ASML: The Selloff Is a Repricing of Expectations, Not a Break in Orders

The consensus story for ASML's summer selloffs is that its monopoly is cracking. On July 27 the stock fell 5.8% after reports that a Shanghai state-backed company had begun mass-producing homegrown chipmaking tools. On August 18 it dropped more than 4% on renewed China concerns. On Friday it closed down 2.2% at $1,696.16 while the S&P 500 lost 0.25%, the latest example of a stock that is up roughly 135% over the past twelve months falling hardest exactly when the AI trade that carried it wobbles. Both fears are real. One is aimed at the wrong business, and the other is a complaint about the price, not about the orders.

ASML effectively operates two businesses, and the two threats hit different ones.

Two markets, one stock

The first business is extreme ultraviolet lithography — the only class of machine that can pattern the most advanced chips, and ASML is the only company that produces it in volume. The second is mature DUV: older-generation machines, including the immersion tools China has been buying for years and now hopes to manufacture itself.

The Chinese machine targets only the second market, and the second market was already closing on its own. The Shanghai tool is built by Aishengna, a state-backed firm founded in 2023 that absorbed teams from SMEE and Yuliangsheng and carries on the SMEE SSA800 project. It is roughly the equivalent of an ASML immersion machine from 2008, aimed at 28-nanometer production, about four generations behind ASML's current line, and still dependent on imported optics and light sources for its most critical parts. Aishengna's output targets are about five units this year and roughly twenty next year, against an ASML DUV-immersion capacity near 130 units this year.

ASML's two markets	Machines	The real threat	State of the order book
Advanced logic and memory: TSMC, Samsung, Intel, SK Hynix, Micron	EUV	AI-capex wobbles	2027 EUV orders near sold out; memory revenue guided up 75% in 2026
Mature nodes, concentrated in China	DUV immersion	Homegrown Aishengna tools	China was ~14% of system sales; export controls had already closed most of it

China was about 14% of ASML's new-system sales in the June quarter, and ASML guides it to roughly 20% of full-year 2026 sales. Those sales were already deflating under export controls, which is why the market lost so little of the company's actual profit stream on the announcement. As SemiAnalysis put it after the July report, a tool ASML cannot legally or physically supply, being built locally, does not subtract from a sold-out order book.

The order book says something different

A week and a half before the China report, ASML raised its 2026 sales forecast to €43–€45 billion — the second raise of the year, up from a €36–€40 billion range in April. It now expects memory-segment revenue to grow 75% in 2026, and it guided the September quarter to €11–€12 billion of sales, a step change from €9.3 billion in the June quarter, when gross margin was 54%. Management said it is close to receiving all orders for next year's EUV systems. South Korea, home to the memory makers buying EUV for HBM, was 43% of June-quarter sales.

The June-quarter beat itself came disproportionately from Installed Base Management — services and upgrades that brought in €2.76 billion, about €300 million above guidance. That is the part of the model that compounds with the installed base even if new-system sales slow, and ASML described order intake for the first half as extremely strong. By every reported measure this is the opposite of a demand-constrained quarter. The China story is a threat to a business the controls had already emptied, being priced against an order book that is as full as it has ever been.

The mechanism

Why the stock keeps falling more than the market is not difficult to isolate: it is the highest-concentration public bet on AI chip capex, and it trades like one. After the run, ASML changes hands at roughly 53 times trailing earnings and about 41 times EBITDA, on a market capitalization near $650 billion. When the AI trade wobbles — as it did around Nvidia's August report — the multiple that already contains years of assumed growth has the furthest to fall. On July 28, analysts at J.P. Morgan called the selloff overblown, and the stock took about a week to reclaim its pre-announcement level. That is what a repricing looks like: the headline supplies the occasion, the multiple supplies the move.

The signal that would change the read

The honest risk is not China; it is whether the AI build-out is paying for itself. Nvidia's August-quarter report, for all its size — roughly $96 billion of revenue and a near-$108 billion forecast for the current quarter — carried details that warrant attention. An analysis of the report noted hyperscale customers' revenue grew only about 13% sequentially against 25% growth elsewhere, as Nvidia stretched payment terms, with receivables reaching $63 billion and days sales outstanding rising from about 45 toward 60 days, against roughly $581 billion in supply commitments. Nvidia's CFO put the cloud-industry backlog above $2 trillion, which is the other side of the same ledger: committed demand, but demand increasingly carried by vendor financing and supply agreements rather than end-customer purchase orders.

ASML's guidance has been raised twice on the strength of that order flow. A capex cut at a hyperscaler, or a pushout at a foundry or memory maker, would show up first in ASML's quarterly bookings and backlog — not in the Chinese press. The key issue is not whether China learns to build a 28-nanometer DUV tool. It is whether the customers writing ASML's orders keep writing them in 2027 and 2028, when the company is planning another 30% expansion of both EUV and DUV capacity. Until a booking shows a break, this is a stock repricing its own expectations — and a stock carrying the full AI trade will keep doing that on every wobble.