ASML Said It Cannot Build Enough Machines for 2027 or 2028. The Shares Fell Anyway.

Here is the sentence that matters, and it arrived in the third bullet block of a quarterly press release: "we now expect that our supply as currently planned for both 2027 and 2028 will not fully meet the demand of our customers."

That is Christophe Fouquet, ASML's chief executive, in the company's Q3 2026 results published this morning in Veldhoven. ASML is the only company in the world that produces extreme ultraviolet lithography machines, the tools TSMC, Samsung, Intel and SK Hynix need to print the most advanced layers of a chip. There is no second supplier to call.

The shares fell anyway.

The quarter that should have ended the argument

The fundamentals were not ambiguous. Record total net sales of EUR 11.66bn, against EUR 9.33bn in the second quarter and EUR 7.52bn a year earlier, according to RTT News. Gross margin of 57.5%, up from 54.0%. Net income of EUR 4.96bn. Basic earnings per share of EUR 12.95, against EUR 7.59 in Q2. ASML sold 113 new lithography systems in the quarter, up from 86. Its installed base business, the service and upgrade line, brought in EUR 3.29bn. End-of-quarter cash and short-term investments stood at EUR 12.74bn, nearly double the EUR 7.58bn it held three months earlier.

Guidance went up for the third time this year, as the Wall Street Journal framed it. Full-year 2026 sales are now EUR 44bn to 45bn, with the bottom of the range lifted from the EUR 43bn ASML guided in July, and a gross margin around 56% instead of a 54% floor. Fourth-quarter revenue is guided to EUR 14.2bn to 15.2bn, above the EUR 14.19bn that analysts polled by LSEG had expected, per Reuters.

Two blemishes on an otherwise clean sheet. Revenue of EUR 11.66bn came in fractionally below LSEG's EUR 11.72bn consensus, so the beat was in profit rather than the top line, as CNBC reported. And the quarter leaned on the high-margin installed base business, which is upgrades and service on machines already in the field, not new tools shipped into new fabs. Strip out those EUR 3.29bn and net system sales land at roughly EUR 8.37bn.

Where the ceiling actually is

This is the part worth sitting with, because supply constraints in this industry are usually somebody else's problem. A foundry. A memory maker. A power utility. This one belongs to ASML.

Take the numbers the company has given the market about its own output. On its Q2 call in July, management said it would ship about 65 low-NA EUV systems in 2026. It committed to lifting capacity by around 30% in 2027, which points to roughly 85 tools. For 2028 it is investigating a further 30%, which would land near 110 machines, but it has not committed to that. The company said as much in the footnote to this morning's release.

Now price the lead time. Each low-NA tool costs around EUR 200m and takes more than a year from order to delivery, per Bloomberg's reporting carried by The Edge Malaysia. ASML has said it is working to cut the time needed to build and test its lithography tools by about a third. None of that compresses into a quarter. The machines that will print 2028's chips are being ordered now, assembled over the next twelve to eighteen months, and installed by field teams that have to be hired and trained first.

The detail that landed hardest, then, was not about customers at all. Fouquet said some of them are expressing interest in placing orders for 2029 and even 2030. Reuters quoted him saying "a lot of our discussions with our customers now are shifting in fact beyond 2027." The backlog has already done its work for the near term: as of July, ASML was close to being fully covered with orders for 2027 low-NA EUV. The 2027 question is not demand. It is output.

What ASML stopped telling you

This quarter's demand signal is harder to read than it looks, for a structural reason. ASML no longer publishes quarterly net bookings. It stopped after 2025, arguing that order intake is lumpy and that its own guidance carries more information. This is the third consecutive quarter without the figure.

That leaves capacity plans and management language carrying the weight a bookings line used to. It also means the most important number in this morning's release is a production target, not an order figure. A production target can slip on hiring, on test time, on factory floor space, in ways a booked order cannot.

Analysts are split on whether the plan is too conservative. UBS thinks the pace of capacity increases could go beyond 30% in both 2027 and 2028. Morgan Stanley models more than 90 tools in 2027. Bernstein's David Dai told Bloomberg the company may hold guidance now and raise it three months later, because "demand will be much clearer then." And there is a reason for the caution. Josep Bori, a global equity analyst at Aviva Investors, said both ASML and TSMC look conservative about ramping supply because "they've been burned before."

Half a memory story

The other thing worth noticing is who is buying. In the second quarter, memory accounted for 49% of ASML's net system sales, essentially level with logic. A year earlier that split was 69/31 in logic's favour. ASML expects memory-related system sales to grow more than 75% this year.

So the toolmaker's order book is now tied to the same DRAM and HBM pricing cycle that has memory makers booked out into 2027. TSMC reports on Thursday, which is the next datapoint on whether the money behind those capacity plans is actually holding.

The market's counter-argument

The scepticism that met this print was not really about ASML. It was about whether its customers can pay for what they want to buy.

On Monday, Broadcom chief executive Hock Tan told a conference in Singapore that many investors "have no clue what they are investing in" and set "very extreme expectations, only to find out that it's almost like they are in a casino." Intel's Lip-Bu Tan struck a similar note, warning that the rapid AI buildout could worsen existing hardware shortages, with memory among the first areas to lose steam. Korean memory names had already slumped to two-month lows by then: Samsung and SK Hynix were down 15% and 13% respectively from Monday's open in Korea, per Morningstar, after Chinese memory maker CXMT said it plans to roll out a new 4F2 DRAM structure by the end of 2026. CXMT has not demonstrated mass production of it, and the gap between a technical announcement and actual fab investment is the difference between a headline and an order.

Layer on top a JPMorgan trading desk note published last week, which found ASML being used as a short to fund purchases of other semiconductor names, along with the line that "a lot of the bull case is now well understood."

The market's question has changed shape. It is no longer whether the demand is real, because ASML just answered that. It is whether the demand can be delivered and whether the customers can fund it through the cycle. Those are different risks, and only the second one explains a 4% drawdown on a record quarter. RTT News reported the shares down about 2.5% in overnight Nasdaq trading after the release, extending a 1.86% decline the previous session, while the Wall Street Journal's quote page showed ASML at minus 4.20%.

One more thing to watch: China. ASML held its forecast that China accounts for about 20% of total 2026 net sales, a figure that includes service revenue, per CNBC. That service revenue is the high-margin part. RFC Capital Research, citing a 26 September count by the Center for Technology & Statecraft, put 343 immersion DUV systems inside Chinese-owned fabs, around 270 of them ASML's NXT:1980i. New shipments to China are already restricted. Restricting service on that installed base would be a different kind of problem.

One caveat on timing: as of 06:30 UTC today, ASML's investor call, scheduled for 15:00 CET, had not yet happened, and the company's full US GAAP statements and investor presentation were still to come. Nothing here should be read as call commentary.

What it means

This is not a demand problem, and that is the interesting part. The constraint ASML is describing is time. Its own assembly and test throughput, its own hiring, its own factory floors. In a buildout where the loudest argument has been about kilowatts and capital, the binding limit on the next two years may turn out to be how many EUR 200m machines one company in the Netherlands can physically finish and install.

For anyone planning silicon or hardware capacity into 2028 and beyond, that reshapes the planning problem. Delivery windows are set years out. Substitution options are limited, because there is no alternative supplier for the layer that matters most. And the difference between a capacity plan that survives contact with reality and one that does not usually comes down to how early the constraint was modelled, not how bullish the demand forecast was.

At DMC we work with hardware companies on exactly that: sourcing strategy, supplier lead-time mapping, and production ramp planning when the equipment you need is on somebody else's schedule. If you are stress-testing a roadmap against a supply chain that now runs two years deep, let's talk.