NextFin News - ASML, the Dutch company that makes the only machines on Earth capable of printing the most advanced computer chips, said on October 14 that its planned supply for 2027 and 2028 will not fully meet customer demand. The warning came as the lithography giant reported record third-quarter sales of €11.7 billion, up 55% from a year earlier, and net income of €5.0 billion. Chief Executive Christophe Fouquet told investors the shortfall is driven by artificial intelligence.
"Our customers continue to increase their short- and medium-term equipment demand forecasts with us, driven primarily by their views on the size and duration of AI-related investments," Fouquet said.Some customers are now asking to reserve tools for 2029 and 2030. The message is stark: the AI buildout has outrun not just the chipmakers, but the equipment maker that sits above them all.
Record Quarter, and a Warning About the Next Two Years
ASML's third-quarter numbers were strong across the board. Total net sales reached €11.7 billion, toward the high end of the company's own guidance and just below the roughly €11.7 billion consensus. Gross margin came in at 57.5%, above the guided range, and income from operations more than doubled to €5.2 billion from €2.5 billion a year earlier. Net income of €5.0 billion includes a €509.7 million gain on the revaluation of ASML's stake in AI startup Mistral, so the cleaner read on the underlying business is operating income, which rose 110%.
The volume of machines tells the same story. ASML shipped 121 lithography systems in the quarter, up from 72 a year earlier, including 17 EUV systems — the most advanced kind — versus nine. Service and upgrade sales jumped 68% to €3.3 billion, a sign that existing customers are paying to squeeze more output out of the tools they already own rather than waiting for new deliveries.
Then came the forward-looking statement that moved the conversation. ASML had already told investors it planned to lift capacity by 30% in 2027 and by another 30% in 2028. Fouquet said that is no longer enough.
"Our supply as currently planned for both 2027 and 2028 will not fully meet the demand of our customers," Fouquet said,adding that the company is investigating ways to push 2028 capacity beyond the previously communicated 30% increase. In the results video he was more blunt:
"30% in 2027, 30% in 2028 will not allow us to meet our customer demand."Elsewhere he put it plainly: customers are asking for more tools every single day.
The demand signal extends well beyond the current planning horizon.
"In light of their demand visibility, some customers are expressing an interest in placing orders with us for 2029 and even 2030," Fouquet said.Customers are not merely complaining about tight supply; they are trying to lock in machines that will not ship for three or four years. That is not a cyclical inventory swing. It is a multi-year capacity reservation.
The market read the news as confirmation rather than a problem. ASML shares rose nearly 2% in New York premarket trading and about 3% in Amsterdam on the day. The stock is up roughly 97% over the past year, and the company now carries a market value of about €615 billion at a price-to-earnings multiple near 58.
Why the Bottleneck Has Moved Upstream
For most of the generative AI boom, the visible shortage was GPUs. Cloud providers fought over Nvidia accelerators, lead times stretched, and allocation became a strategic asset. The constraint then shifted to high-bandwidth memory, squeezing DRAM and NAND capacity so hard that smartphone and PC shipments fell as memory makers redirected output to data-center customers paying more.
ASML's warning moves the bottleneck one layer further upstream, to the machines that create wafer capacity in the first place. A GPU shortage can eventually be eased by reallocating wafer and packaging capacity. An EUV shortage affects the industry's ability to build that capacity at all. ASML's extreme ultraviolet scanners are the sole route to the smallest features on the processors that train and run large language models, and on the high-bandwidth memory stacked beside them. There is no second supplier. When the sole supplier says it cannot build fast enough, everyone downstream — TSMC, Samsung, Intel, Micron, Nvidia, and the cloud companies funding the buildout — inherits the constraint.
The balance sheet shows how far ahead customers are reaching. The long-term portion of money customers have paid or been billed ahead of delivery jumped €5.7 billion in the quarter to €9.3 billion. ASML no longer reports quarterly bookings, but the last disclosed figure was €13.2 billion in the fourth quarter of 2025, with a year-end backlog of about €38.8 billion. Prepayments for tools that will not arrive until 2028 are exactly where money for a 2028 delivery would show up.
The downstream commitments line up with the timeline ASML now says it cannot fully supply. Samsung and TSMC signed on in September to deploy ASML's newest High-NA EUV machines in their most advanced processes, with Samsung eyeing DRAM production by 2028 and TSMC planning high-volume advanced-node output from 2030. Intel became the first chipmaker to ship high-volume logic built with High-NA EUV in July, using select layers of its Core Ultra Series 3 processors on the 18A node. Those node transitions sit squarely inside the window ASML is struggling to cover. Any slippage in ASML's rollout could cascade into delayed node transitions at the moment AI chip designers are counting on smaller, denser transistors to keep performance scaling.
The Capacity Math: 30% Plus 30% Is Not Enough
ASML's expansion plan was already aggressive. The company said its 2026 Low-NA EUV production capacity is approximately 65 systems, and it is planning a similar expansion in DUV immersion capacity from a 2026 base of about 130 systems. Adding 30% in 2027 and another 30% in 2028 would take Low-NA EUV capacity to roughly 110 systems by 2028 — a figure JPMorgan analysts have flagged in market commentary, though ASML has not confirmed it as a formal target.
Even that would not be enough. The reason sits with ASML's customers, who expect that their supply will not meet demand at least until 2028, so they are adding factories faster. Memory makers are the clearest example. Micron reported net capital spending of $27.4 billion for fiscal 2026, against $13.8 billion a year earlier — roughly double. ASML said its foundry customers are adding capacity at the 3-nanometer node "in support of the latest generation of AI accelerators," and that memory makers are expanding because supply challenges are "driving up both DDR and HBM prices."
ASML is also growing faster than its U.S. equipment peers, whose tools etch, deposit, and inspect the layers that ASML's scanners pattern. In their latest reported quarters, ASML's revenue rose 55% year over year, compared with 30% at Lam Research, 25% at Applied Materials, and 15% at KLA. The gap is the signature of a chokepoint supplier: when the entire industry is racing to add capacity, the sole provider of the most critical tool captures the tightest part of the cycle.
There is a second lever ASML is pulling besides building more machines. Its productivity upgrade portfolio lets customers extract more wafer output from installed tools. The 68% jump in service and upgrade sales shows customers are using it. In a supply-constrained market, an upgrade that adds a few percent of throughput is worth paying for because it arrives months before a new scanner could.
This Is Structural, Not Cyclical
The critical question is whether this is a cyclical overordering that will reverse, or a structural shift that will not. Three pieces of evidence point to structural.
First, the demand is anchored to a multi-year infrastructure buildout, not an inventory restock. Customers are expressing interest in orders for 2029 and 2030. When customers reserve capacity that far ahead, they are planning for a technology roadmap — the 2-nanometer-class logic and advanced DRAM nodes that AI workloads will run on through the end of the decade.
Second, the supply side cannot respond quickly. ASML's machines are among the most complex products ever built, with hundreds of thousands of parts and a supply chain that spans continents. Ramping capacity takes years, not quarters. A 30% annual increase is already at the edge of what the company has said it can execute; going beyond that in 2028 means adding suppliers, factory space, and trained engineers on an accelerated clock.
Third, there is no substitute. ASML is the only company that makes EUV lithography systems. If a customer cannot get an EUV tool, it cannot make the most advanced chips. It can fall back on older deep ultraviolet machines with multi-patterning — more steps, lower yield, higher cost — but it cannot replicate EUV. That is the definition of a structural bottleneck: a constraint that cannot be arbitraged away by switching suppliers.
The counter-thesis deserves a hearing. Demand this hot invites front-loading and double ordering. If AI capital spending slows because returns disappoint, or if the hyperscalers pause after absorbing this year's GPU deliveries, some of those 2029 and 2030 expressions of interest could evaporate before they become firm orders. ASML still expects China to represent 20% of fourth-quarter revenue, and export controls on advanced tools to China remain a live geopolitical risk that could remove a large revenue stream overnight. And at nearly 58 times earnings, the stock prices in a long period of flawless execution. Any stumble — a High-NA ramp delay, a component shortage, a customer deferral — would be punished.
The strongest version of that counter-argument is that ASML's own customers may be manufacturing the shortage. If TSMC, Samsung, and the memory makers are all ordering extra tools to secure allocation in a tight market, the order book overstates end demand. The falsifying signal is specific: if ASML's long-term prepayments — now €9.3 billion — stop growing or decline for two consecutive quarters while customers simultaneously announce capital-spending cuts, the structural thesis is wrong and this is a front-loaded cycle. A second signal would be a shift in the order mix away from EUV and toward productivity upgrades and DUV, which would indicate customers are making do rather than expanding.
Who Benefits, Who Is Exposed
The immediate beneficiaries are the companies that already hold EUV allocation. TSMC, Samsung, and Intel have committed to High-NA EUV roadmaps, and Intel's early high-volume shipments give it a claim on the tightest capacity. Memory makers expanding for HBM and advanced DRAM — Micron, Samsung, SK Hynix — are bidding for the same tools. The equipment suppliers adjacent to ASML also benefit from the same capital-spending wave: Lam Research, Applied Materials, KLA, and the firms that supply ASML's own supply chain all ride the same spending cycle, though none enjoys ASML's monopoly position.
The exposed are the companies without secured tool access. Smaller foundries, late-moving AI chip startups, and any designer depending on a node transition in 2028 or 2029 face the risk of delayed ramps. Cloud providers that have budgeted for data-center capacity predicated on a certain flow of advanced chips may find their timelines pushed out. And consumers of advanced chips outside AI — smartphone makers, automotive suppliers — could see advanced-node capacity priced out of reach as AI customers outbid them.
What to Watch Next
ASML's fourth-quarter guidance calls for net sales of €14.2 billion to €15.2 billion with a gross margin around 57%, and full-year sales of €44 billion to €45 billion with a gross margin near 56%. China is still expected to represent 20% of total revenue, unchanged from the earlier projection. The company bought back around €4.1 billion of shares in the quarter under its 2026–2028 program and declared an interim dividend of €1.88 per share, payable November 4.
For the structural call to hold, three things need to happen. Capacity for 2028 must expand beyond 30%, and ASML must confirm a concrete number rather than an investigation. Long-term prepayments must keep rising, confirming that 2029 and 2030 interest is converting into committed orders. And High-NA EUV deliveries must ramp on schedule to support the Samsung 2028 and TSMC 2030 timelines.
The short-term view is that ASML's order momentum and margin strength leave little room for disappointment in the next few quarters. The medium-term view is that the supply-demand gap supports pricing power and high utilization through 2028. The long-term view is the one that matters most: if AI infrastructure spending proves durable, ASML's capacity constraint is a feature, not a bug — a durable moat that keeps returns high. If it proves cyclical, the moat becomes a mirror, reflecting a capital-spending cycle that eventually turns.
The takeaway is simple. For most of the AI boom, the market asked how many chips Nvidia and its peers could sell. ASML's October 14 disclosure changes the question. The constraint is no longer how many chips customers want to buy; it is how quickly the semiconductor industry can build the capacity to make them. And for the next two years at least, the answer is: not fast enough.
Explore more exclusive insights at nextfin.ai.
