ASML Holding N.V. ASML

Key data

The business

ASML builds the machines that print circuitry onto silicon wafers, and it is the only company on Earth that makes the extreme-ultraviolet (EUV) machines needed for the most advanced chips. Every leading-edge processor and every high-bandwidth memory stack that feeds an AI accelerator is patterned on its tools. The customer list is short: TSMC, Samsung, SK Hynix, Intel, Micron. A single EUV system sells for hundreds of millions of euros and takes months to build, ships in crates, and takes weeks to install. The moat is thirty years of physics nobody has replicated: the light source, the Zeiss optics, and a supply chain assembled over decades. About a third of sales now come from servicing the installed base, a recurring stream that softens the swings in machine orders. What you are buying is a genuine monopoly on the single most important tool in the chip industry, sold to five customers who cannot make advanced chips without it.

The numbers

The record is a cyclical business riding a secular ramp.

Revenue compounded 15% a year from 2021 to 2025 and EPS about the same, with buybacks adding the difference. But the line was not smooth: 2022 and 2024 were flat-to-down cycle years, 2025 reaccelerated to +16%, and 2026 is guided to €43-45B, roughly +35%, the steepest jump in the company's history and pure AI pull. Gross margin ran 53.5% in the first half and is guided to 54-56% for the year.

Earnings troughed at $6.41 in Q3 2025 and have climbed since; three of the last four quarters cleared expectations, with Q4 2025 the lone shortfall on softer shipments. Q2 2026 delivered €9.3B of sales and €2.9B of net income, and Q3 is guided to €11-12B, a sharp step up.

The mix tells the real story. South Korea, where the HBM memory for AI is made, jumped about 60% and is now the single largest market; Taiwan grew 17%. China fell to €2.9B, about 16% of sales from 24% a year earlier, as export controls bite. The worry from a month ago, that China's shrinking mix would go unoffset, broke the other way: Korea's surge more than covered the China decline, and total sales still rose 17%.

At $1,714.88 you pay about 53 times the last twelve months' earnings, against a typical band of 30 to 38 times over the past decade and a prior peak near 49 times, so this is a record multiple. It only holds together on the forward: consensus has earnings roughly doubling to about $68 a share by 2028, which puts the stock near 25 times those numbers. The variant here is that the China cliff everyone fears is largely already in the numbers, not ahead of them; what actually decides the outcome is whether Korea's AI-memory orders stay hot, and the prints that settle it are quarterly bookings and the Korea-versus-China mix.

Management

Insiders neither bought nor sold in the past twelve months, so there is nothing to read in either direction. The tell is capital return: after two light years, ASML repurchased €6.0B of stock in 2025 and paid its dividend, which halved cash to €6.7B by mid-2026 from €12.9B at year-end, though debt stays negligible (debt-to-equity 0.09) and the balance sheet is still roughly net cash. R&D ran 13.7% of sales, the spend that keeps the monopoly intact, and management guides conservatively and has generally cleared its own bar. Executive pay against profit was not disclosed in this pack.

How it fails or surprises you

The AI-memory ramp overshoots (right tail). Consensus has sales at €43-45B this year and earnings doubling by 2028. If Korea's HBM buildout and the first High-NA installs keep the order book climbing into 2027, the "peak" is not a peak, 2028 estimates prove low, and the 25x forward multiple looks cheap in hindsight. First sign: second-half 2026 bookings holding above expectations.

A booking air-pocket. China is capped near 16% of sales by export rules and Korea is now doing the heavy lifting. If AI-memory spending pauses even a quarter while China stays restricted, orders can fall fast in a business where each machine is worth hundreds of millions. Watch quarterly net bookings and the Korea line together.

You are paying a peak multiple on peak earnings. 53x trailing, above the 12-year high, for a business whose 2026 sales jump 35% off an AI wave. The fact this memo explains least well is why anyone pays a record price at what could be a cyclical top; if 2027 bookings roll over, the earnings and the multiple compress at the same time.

Closing thoughts

The business does not break: it is a monopoly with a net-cash balance sheet, and the only real question is whether you overpaid at the top of a cycle. The fatter tail is the upside, because AI demand has repeatedly run ahead of forecasts and ASML captures it with almost no chance of losing the socket. The risk is timing and price, a de-rating from 53x if orders stall, not impairment. If the downside linchpin breaks you give back perhaps a third as the multiple normalizes; if the upside lands, 2028 earnings clear consensus and today's price looks ordinary.

The bet is still that the world keeps building AI chips, and anyone making the most advanced ones has to buy ASML's machines, with Korea and Taiwan more than covering what China loses to export rules. What breaks it is a stall in AI-memory orders while China stays capped. The one pair of numbers that tells you first is quarterly net bookings and the Korea-versus-China revenue mix. If those two turn down together for two quarters running, the record multiple has nothing left to stand on.