Marché mondial de la lithographie EUV 2025-2033

1. Demande croissante pour les nœuds semi-conducteurs avancés Les applications de calcul haute performance, d'IA et d'Internet des objets nécessitent des puces dotées de géométries plus petites et d'une densité de transistors plus élevée.
2. Intégration de la 5G et de l'IA La demande de processeurs et de solutions mémoire plus rapides accélère l'adoption de la lithographie EUV.
3. Investissements gouvernementaux et industriels Initiatives stratégiques aux États-Unis, en Europe, à Taïwan, en Corée du Sud et en Chine pour renforcer l'autosuffisance en semi-conducteurs.
4. Miniaturisation de l'électronique Les smartphones, les objets connectés portables et les appareils électroniques grand public haute performance reposent sur une fabrication basée sur la lithographie EUV.
5. Expansion de la chaîne d'approvisionnement Une collaboration accrue entre les fonderies, les fournisseurs d'équipements et les entreprises de matériaux garantit l'évolutivité.

• Source de lumière
• Miroirs et optique
• Équipement de masque et de pellicule
• Systèmes de lithographie EUV
• Autres

• Fabrication de semi-conducteurs logiques
• Mémoire (DRAM, NAND)
• Fonderies
• Fabricants de dispositifs intégrés (IDM)

• Électronique grand public
• Automobile & Véhicules autonomes
• Télécommunications et 5G
• Centres de données et Cloud Computing
• Intelligence artificielle et apprentissage automatique

• Asie-Pacifique ??? le plus grand marché Taïwan, la Corée du Sud, la Chine et le Japon dominent grâce à de solides bases de fabrication de semi-conducteurs (TSMC, Samsung, SMIC).
• Amérique du Nord Les investissements américains dans le cadre de la loi CHIPS renforcent les usines de semi-conducteurs nationales et l'adoption de la technologie EUV.
• Europe Présence marquée de fabricants d'équipements comme ASML et investissements croissants dans la production de puces en Allemagne et aux Pays-Bas.
• Amérique latine et Moyen-Orient & Afrique Marchés émergents à adoption progressive, principalement axés sur les applications électroniques et automobiles.

• ASML Holding N.V.
• Taiwan Semiconductor Manufacturing Company (TSMC)
• Samsung Electronics Co., Ltd.
• Intel Corporation
• Applied Materials Inc.
• Société Nikon
• Canon Inc.
• GlobalFoundries
• Lam Research Corporation
• SK Hynix Inc.

• Moteur de prévision de la demande Pheonixindique une forte croissance tirée par l’IA, la 5G et les applications axées sur les données.
• Innovation Tracker technologiquemet en lumière les progrès continus dans la conception des pellicules, l’optique et la détection des défauts.
• Analyseur de sentiment des consommateurssuggère une demande solide pour les appareils alimentés par des puces plus petites, plus rapides et plus économes en énergie.
• Les cinq forces de Portermontre des barrières à l’entrée élevées en raison d’une R&D à forte intensité de capital, d’un nombre limité de fournisseurs d’équipements et d’un pouvoir de négociation élevé des acheteurs, concentré parmi les principales fonderies.

• La lithographie EUV est essentielle pour fabrication de puces de nouvelle génération.
• Demande croissante pour l’IA, 5G, et les technologies de conduite autonome rend la technologie EUV indispensable.
• Les investissements stratégiques des gouvernements et des entreprises garantissent la résilience des chaînes d'approvisionnement mondiales.
• Les innovations en matière d'optique EUV et de contrôle des défauts sont essentielles pour atteindre des rendements plus élevés et augmenter la production à l'échelle.

Table des matières

• Aperçu

• Principaux moteurs de la croissance du marché

  • Demande croissante pour les nœuds semi-conducteurs avancés

  • Intégration de la 5G et de l'IA

  • Investissements gouvernementaux et industriels

  • Miniaturisation de l'électronique

  • Expansion de la chaîne d'approvisionnement

• Segmentation du marché

  • Par type d’équipement

    • Source de lumière

    • Miroirs et optique

    • Équipement de masques et de pellicules

    • Systèmes de lithographie EUV

    • Autres

  • Par Secteur d'Utilisation Finale

    • Fabrication de semi-conducteurs logiques

    • Mémoire (DRAM, NAND)

    • Fonderies

    • Fabricants de dispositifs intégrés (IDM)

  • Par application

    • Électronique grand public

    • Automobile et véhicules autonomes

    • Télécommunications et 5G

    • Centres de données et Cloud Computing

    • Intelligence artificielle et apprentissage automatique

• Insights au niveau régional

  • Asie-Pacifique ??? le plus grand marché

  • Amérique du Nord

  • Europe

  • Amérique latine et Moyen-Orient & Afrique

• Principales entreprises du marché

• Intelligence stratégique et perspectives basées sur l’IA

• Aperçu des prévisions : 2025–2033

• Pourquoi le marché reste essentiel

• Conclusion

Paysage concurrentiel

There is no share to win, so the chapter has to be about something else

Every other report in this library opens its competitive chapter by dividing a market between firms. This one cannot, because the division is 100 to zero. ASML has supplied every commercial EUV scanner in high-volume manufacturing, and its 2025 EUV system revenue of €11.6 billion represented 45.5 per cent of the company’s entire €32.7 billion of net sales. Nikon and Canon, the two firms that between them hold most of the world’s lithography patents, do not sell an EUV production tool. The duopoly that defined lithography for thirty years simply ended at 13.5 nanometres.

So the useful questions are different. Who can say no, and what does it cost them? That is the real competitive structure, and it has four layers that never bid against each other.

Layer one: the customer, which is the only competitive constraint that exists

The buyer side is four companies. Net system sales in the fourth quarter of 2025 were 70 per cent logic and 30 per cent memory; by the third quarter of 2026 the split was 54 per cent logic and 46 per cent memory. That single swing is the most important competitive event of the year, and it was not caused by any vendor.

What it shows is that EUV demand is not one market but two, with uncorrelated clocks. Logic demand is set by leading-edge foundry capacity — TSMC’s N2 ramp and its A16 successor, Intel’s 18A and 14A. Memory demand is set by DRAM node transitions and by high-bandwidth memory, which needs die stacking rather than more lithography but drags the whole DRAM wafer base with it. When memory turns, as it did through 2026, it arrives as a step change rather than a slope, because a DRAM maker converting to EUV converts whole fabs.

The consequence for pricing is peculiar. In a normal monopoly the supplier extracts; here the supplier allocates. ASML’s own words in October 2026 were that supply “as currently planned for both 2027 and 2028 will not fully meet the demand of our customers”, and that some customers were asking to place orders for 2029 and even 2030. When allocation replaces price as the clearing mechanism, a customer’s competitive position depends on its relationship with one supplier — which is why TSMC, Samsung and Intel have all taken equity-like strategic positions in the roadmap rather than simply buying tools, and why they committed publicly to High-NA in September 2026.

Layer two: the optics and source oligopoly beneath the monopolist

ASML is a monopolist with a single point of failure, and it has said so in its own filings for over a decade. Its annual report language is unusually direct: the company relies on outside vendors for components and subassemblies “each of which is obtained from a sole supplier or a limited number of suppliers”, and in particular “the number of systems ASML has been able to produce has been limited by the production capacity of Zeiss”. Zeiss is the sole external supplier of the projection optics, and can produce them only at Oberkochen and Wetzlar in Germany.

The source is the other half. In 2013 ASML bought Cymer, its EUV light-source supplier, and the filing language explaining why is worth reading twice: without the acquisition, ASML did not believe Cymer “would have sufficient resources to complete the development of the EUV source”. The monopoly was not built by out-competing a supplier. It was built by buying one to keep the roadmap alive. Trumpf supplies the CO2 drive laser that generates the tin plasma.

This is the honest reading of the “tier” question. There is one tier-1 player in EUV systems. Beneath it sits a supply chain of perhaps six firms, several of them sole-source, none of them substitutable inside the forecast window. A competitor to ASML would not need to out-engineer ASML; it would need to recreate Zeiss.

Layer three: the mask infrastructure, which is where the real second supply base lives

The only genuinely multi-vendor part of the EUV stack is the part that never enters the scanner: mask blanks, mask inspection, pellicles and resist. A chipmaker buying an EUV tool has one choice of vendor. A chipmaker qualifying an EUV layer has several choices of blank supplier, several of resist chemistry, and a partly contestable inspection market. This is discussed in the Value Chain chapter, but it belongs in the competitive chapter too, because it inverts the usual logic: the highest-barrier component is not the scarcest. ASML can build the tool; the constraint on using it moves to whether the mask is good enough.

Layer four: the competition ASML actually runs, which is against itself

Because no one else bids, the competitive boundary is internal, and it is priced. On 2025 realised revenue, a low-NA NXE system costs €1.078 million per wafer-per-hour of capacity at 220 WPH. A High-NA EXE system costs €2.556 million per WPH at 135 WPH — 2.37 times the capital intensity per unit of throughput.

That ratio is the whole of the High-NA adoption argument. High-NA does not win because it is better; it wins only on the layers where 0.33 NA cannot print the pattern in a single exposure and the customer would otherwise pay for multi-patterning, which costs more than 2.37 times in yield, cycle time and defect density. Every High-NA unit in this forecast is therefore a decision by a customer that a specific layer has crossed that line — and the roadmap closes the gap deliberately, from 135 WPH today to a targeted 250-plus on the planned EXE:5600.

The same logic governs the low-NA catalogue. The NXE:3800F, due in 2027, raises throughput from 230 to 260 WPH. It is not a new market; it is the same market at 13 per cent more output per euro of tool. ASML competes with its own installed base by making it cheaper to buy a new tool than to keep running the old one — which is precisely why it recorded so many 3800E field upgrades in 2025 that it disclosed a shift of EUV revenue from systems into installed base.

How the structure scores

The market is highly consolidated in the only sense that matters — one supplier — with one tier-1 participant. Competitive intensity is low, and saying so is not complacency: there is no price war, no share shift and no second source, so the usual intensity metrics return nothing. What is intense is the queue. The correct reading of a low-intensity, single-supplier market is not that it is calm. It is that its failure modes are all supply-side, and the Risk chapter treats them that way.

Chaîne de valeur

Seven gates, and the hinge is the mirror

The EUV value chain is usually described from the laser backwards. That is the wrong direction, because the laser is not what makes EUV difficult. What makes it difficult is that at 13.5 nanometres every material absorbs. There is no lens, no transmission, no refractive optic anywhere in the system. The light must bounce.

That single physical fact generates the entire cost structure. EUV optics are molybdenum-silicon multilayer mirrors, each reflecting roughly 70 per cent of the light that hits it. A production scanner has on the order of ten of them in the path from source to wafer. Compound that and a few per cent of the light generated reaches the resist. This is why a 500-watt source on the NXE:3800E was an engineering milestone rather than a specification bump, and why throughput, not resolution, is the commercial variable in this industry. Every euro of EUV economics is a fight against absorption.

Gate one and two: the blank

An EUV mask is not a photographic plate. It is a low-thermal-expansion substrate — Corning’s ULE glass-ceramic is the reference material — coated with dozens of alternating Mo/Si layers to form a reflective stack, then capped with an absorber. The substrate cannot expand measurably when the mask is heated by the beam, because a nanometre of distortion at the mask prints as a defect at the wafer. Mask blanks are supplied by a very small group, with Hoya and AGC the substantive names.

The commercial consequence is peculiar: a blank is worth more than its materials by orders of magnitude, and its lead time is measured in months. A blank defect discovered late destroys a mask that took weeks to write. This is why the mask layer, not the scanner, is where a fab’s ramp schedule actually breaks.

Gate three and four: write, repair, and the pellicle

An EUV mask is written by e-beam, pattern by pattern, over days; then inspected, then repaired. The inspection step is the one genuinely multi-vendor node in the stack, because it must be actinic — it must look at the mask with 13.5nm light, since a DUV inspection tool cannot see the defects that matter. Then the pellicle: a membrane thin enough to transmit EUV and robust enough to survive a 500-watt source, standing off the mask so that a falling particle lands on the membrane rather than the pattern.

For most of EUV’s history the pellicle did not exist, and fabs ran without one, accepting defectivity. It exists now. The pellicle is a good example of a component that is small in revenue, decisive in yield, and supplied by a handful of firms including ASML itself, Mitsui and Korea’s S&S Tech.

Gate five: the scanner, which is really three supply chains bolted together

Inside the tool, three things arrive from three places and must be integrated to sub-nanometre tolerance. The drive laser comes from Trumpf. The source — tin droplets, blasted into plasma at 500 watts, collected by an ellipsoidal mirror — comes from Cymer, which is ASML. The projection optics come from Zeiss SMT, which is the only supplier on earth that can make them and, as ASML’s own filings state, has historically been the limit on how many systems the industry could produce.

This is the structural asymmetry the competitive chapter identified: the monopolist is a systems integrator standing on a sole-source base. Zeiss is not a vendor that can be second-sourced. It is the reason there is a monopoly.

Gate six: integration into the fab, where the tool becomes a process

An EUV scanner installed is not an EUV layer qualified. Between the two sit overlay metrology, focus control, computational lithography, and defect inspection — ASML’s holistic lithography portfolio, plus YieldStar optical metrology and multibeam e-beam inspection, which the company reports is being adopted specifically because 3D device structures produce defects that cannot be seen optically. In 2025 ASML’s metrology and inspection system sales were €825 million, up 28 per cent.

This gate is where EUV’s cost actually lands. A fab that buys the tool but not the control layer gets the resolution and none of the yield, and the difference between a 90 per cent and a 93 per cent available tool is measured in wafer starts, not specifications.

Gate seven: the field, which is where the money compounds

The last gate is the installed base itself, and it is the one the industry persistently undervalues. In 2025 ASML carried out what it described as “a significant number of TWINSCAN NXE:3800E field upgrades”, and disclosed the accounting consequence plainly: a substantial portion of EUV system revenue shifted into installed base revenue. Net service and field-option sales rose 26.2 per cent to €8.19 billion in 2025, and the company expects roughly 50 per cent growth in that line in 2026.

On this report’s attribution of 40 per cent of that revenue to EUV, an installed EUV system generated €9.53 million of service and upgrade revenue in 2025 — about 4 per cent of its original purchase price, every year, recurring. That is the number that makes EUV a durable business rather than a capex cycle, and it is why the service line supplies 34.2 per cent of this forecast’s absolute growth from 22.0 per cent of its 2025 value.

The next structural change is the mask format

In September 2026 ASML, TSMC, Samsung and Intel committed to an industry initiative to move EUV photomasks from the current 6-inch format to 12-inch. The stated benefits are productivity and lower chipmaking cost. The mechanism is straightforward: a larger mask exposes more die per pass, which reduces the number of exposures per wafer and attacks the same absorption arithmetic from the other side.

It is also a warning about the shape of this value chain. A format change resets the mask infrastructure layer — blanks, writers, inspection, pellicles — while leaving the scanner largely untouched. The next decade of EUV cost reduction may come less from the tool than from the plate that feeds it.

What this chapter does not size, and why

The attached spend on EUV masks, blanks, pellicles and resist is real and growing, and this report deliberately declines to put a number on it. No published series isolates EUV-specific mask and materials spend; the figures that circulate are derived from total photomask market estimates, which are dominated by mature-node DUV volume and therefore say almost nothing about the EUV subset. An EUV mask set is a very different object from a 28nm mask set and cannot be scaled from it. Stating a figure would require inventing a method, and this report would rather name the gap: the scanner is measurable, the ecosystem around it is not, and the ecosystem is where the next constraint will appear.

The value-chain model that this report does use is hybrid: vertically integrated at the scanner vendor, where source and system are one company, and distributor-independent at the mask and materials layer, where fabs buy direct from a small panel of qualified suppliers. Supply chain complexity is high for a structural reason, not a commercial one — the number of firms capable of making any one critical component is in single digits, and for the optics it is one.

Activité d'investissement

Where the capital actually goes, which is not where the industry says

ASML spent €4.7 billion on research and development in 2025, which was about 14 per cent of net sales, and a further €1.3 billion on selling, general and administrative costs, about 4 per cent. Those two numbers describe a company whose competitive moat is built out of engineers rather than out of factories. For comparison, the customers who buy the tool spend tens of billions of dollars a year on fabs; the supplier that makes the tool spendable spends billions on design. The asymmetry is the point: the scarcity in EUV is intellectual, not industrial.

That is why the headline investment number in this market is not a capex figure at all. It is the capacity plan, expressed in units, because ASML’s binding constraint has historically been Zeiss’s optics output and its own cleanroom integration capacity rather than money.

The committed capacity, which is the only hard investment fact in the report

Alongside it, ASML plans to add 30 per cent to its 2026 DUV immersion capacity of around 130 units for 2027. In October 2026 the CEO stated that 30 per cent in 2027 and 30 per cent in 2028 “will not allow us to meet our customer demand”, that the company was looking at ways to extend capacity further “especially in 2028 and beyond”, and that a second 30 per cent step for 2028 was being investigated beyond the one already communicated.

This report’s baseline takes the announced steps and then decelerates hard: low-NA units reach 96 in 2028 and only 105 by 2033. That is a deliberate conservatism against a company that has said it intends to go further. It is also the reason the conservative scenario exists as a separate path rather than as a footnote.

The order book is the investment, seen from the other side

ASML ended 2025 with a backlog of €38.8 billion, after €28.0 billion of net bookings during the year, of which the fourth quarter alone contributed €13.2 billion — €7.4 billion of it EUV. In October 2026 the company reported order momentum increasing quarter on quarter and said some customers were seeking to place orders for 2029 and even 2030.

For a market report, the significance is not the backlog size. It is that the backlog converts the next three years of this forecast from a projection into a delivery schedule. A customer that has ordered a tool for 2028 has already made the investment decision that the forecast assumes. That is why the near-term half of the S-curve in this report is far more reliable than the far half, and why the 50.6 per cent of growth landing by 2028 is the most defensible number in it.

What the cash is being used for, and what that signals

ASML generated €11.0 billion of free cash flow in 2025 and returned €8.5 billion to shareholders — €2.6 billion of dividends and €5.9 billion of buybacks — while announcing a new buyback programme of up to €12 billion running to the end of 2028. A monopoly that returns three quarters of its free cash flow is telling you something specific: it does not believe it can deploy more capital productively inside its own moat, because the moat is not capital-hungry.

The one investment that cuts against that reading is the approximately 11 per cent fully diluted stake in Mistral AI taken during 2025. ASML’s own explanation was that the model supports software coding and therefore the development of its systems. Whether or not that is the real rationale, it is a signal about where the next defensibility is expected to come from: not from harder optics, but from the software that makes a hard tool productive.

On the ground, the company broke ground in September 2026 on a major new Eindhoven site intended to eventually accommodate up to 20,000 workers, including offices, a logistics hub and manufacturing facilities. That is a decade-scale commitment to the Netherlands, and it is also a statement that the constraint ASML expects to hit is skilled people, not demand.

The customer-side commitments that date the High-NA line

Three announcements in 2026 do more for the forecast than any capex figure. Intel moved High-NA into high-volume manufacturing on its 18A process in July 2026, with Panther Lake the first logic product to reach volume production using the technology, and reported processing more than a million wafers on High-NA equipment. Samsung committed to High-NA for DRAM high-volume manufacturing from 2028. TSMC committed to adopt it in 2030.

Those are dated intentions from the only three companies that matter, and they sit inside this report’s High-NA unit path — 5 systems in 2026, 12 in 2028, 24 in 2033. A reader who thinks the 2028 to 2030 gap is too slow should note that it is the customers’ own schedule, not this model’s.

How the investment profile scores

Investment trend is rising — unambiguously, and across every line: capacity, R&D, backlog, and customer commitments. Capital intensity is high, but the intensity sits with the buyer, not the seller: an EUV layer is the most capital-intensive thing a leading-edge fab does, at €1.078 million per wafer-per-hour on low-NA and €2.556 million on High-NA. Recent M&A activity is none: the last transformative transaction was Cymer in 2013, and the reason this market consolidated into one firm is that the consolidation happened fifteen years ago and has not been contestable since.

Technologie et innovation

The molecule does not have a roadmap. The machine does.

EUV lithography sits at a fixed wavelength — 13.5 nanometres — and that number will not change. So unlike most technology markets, there is no generational leap to anticipate, no chemistry substitution, no architecture reset. What there is, and what the entire innovation story consists of, is a set of productivity curves and one very stubborn reliability curve.

That makes EUV unusual in another way: innovation here is measured in wafers per hour and in uptime percentage, not in resolution. Resolution has been sufficient since 2019, when the first NXE systems entered high-volume manufacturing at 7nm. Everything since has been about making the same pattern cheaper.

The low-NA curve, which is where nearly all the volume is

The NXE platform has improved throughput in four steps: 137 wafers per hour on the 3400B, 145 on the 3400C, 160 on the 3600D, and 220 on the 3800E — a 38 per cent gain from the 3600D to the 3800E, alongside a 13 per cent overlay improvement. The NXE:3800F, due in 2027, takes that to 260 WPH.

This is the least glamorous and most consequential part of the report. Low-NA is 73.3 per cent of the 2025 market and still 53.5 per cent in 2033. A 13 per cent throughput gain on the 3800F is worth more to a fab than any High-NA milestone, because it applies to every critical layer rather than the handful that need 0.55 NA.

Reliability is the other half, and it is the honest measure of maturity. ASML’s 0.33 NA fleet exceeded 93.5 per cent worldwide average availability in the third quarter of 2024 and the company was moving the fleet toward 95 per cent. For a tool costing a quarter of a billion euros, a point of availability is worth more than a point of throughput, because an idle scanner is a fab losing wafer starts it cannot recover.

The High-NA curve, and the 2.37-times problem

The EXE platform moves numerical aperture from 0.33 to 0.55, which shrinks printable features by up to two thirds at the same pitch and permits single-exposure patterning below 2nm. It arrives with a cost problem that the roadmap is explicitly designed to solve.

At 2025 realised prices, low-NA costs €1.078 million per wafer-per-hour and High-NA €2.556 million — 2.37 times. That gap is not a defect; it is the deliberate pricing of a capability, and it is why High-NA will never be deployed across a whole device. It goes to the layers where 0.33 NA would require multi-patterning, and ASML’s own assessment is that on those critical layers the 0.55 NA platform lowers cost by 20 to 35 per cent against 0.33 NA through dose reduction, single exposure and simpler 2D layouts.

In other words, High-NA is not more economical per wafer-hour. It removes wafer-hours from the process. That distinction is the whole of its business case, and it is why the 2028 and 2030 customer commitments are layer-specific rather than fab-wide.

Availability is the number that decides whether High-NA is real

The most significant technology number published in 2026 was not a throughput figure. It was 84 per cent — the High-NA fleet availability reported in July 2026 — against a target of 90 per cent in the fourth quarter of 2026 and a longer-term goal of 93 per cent. ASML’s head of High-NA product management said the company believed it could reach a stable level meeting customer requirements as they move into volume manufacturing by the end of the year.

That is the honest scoreboard. A tool at 84 per cent is a development tool. A tool at 90 per cent is a production tool. The gap between those two numbers contains most of this report’s High-NA risk, and it is a reliability engineering problem, not a physics problem — which is why the installed base matters: by mid-July 2026 the EXE fleet had accumulated more than 1.35 million wafer exposures, giving ASML the operational dataset it needed to find and fix the failure modes.

Why EUV intensity keeps rising, which is the demand-side innovation

The most underrated technology driver in this market is not in the tool at all. It is the number of EUV layers per device, and it is going up for two independent reasons.

The first is DRAM. Memory makers are migrating multi-patterning DUV layers to single-exposure EUV, and the shift from 6F² to 4F² cell architecture increases EUV intensity further. This is why memory went from 30 per cent of ASML’s system sales in the fourth quarter of 2025 to 46 per cent in the third quarter of 2026, and why the Korean line in this forecast grows at 18.11 per cent.

The second is transistor architecture. As logic moves to gate-all-around and then to stacked structures, the number of critical exposures per device rises even when the node number barely changes. ASML describes this as lithography intensity, and it is the reason a market growing 13.15 per cent in units can grow 17.06 per cent in revenue.

Hyper-NA, and why it is not in this forecast

ASML’s roadmap extends to a 0.75 NA “Hyper NA” concept beyond 0.55. It appears in the company’s own long-range material alongside 0.33 and 0.55, and it is the natural next step in the same arithmetic: more numerical aperture, fewer exposures.

It is deliberately absent from this model. No timeline has been disclosed, no customer has committed, and the optical and resist requirements are unsolved. Including it would be forecasting a product announcement rather than a market. If High-NA is the 2026 to 2030 story, Hyper-NA is the 2033-plus story, and this report ends at 2033.

How the technology profile scores

Innovation intensity is high, and the R&D supporting it is €4.7 billion a year. Patent activity is high — stated as judgement rather than measurement, because no published series isolates EUV-specific filings, and the moat here is process know-how held in a small number of firms rather than a licensable portfolio. Technology maturity is mature for the 0.33 NA platform, which has been in volume manufacturing since 2019 and exceeds 93.5 per cent availability, and growth-stage for 0.55 NA. The single headline value understates a market that contains both; the split matters more than the label.

Risque de marché

Six risks, ranked, and the dominant one is not the obvious one

The risks normally associated with a monopoly are the wrong ones here. There is no competitor to lose share to, no price war, no commoditisation, and no realistic substitution inside the forecast window. What can go wrong is narrower and much more specific, and it sits almost entirely on the demand side and on the ramp.

1. Customer concentration — dominant

Four buyers. That is the whole of this market’s demand, and the swing between them is violent. ASML’s net system sales were 70 per cent logic and 30 per cent memory in the fourth quarter of 2025 and 54 per cent logic and 46 per cent memory in the third quarter of 2026. A 16-point shift in two quarters, with no change in total demand, is a warning about how thin the diversification really is.

This matters more than it looks because the two clocks are not synchronised and neither is visible far ahead. A foundry’s node transition is a multi-year plan measured in fab shells and wafer starts; a DRAM maker’s EUV conversion is a fab-wide decision taken when the memory price supports it. If a single leading-edge logic customer slips a node by two quarters, several EUV tools move with it, and there is no second buyer waiting for that capacity — because the queue is the market.

The asymmetry is that this risk is not symmetric in time. Concentration is benign while demand exceeds supply, which is the situation through 2028. It becomes severe precisely when supply catches up, because that is when a single customer’s pause becomes idle capacity rather than a shorter queue. The concentration risk is therefore latent until the capacity risk crystallises, and then the two compound.

2. High-NA adoption timing — the model’s single largest assumption

High-NA supplies 20.2 per cent of this forecast’s absolute growth from 4.6 per cent of its 2025 value, growing at 36.43 per cent against the market’s 17.06. It is also the line with the least evidence behind it, because it depends on three customers’ dated intentions rather than on orders already placed.

Those intentions are specific and they are late: Intel is in high-volume manufacturing now and reported more than a million wafers processed; Samsung has committed to High-NA for DRAM high-volume manufacturing from 2028; TSMC has committed to adopt it in 2030. Between 2026 and 2030, therefore, High-NA revenue is largely development, qualification and early-production tooling.

If High-NA units in 2033 land at half the baseline — 12 instead of 24 — the market loses €4.14 billion, or USD 4.76 billion, and the total 2033 figure falls from USD 60.33 billion to roughly USD 55.6 billion, which is about 1.5 points of CAGR. That is a real but survivable miss, and it is the correct size to put on it. The risk that would not be survivable is different: if High-NA stalls entirely and the industry stays on 0.33 NA multi-patterning, the cost-of-technology curve that justifies each node’s economics stops working, and the demand for low-NA tools falls with it.

3. Capacity execution — the one that decides the decade

This is the risk the conservative scenario actually prices. The baseline assumes ASML builds 105 low-NA and 24 High-NA systems in 2033; the conservative case assumes the industry builds what it has already promised and then stops, at 87 and 15. The difference is USD 19.29 billion in 2033 — 60.33 against 41.03 — and it is almost entirely a single question: whether the 2028 capacity ceiling is a ceiling.

ASML has said it is investigating ways to go beyond the announced 30 per cent steps for 2028, and has said plainly that the announced steps will not meet demand. But the constraint it is pushing against is not money. It is Zeiss’s optics output, the availability of qualified integration staff, and the physical build-out in Eindhoven and Veldhoven. A capacity plan that depends on a sole-source optics supplier doubling its output is a plan with a single point of failure in it, and that is the honest reason this report does not extrapolate the 30 per cent steps beyond 2028.

4. Export controls and geopolitics — high, but not in the way usually reported

The direct exposure is zero. No EUV scanner has ever been exported to China under any licence, so there is no revenue to lose. Any analysis that treats China restrictions as an EUV revenue risk is describing DUV.

The indirect exposure is substantial. In April 2026 the MATCH Act was introduced in the United States House, seeking to force allied countries — explicitly including the Netherlands and Japan — to apply US-equivalent export restrictions, and naming ASML equipment. The Dutch government formally objected: trade minister Sjoerd Sjoerdsma’s written answers of 12 May 2026 confirm that the Prime Minister and senior officials raised the concern with Washington during the April state visit. ASML expects China to be around 20 per cent of its total net sales in 2026, essentially all of it DUV.

So the exposure is asymmetric: EUV has nothing to lose and DUV immersion has a fifth of the company behind it. The retaliation risk is the harder one to model, because it would arrive as a restriction on the materials and components ASML buys rather than on the tools it sells.

5. AI capex digestion

2026 is a 36 per cent growth year for the company and a 46.6 per cent growth year for EUV, driven by AI infrastructure spending that is itself concentrated in a handful of hyperscalers. ASML’s stated position is that customer demand will exceed its supply at least through 2028, and that some customers want to order for 2029 and 2030.

That is a strong position and it is also the classic setup for an air pocket. The specific mechanic to watch is not demand falling but demand arriving early: if customers pull 2029 orders into 2027 and 2028 to secure allocation, the backlog that makes the near term safe is depleted exactly when the capacity expansion lands. This report’s deceleration after 2028 — 96 low-NA units in 2028 rising only to 105 by 2033 — is partly a hedge against that pattern.

6. Supply chain single point of failure

ASML’s own filings are unusually frank: components are obtained from a sole supplier or a limited number of suppliers, and historically “the number of systems ASML has been able to produce has been limited by the production capacity of Zeiss”. An interruption at Oberkochen or Wetzlar — industrial action, an energy shock, a fire, a single failed qualification batch — is not hedgeable, insurable at sensible cost, or substitutable inside any planning horizon.

This risk does not show up in a forecast because it is a step function rather than a distribution. It belongs in the report anyway, because a reader buying exposure to this market is buying exposure to that concentration whether or not the model says so.

What is not a risk

Substitution is low. Nothing replaces EUV for sub-7nm critical layers in the forecast window. The theoretical alternatives — nanoimprint, directed self-assembly, e-beam direct write — are either throughput-limited by orders of magnitude or confined to niche layers. The realistic fallback is DUV multi-patterning, which is not a substitute but a more expensive way of doing the same job, and which is exactly what the 0.55 NA platform is designed to displace.

Overall market risk is moderate: the demand is real, the backlog exists, and the supplier is not going to be displaced. Geopolitical exposure is high, not because of EUV’s own China exposure, which is nil, but because the company carrying this market has a fifth of its revenue in the one jurisdiction the controls are aimed at.

Paysage réglementaire

The only regulation that matters here is an export licence

EUV lithography is the most heavily controlled machine in civilian manufacturing, and the regime that controls it is not a product-safety regime. There is no type approval to obtain, no notified body, no CE marking process of the kind that governs medical devices or lifts. A scanner costing €345 million is regulated by whether a government will let it leave the country and who is allowed to receive it.

That inverts the usual structure of a regulatory chapter. In most capital-goods markets, regulation constrains the design of the product and compliance is a cost of market access. Here, regulation constrains the customer list, and compliance is a condition of having any market at all. It also means the regulatory risk and the commercial risk are the same risk, seen from two directions.

Three stacked regimes, all pointing the same way

Multilateral. EUV controls descend from the Wassenaar Arrangement on dual-use goods and technologies, which for two decades set the baseline for what could be exported where. In practice, national controls have moved faster and further than the multilateral text, so Wassenaar now functions more as a floor than as an operating rule.

European and Dutch. EU dual-use Regulation 2021/821 establishes the Union framework, and it is implemented nationally — in ASML’s case by the Netherlands, which issues the licences. This is why the Dutch government’s position matters commercially and not merely diplomatically: it is the licensing authority for the only EUV supplier on earth.

American. The United States Export Administration Regulations, the Entity List, and the foreign direct product rule have driven the pace of tightening since 2018, and they reach beyond US-origin goods to products made abroad with US technology or tooling.

The 2026 development, and why it matters more for DUV than for EUV

In April 2026 the MATCH Act was introduced in the United States House of Representatives, with Republican and Democratic sponsors, seeking to require allied countries — the Netherlands and Japan named — to apply the same restrictions to their own firms that the United States applies to American ones. ASML’s equipment is named in the legislation.

The Dutch response was formal and on the record. Trade minister Sjoerd Sjoerdsma’s answers to parliament, dated 12 May 2026, confirm that the Netherlands objected to the proposal, and that Prime Minister Rob Jetten and other senior officials raised the concern during the April state visit to the United States. The minister acknowledged that the Act could in practice prevent ASML from selling immersion DUV lithography machines to China.

Note what that sentence is about. Not EUV. Immersion DUV. EUV has never been licensed for China and is not the subject of the debate; the entire legislative fight concerns the tool one generation below it. That is the correct reading of the regulatory risk in this market, and it is why this chapter and the geopolitics section of the Risk chapter say the same thing in different registers.

Where regulation does shape the forecast: location

Subsidy regimes do not restrict who can buy an EUV tool, they determine where a fab is built — and therefore which region books the tool. The American CHIPS programme, the European Chips Act and Japan’s semiconductor support have all been justified explicitly as reducing dependence on a concentrated manufacturing base, and all of them subsidise the construction of exactly the leading-edge capacity that buys EUV.

This is the mechanism behind the regional shift in this report: North America grows at 20.03 per cent against Taiwan’s 14.71 and Europe’s 14.83, taking its share from 18.0 to 22.0 per cent, because the subsidised capacity is being built there. It is also the reason Japan appears at all as a region, growing 18.80 per cent from a small base.

The standard that is not a regulation

In September 2026 ASML, TSMC, Samsung and Intel committed to an industry initiative to move EUV photomasks from the 6-inch to the 12-inch format. No government required it. It is a de facto standard agreed among the four firms that can make it matter, and it will govern blank supply, mask writing, inspection and pellicle design for the next decade.

This is worth stating in a regulatory chapter because it shows where the real coordination happens in this industry. On export control, governments lead and firms comply. On the technical standards that determine cost, firms lead and governments are absent. A reader tracking this market should watch the second far more closely than the first, because a mask format change moves the value chain and a licence change moves a press release.

How the regulatory profile scores

Regulatory complexity is high — three stacked regimes, one licensing authority, and a live legislative proposal that would change the rules for the generation below the product this report covers. The approval pathway is export-licence controlled: not a certification of the product, but an authorisation of the transaction, granted per destination and revocable in a way that a CE mark never is.

The practical implication for anyone using this report is that the EUV forecast does not carry a regulatory discount, because there is no realistic regulatory scenario in which EUV tools are sold to a new jurisdiction. The discount belongs on the DUV and installed-base lines, and it is carried there, in the company’s own results rather than in this market’s numbers.