A lithography machine is ordered years before it is needed, built to a customer's process, and installed by a crew that flies in with the tool. There is no spot market, no second-hand bargain that arrives in time, and no way to buy capacity you did not reserve.
That is why the interesting number in a chipmaker's earnings season is often not profit. It is the queue.
ASML published its third-quarter figures on 14 October, and the queue is the constraint.
Key takeaways
- ASML reported €11.7 billion of third-quarter net sales, a 57.5% gross margin and €8.4 billion of net system sales.
- China took 23% of system sales by ship-to location in the quarter; ASML guides to roughly 20% for the full year.
- EUV was 49% of system revenue and immersion DUV 39%; logic was 54% of the mix and memory 46%.
- The chief executive said supply planned for 2027 and 2028 will not meet demand; some customers are asking about 2029 and 2030, which is interest, not orders.
- China cannot buy EUV at all, so its expansion runs on immersion DUV and multi-patterning — a structurally more expensive wafer.
The quarter, in one paragraph
Net sales of €11.7 billion (≈ RMB 87.6 billion) against €9.3 billion in the second quarter and roughly €7.5 billion a year earlier. Gross margin 57.5%, operating margin 44.4%, net income equal to 42.5% of sales. Installed-base management — service and field upgrades — added €3.3 billion, more than a quarter of the total.
The company guided fourth-quarter sales to €14.2–15.2 billion and full-year 2026 to €44–45 billion at a gross margin near 56%. The company expects full-year growth of about 36% on 2025, with EUV growing faster than the rest.
Where the wafers are going
The regional split is the part Chinese buyers should read twice. Taiwan took 32% of system sales, South Korea 31%, China 23%, the United States 13%. China's share was 42% a year earlier. End-use split almost evenly between logic at 54% and memory at 46%, the highest memory share in years and a direct consequence of the HBM and server-DRAM buildout.
China's 23% is a reminder of something easily lost in export-control coverage: the restrictions are narrow at the top and porous lower down. EUV systems are not licensed for export to China and have not been for years. Immersion DUV is a different product with a different licence, and it is what Chinese fabs actually run.
That is why ASML's chief financial officer argued publicly against a proposal in the United States Congress to tighten DUV restrictions, calling it a bad idea that would only accelerate China's own tool programme, Reuters reported from the earnings call. A supplier that cannot build fast enough does not usually fight for a customer. It does when the customer is a fifth of revenue and the alternative is a domestic competitor funded either way.
Why the queue is the story
On the call, chief executive Christophe Fouquet said supply "as currently planned for both 2027 and 2028 will not fully meet the demand of our customers". ASML had planned 30% more capacity in 2027 and is looking for another 30%-plus in 2028. Its new Eindhoven plant will not finish its first phase until 2029. Fouquet added that some customers are expressing interest in ordering for 2029 and even 2030. Interest is not an order, and ASML stopped reporting quarterly bookings this year, so none of it appears as backlog.
For anyone modelling Chinese AI capacity, that admission does more work than any benchmark. A fab's output in 2028 is not set by its cleanroom shell, its power contract or its design team. It is set by which tools were booked in 2025 and 2026. Orders placed today land after the current Five-Year Plan ends.
The corollary is that China's accelerator supply in 2027 and 2028 is largely already determined. Huawei can design a faster Ascend; it cannot conjure the scanners, and every additional wafer of leading-edge logic competes for the same finite tool output as TSMC's and Samsung's expansions.
The DUV tax
There is a second-order effect that rarely makes it into market-share charts. A fab that cannot buy EUV reaches a given density by exposing the same layer several times with immersion DUV. Each extra pass costs a scanner hour, a mask set, and yield.
So China's leading-edge capacity carries a structural cost premium per good wafer, and it consumes more of the scarce machines per wafer than a competitor using EUV. In a shortage that matters twice, and the constraint compounds.
This is the honest version of the self-sufficiency story. China's fabs are running and its equipment makers ship more each year, but multi-patterning is a tax that does not disappear when a domestic etcher improves.
Honest limitations
All figures come from ASML's own quarterly release and the earnings call as reported by Reuters and by Chinese financial media; I have not seen the full transcript. The 2029–2030 customer interest and the capacity figures are management commentary, not contracted backlog.
A 23% regional share can also mean a handful of large orders landing in one quarter, which makes single-quarter comparisons noisy. And the claim that China's capacity is "already determined" is an inference from tool lead times, not a measurement — Chinese fabs hold inventories of scanners bought during the 2022–2024 stockpiling wave, and nobody outside those fabs knows how much idle tool capacity remains. Nor does this article assess whether Chinese fabs can substitute domestically built DUV tools at volume; that is a separate question with much less public data.
What readers can do now
- If your supply model assumes Chinese leading-edge capacity grows on demand, replace that assumption with tool delivery dates. The two diverge by two to three years.
- Track ASML's regional split each quarter rather than annually. Single quarters swing on a few large orders, but a falling China share sustained over four quarters is the earliest public signal of a capacity slowdown.
- For cost modelling, assume a multi-patterning penalty on any Chinese wafer below the DUV limit, and ask suppliers whether their price assumes it.