +1.3% against its $1,752.87 basis. The skew improved this morning: the raise de-risked the downside tail, and the muted reaction means the upside wasn’t spent celebrating.

8. The honesty section — a lean-in, not a diversifier

Here is the criticism a good committee would raise, so we’ll raise it ourselves: you already own Micron, a DRAM basket, and a book that is overwhelmingly a bet on AI capital spending. Why does the next position need a 2.3 beta and a ~94% correlation to the same factor?

Because it’s the highest-quality expression of a bet we already hold with conviction — and for no other reason. ASML deepens our concentration; it does not hedge it. On the day the AI-capex thesis breaks, ASML goes down with the rest of the book, beta-adjusted and then some. We sized it accordingly (a starter), and the correlation-budget discipline that governs this portfolio says the next new dollar goes to the true diversifiers on our bench — the power name, the EDA name, the software name — not to more of this factor. Naming that trade-off out loud is the difference between a thesis and a mood.

9. Levels, triggers, and what changes our mind

Upgrade pressure (toward outright Buy): a pullback into the add-zone with the guide intact; or evidence the Q3 €11–12B is conservative (listen to today’s 15:00 CEST call for backlog and order-momentum language — the bookings replacement signal — and High-NA cadence).

Downgrade pressure (toward Hold/Trim): the 2027 capacity expansion meeting soft order commentary within two quarters (capacity into fading demand is how supercycles end); or the stock running 20%+ above the probability-weighted target on multiple expansion alone.

The watch list from here: TSMC’s print tomorrow · hyperscaler capex, July 22–31 · Micron’s next HBM4 share datapoint · the December High-NA shipment cadence.