ASML earns more from the leading edge while shipping fewer machines into it

The only company that makes the machine

ASML shipped 48 extreme-ultraviolet lithography systems in 2025.¹ Every other equipment maker on earth shipped none, because no other equipment maker knows how to build one. ASML is the sole producer of EUV, roughly 100% of a market it has entirely to itself, and its share of lithography overall runs above 90%; Nikon and Canon, the two firms that once competed with it, are confined to the older deep-ultraviolet and i-line tools that pattern mature chips.² EUV is the only method fine enough to print the smallest features on leading-edge logic and memory, which is another way of saying that the processors inside every recent phone, and the accelerators inside every AI data center, exist because one Dutch company can aim a kind of light nobody else can.³

A monopoly on a step every advanced chip must pass through is why the market prices ASML as the toll collector of Moore's Law. The company earns from that position in two separate ways. It sells the machines, €24.5 billion of systems in 2025. And it runs an annuity on the machines already in the field, €8.2 billion of service and upgrades that year, up 26%.³ The toll is real and it is growing: second-quarter 2026 revenue reached €9.3 billion at a 54.0% gross margin, and management raised full-year guidance to €43–45 billion.⁴

Two revenue engines running on different clocks

The system sale and the installed-base annuity keep different time. A system sale is lumpy and cyclical: a fab commits hundreds of millions of euros when it is adding capacity and goes quiet when it is not. The annuity is the opposite. Once a machine is bolted to a cleanroom floor it needs decades of service, spare parts, and field upgrades, and that revenue accrues whether or not the customer is buying anything new. In 2025 the two engines pulled in opposite directions within the systems line itself: EUV system sales grew 39% to €11.6 billion, while the older DUV systems fell 6% to €12.0 billion, dragged down by a cooling Chinese market.⁵

The tell in those numbers is the crossover. The annuity grew 26% in 2025 against 12.4% for system sales, so the maintenance base is expanding faster than the machines being sold into it.³ System sales lead the cycle, but every system shipped becomes a future annuity obligation, so over time the steadier installed-base revenue grows into a larger and more reliable share of the whole. It is the part of ASML that does not care what the order book does next quarter. The table below is the full-year shape, and the two engines sit side by side in it.

Revenue climbed while shipments fell

Read the table's last few rows together and the toll-taker story acquires a complication. Revenue rose 15.6% in 2025, but the company shipped 327 lithography systems against 418 in 2024, a drop of nearly 22%.⁵ More money on fewer machines has one arithmetic explanation: the average system got far more expensive. Net system sales of €24.5 billion across 327 units works out to about €74.8 million of net system sales per litho system, up from €21.8 billion across 418 units, or €52.1 million, in 2024. The average price of an ASML system rose more than 40% in a single year, entirely on mix, as EUV and its priciest configurations displaced cheaper DUV volume.

A toll collector that raises the toll while fewer vehicles pass is not obviously a growth story; it can be a cyclical one wearing a secular costume. The cycle is visible in the order book. Net bookings peaked at €30.7 billion in 2022, collapsed to €18.9 billion in 2024, then recovered to €28.0 billion in 2025.⁵⁶ That is a swing of the kind semiconductor-equipment demand has always produced, and it sits directly underneath the smooth revenue line the headline number shows. The secular case, more compute forever, is real. The cyclical case, that fabs order in waves and ASML rides them, is also real, and 2025's falling unit count is what the second one looks like when the first one is winning on price.

The demand tell just went dark

For a company whose revenue lags its orders by a year or more, the order book is the number that matters, because it reads demand before demand shows up in sales. ASML's fourth quarter of 2025 was a statement of intent: €13.2 billion of net bookings in three months, of which €7.4 billion was EUV, taking year-end backlog to €38.8 billion.¹ That single quarter was most of what made 2025's bookings recovery look like a boom.

Then the tell went dark. Starting with the first quarter of 2026, ASML stopped publishing quarterly net bookings, on the reasoning that large orders arrive unevenly and a single lumpy quarter can distort the read on underlying momentum.⁷ The July 2026 second-quarter release and its investor deck carry no bookings figure at all; management now characterizes order intake only in words, calling it extremely strong.⁴⁸ The timing is its own data point. The most direct instrument for judging whether the AI order wave is durable or a pull-forward was retired at the exact moment that question became most valuable to answer. The last hard reads, €28.0 billion of 2025 bookings and €38.8 billion of backlog, are now the anchors an investor has to work from, and everything since is adjective.

China's share is being cut in half

The clearest exposure is geographic. As US-led export controls closed in on advanced tools, Chinese fabs bought every unrestricted DUV machine they could, and ASML's China revenue share climbed from 14% of sales in 2022 to 29% in 2023.⁹ It kept climbing, to about 41% in 2024, then began to reverse, to roughly 33% in 2025.¹⁰ For 2026 ASML guides China to about 20% of sales, a level management reaffirmed on the second-quarter call, describing the pull-forward as normalizing and the remaining Chinese demand as concentrated in domestic-led logic.¹¹

One caution about the figure. ASML's geographic split lives in chart images inside its decks rather than in a table that extracts cleanly, so these percentages are ASML's own disclosure as reported by TrendForce and the South China Morning Post, on a share-of-total-net-sales basis; third-party aggregators that use a different denominator show a lower China number for 2024, near 36%. The trajectory is what matters and it is not in dispute: China roughly tripled its share, then is being cut in half. That is the cliff after the pull-forward. It falls almost entirely on the DUV engine, which explains the 6% DUV decline in a year when EUV grew 39%, and it removes a chunk of demand that inflated the last two years' comparisons.

The €350 million bet on High-NA

The other exposure sits inside the growth engine itself. ASML's next EUV generation, High-NA, lists at around €350 million a system, roughly double the €170–180 million of the current Low-NA EUV machines, and its order book so far is a handful, on the order of 10 to 20 units to Intel and SK hynix, ramping toward perhaps 20 a year by 2028.¹² The technology is genuinely in a fab. Of the 48 EUV systems ASML shipped in 2025, four were High-NA, and by year-end customers had run more than 400,000 wafers on them; Intel says it has qualified High-NA on 18A product layers and is using it in production.³⁸¹¹

The bet is not whether High-NA works. It is whether customers adopt it on ASML's timeline, and the strongest independent analysis says they may not. SemiAnalysis argues the economics favor the older tool: Low-NA EUV run as a double exposure delivers a lower cost per wafer than a High-NA single exposure, because High-NA throughput is limited by the light dose each wafer needs. In their words, "despite requiring twice as many wafer passes through the scanner, the lithography costs are lower than a high-NA single exposure."¹³ On their modeling the Low-NA cost advantage widens through the 2nm and 1.4nm nodes and only reaches parity around the 1nm generation, near the end of the decade, when scaling shifts vertical. They note that ASML's own High-NA pitch quietly moved from "cost advantage" to "process-complexity reduction," which is the language of a tool that has to be justified on something other than price per wafer.

Both pictures are true at once. High-NA is live in one Intel fab, and the volume-adoption decision at the larger buyers, TSMC and Samsung, remains open, with the independent read pointing to deferral toward the 1nm node. A machine ASML is only investigating building more of, weighing a 30% increase in EUV capacity for 2028, is a machine its biggest customers may choose to buy slowly.¹¹

The monopoly holds; the case around it is thinner

The monopoly is not in question. ASML will make every EUV machine sold for the foreseeable future, and the installed-base annuity underneath it compounds regardless of the cycle, which is the sturdiest part of the case and the part that most resembles the toll-taker the market pays for. Its own 2030 model reaches for €44–60 billion of revenue at 56–60% gross margins, on a semiconductor market it expects to pass a trillion dollars.⁵ The near term supports the direction: guidance was raised twice in 2026, to €43–45 billion, on AI demand.⁴

What the price actually pays for is thinner than the monopoly makes it sound, because three of the four things that carry the story from here have turned the wrong way at once. Unit volume is falling, and the revenue line is held up by a mix shift that cannot repeat every year; the average system cannot rise 40% annually forever. China, a fifth to two-fifths of sales across the cycle, is being cut in half as the export-control pull-forward unwinds. And High-NA, the generation ASML is tooling up to sell, faces an independent cost case that its largest customers can act on by simply waiting. Against all three, the one instrument that would show whether the order wave is durable, the quarterly bookings print, was switched off this year. The toll is real, and it is still being paid. The question the buyer of the stock is now answering, with less disclosure than a year ago, is whether it is being paid by more customers or merely at a higher price by fewer.