The Pickaxe Maker Says The Buildout Is Booked Through 2027

ASML reported Q2 net sales of EUR 9.3 billion with a 54% gross margin and EUR 2.9 billion net income, beating its own guidance, and raised full-year 2026 to EUR 43-45 billion (its second hike this year) with a Q3 guide stepping up to EUR 11-12 billion. For 2026 it guided EUV revenue +45%, advanced foundry logic +25%, installed base +30%, and memory revenue +75%. CEO Fouquet said DDR and HBM pricing "indicates a clear need for additional supply," translating into accelerated customer capacity plans, and that logic customers are ramping 2nm "as aggressively as possible" while starting to look at 1.4nm. On visibility: he said end-market demand has motivated customers to increase capex and accelerate their plans, and that their own longer-than-usual visibility into their customers is leading them to commit further out. ASML is close to receiving all the EUV orders it needs for 2027 even while adding ~30% capacity, has a large number of 2028 orders in hand, and is investigating another +30% for 2028. High-NA reached an Intel production milestone. This lands against a SOX that had fallen ~16% from its June record on "peak AI" fears; Bernstein and BofA raised targets to $2,623 and $2,345 into the print. The stock rose only ~2.9%.

This is the cleanest rebuttal available to the "AI capex is about to roll over" case, delivered from the chokepoint of the entire supply chain. Orders booked years out and a second full-year raise say the buildout is not a one-quarter sugar high, and the memory line does double duty: 75% memory-revenue growth plus "a clear need for additional supply" is independent, upstream corroboration of the memory tightness the Charizard thesis rests on, seen from the equipment side rather than the operator side. But the honesty is in the tension, not the confirmation. ASML confirming that customers are committing capex through 2027-2028 argues directly against the Layer 3 / Layer 4 bear case that the monetization gap forces an imminent capex cut — you cannot hold "ASML's 2027 is sold out" and "hyperscalers slash capex in 2026" at the same conviction, and this print tilts against the cut. The counterweights: bookings are lagging multi-year commitments that reflect last quarter's confidence, not today's; a nearly-fully-booked 2027 reads equally as the top-tick of the supercycle; and it sits unresolved against the IBM double-ordering worry and Zuckerberg's own admission — if enterprises are panic-ordering memory, hyperscalers may be over-committing litho. The muted 2.9% reaction on a beat-and-raise says the market already assumes a lot of this.