This group makes the machines that print the circuits onto silicon chips. The product is a lithography tool — a room-sized machine that uses light to project a chip's circuit pattern onto a silicon wafer, the way a stencil sprays paint, but at a scale measured in billionths of a metre. The newest kind, EUV (extreme ultraviolet — light with a very short 13.5-nanometre wavelength), prints features at the 3-nanometre node and below, and is widely described as the most complex machine ever built (source: 500-stocks scan). Without lithography, no advanced AI chip — no GPU (graphics processing unit, the main AI compute chip), no HBM (high-bandwidth memory) memory, no AI accelerator — can be made. One company, ASML of the Netherlands, makes every EUV machine on Earth; a small Japanese supplier, Lasertec, supplies most of the inspection tools that check the masks (the patterned templates) those machines use. So the "owner" question here is unusually clean: you are buying exposure to a near-monopoly on one of the hardest-to-replicate bottlenecks in the chip supply chain.
Demand for these machines has, on the structural facts presented, been running ahead of supply: every new AI-chip fab (fabrication plant) needs dozens of EUV tools, while only one company can build EUV and only at roughly 50-60 units a year (est.). Supply is limited not mainly by money but by physics and accumulated know-how — the optics, the lasers, and decades of engineering cannot be reproduced quickly. The market currently prices this scarcity at roughly the same dollars of market value per dollar of revenue as a fast-growing software firm (est.), even though this is a heavy-machinery business. This sheet lays out the arithmetic; the reader judges whether that price is warranted.
The product is a lithography system — sold one machine at a time. There are two broad families:
The companies earn cash three ways. (1) Selling the machine — a single large payment when a tool ships, the bulk of revenue. (2) Service and upgrades — once a fab owns 1,000+ of these machines, it pays ASML every year to keep them running and to boost their output; this "installed base" revenue (the recurring income earned from machines already sold and in use) is recurring and high-margin, the closest thing to a subscription in heavy industry. (3) Field upgrades — selling more performance into machines already on a customer's floor. The plain-English point for an owner: because the installed base only grows, the recurring service stream grows with it, which softens the natural boom-bust of a machinery business. EBITDA (earnings before interest, tax, depreciation and amortisation — a rough proxy for the operating cash the business throws off before financing and accounting items) margins for the leader have been reported in roughly the high-30s to low-40s percent range est. — software-like for a maker of physical equipment.
Source: 500-stocks scan (machine cost, monopoly structure); machine pricing and margin figures from general knowledge, not live-verified.
Demand for lithography tools is a derivative of demand for chips, which is itself a derivative of demand for compute. The chain is: more AI compute → more chips → more wafer factories ("fabs") → more lithography machines. The scan describes the AI link as "Direct": every new advanced-chip fab needs dozens of EUV tools, and TSMC's new Arizona, Japan and other fabs all require ASML's latest machines (source: scan).
Who the buyers are. A very short list: the leading-edge logic foundries (chip factories that build chips for others — TSMC, Samsung, Intel), the memory makers (SK Hynix, Samsung, Micron), and a handful of Chinese fabs (whose access to EUV is blocked by export controls, so they buy DUV instead). On a rough order of magnitude, around a dozen customers worldwide account for nearly all EUV demand (est.). This concentration cuts both ways: very large orders, but few buyers — if two or three of them pause capex (capital expenditure — money a company spends on big, long-lived assets like factories and machines), order flow can swing hard.
Contracted vs forecast. What is genuinely contracted is the order backlog — machines customers have already committed to buy, reported by ASML to stretch "years" out (source: scan). contracted That backlog is the firmest demand evidence available. Everything beyond it is forecast.
Forecast (AGI lens). forecast Reasoning from the premise that AGI is arriving: compute demand does not plateau — it compounds, as frontier labs scale training and, increasingly, inference for agents and physical-AI (robotics). Every doubling of installed AI compute must ultimately be fabricated, and the leading-edge transistors inside those chips can only be printed by EUV. On that premise, the structural demand curve for the most advanced lithography points up and to the right for as far out as is reasonable to forecast. The cyclical risk is timing, not direction: a digestion pause after a capex surge can still cut tool orders sharply for a year or two even while the long-run trend rises. This paragraph is a forecast, not a contracted fact.
✓ VERIFIED — the following figures were confirmed from primary sources after initial publication:
Current capacity. ASML produces roughly 50-60 EUV systems per year (est.), and the order backlog stretches years beyond that (source: scan). DUV output is much higher (hundreds of units a year across makers, est.) but EUV is the binding constraint for the AI-relevant leading edge.
What actually limits supply — and why money cannot fix it fast.
Market-share structure (who controls supply). est. EUV: ASML approximately 100%. DUV: ASML holds a large majority (well over half), with Nikon and Canon splitting the rest. Mask-inspection (the tooling that checks EUV photomasks for defects): Japan's Lasertec is the dominant supplier. On these figures, this group's supply is among the most concentrated in the economy — closer in structure to a regulated monopoly than to a normal competitive market.
Source: 500-stocks scan (capacity, backlog, monopoly, hardest-bottleneck framing); supplier names and DUV share splits from general knowledge, not live-verified.
Put the two sides together and, on the structural facts presented, the product is short at the leading edge: one supplier, a physical cap of roughly 50-60 EUV units a year (est.), and a multi-year order backlog. The evidence of shortage is the backlog itself plus reported high utilisation — customers wait in line rather than shop around, because there is nowhere else to shop.
| Signal | What it shows | Reading |
|---|---|---|
| Order backlog | Multiple years (source: scan) | Demand booked well past current output → short |
| EUV annual output | ~50-60 units est. | Physical ceiling, slow to lift → short |
| Number of EUV makers | 1 | No competitive supply relief available → short |
| Pricing trend | Rising (High-NA > standard EUV) est. | Higher price per tool over time → short |
| DUV (non-leading-edge) | Several makers, higher volume | Closer to balanced; less scarce |
When could it flip to oversupply? forecast Not through new competition — there is no credible second EUV maker on any near horizon. The realistic ways the gap could narrow are: (1) a cyclical demand pause — if the big foundries and memory makers over-build and then digest, tool orders can drop sharply for a year or two even though the long-run trend is up; (2) China's DUV market being filled by domestic tool-makers over time, eroding ASML's DUV (not EUV) volume; (3) a much slower compute build-out than the AGI premise implies. Under the premise that AGI is arriving, scenario (3) is the least likely of the three, which leaves the cyclical pause as the main exposure for an owner. These are forecasts, not contracted facts.
Source: 500-stocks scan; cyclical and China-DUV dynamics from general knowledge, not live-verified.
This is the bridge to individual deep-dives. The group is dominated by one name; the others are either small pure-plays or large companies where lithography is one slice.
| Company (ticker) | What it makes | Exposure to this product | Rough size | Position |
|---|---|---|---|---|
| ASML (ASML) | EUV + DUV lithography systems, plus service | Pure-play (~100% of revenue is lithography) | ~$30-35B revenue est. | Sole EUV maker; DUV majority; multi-year backlog. |
| Lasertec (LSRCY) | EUV photomask inspection tools | Near pure-play on the EUV ecosystem | Small / mid-cap est. | Dominant supplier in EUV mask inspection — a single-product niche. |
| Onto Innovation (ONTO) | Metrology & inspection (process control — measuring and checking chips as they are made), not lithography itself | Adjacent, not lithography per se | Small / mid-cap est. | Sells into the same fabs; revenue tracks wafer-start (the number of wafers entering production) growth. |
| Nikon / Canon (Japan) | DUV lithography (no EUV) | Small slice of large camera/optics conglomerates | Large, diversified est. | Minority DUV share; exited the EUV race years ago. |
| Zeiss SMT (private, German) | EUV optics / mirrors | Sole critical optics supplier to ASML | Private est. | Not directly investable; ASML owns a stake. Central to the EUV bottleneck. |
Source: 500-stocks scan (ASML, Lasertec, Onto as US-listed names); Nikon/Canon/Zeiss roles and sizes from general knowledge, not live-verified.
In plain money terms: as an owner you are buying a stream of future machine sales plus a growing recurring service stream. The natural questions are how many dollars of market value you pay per dollar of revenue, and how much of that revenue actually turns into owner cash.
The money-in / money-out shape. ASML's structure is unusual: it makes one of the heaviest capital goods in the world, yet it is itself relatively asset-light — it designs and integrates the machine and pushes much of the heavy manufacturing onto its supplier network. On the reported figures, a high share of revenue converts to free cash, and the company has historically returned large amounts to owners via dividends and buybacks. So you pay a high multiple of revenue (roughly software-like, not machinery-like), in exchange for that cash-conversion profile. The smaller names (Lasertec, Onto) have tended to move more sharply than ASML for a given change in industry demand, with narrower competitive barriers than ASML and more cyclicality.
Neutral arithmetic to internalise. If you pay ~$10 of market value per $1 of revenue and the business keeps ~40 cents of EBITDA per revenue dollar, you are paying roughly 25× current operating cash (10 ÷ 0.40) est. — a level that is only supported if revenue grows for years. The question for an owner reduces to: how long, and how steeply, does the EUV machine count compound? That is exactly the demand-vs-supply gap above. The reader judges whether the price matches that growth.
Source: price-to-sales, margin and capital-intensity figures from general knowledge, not live-verified; verify against ASML's latest filings.
Where a company-level deep-dive would be most informative, factually:
Source: 500-stocks scan (player list); characterisations from general knowledge, not live-verified.
What was used:
/Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/02-semiconductors.html. Grounds the hard structural facts: ASML's EUV monopoly, ~50-60 systems/year, multi-year backlog, ~$350M+ per machine, hardest-bottleneck-to-replicate framing, and the US-listed player list (ASML, Lasertec/LSRCY, Onto/ONTO).Hard vs approximate:
Confidence: Higher on the qualitative structure (single EUV maker, structurally short on the facts presented, asset-light cash-conversion profile, AI-direct demand link). Lower on exact figures, which were not live-verified — reconfirm revenue, order intake, backlog and valuation against ASML's most recent quarterly report before acting.
Source: 500-stocks semiconductors scan section; general knowledge as labelled. Live web retrieval was unavailable at build time.