Lesson 17 · Chips & LLMs

ASML & EUV: The Monopoly at the Foot of the Chain

Why one Dutch company is the single most unreplaceable link in the entire semiconductor value chain — and the most decisive geopolitical lever in the China question.

Builds on: L1 (value chain), L3 (China), L4 (TSMC) Confirms: THESIS.md — ASML

In your checkpoint, the "sand → served token" trace started with lithography, and your value-chain card named ASML as the one true monopoly in the stack. This lesson earns that claim from first principles. The goal: be able to explain, to a skeptic, exactly why EUV is unreplicable, why that translates into durable pricing power, and why export controls on a single machine can throttle a nation's chip ambitions.

Core thesis: Every transistor's size is set by how finely you can print a pattern, and printing resolution is governed by the wavelength of light you use. The industry was stuck at 193 nm light for ~20 years. ASML is the only company on earth that productized 13.5 nm (EUV) light — a feat requiring tin-plasma sources, all-mirror optics, and a vacuum the size of a bus. There is no second supplier, no near-term substitute, and no path to leading-edge chips without them. That is the deepest moat in the industry.

01 — The One Equation That Governs Everything

Resolution Is Set by Wavelength

Photolithography prints a circuit pattern by shining light through a mask (the stencil) onto a light-sensitive photoresist on the wafer. The smallest feature you can resolve follows the Rayleigh criterion:

CD = k₁ × ( λ / NA ) CD = critical dimension (smallest printable feature) · λ = wavelength of light · NA = numerical aperture of the optics · k₁ = process factor (~0.25 floor)

To print smaller features you have only three levers: shorter wavelength (λ), bigger numerical aperture (NA), or a lower k₁ (process tricks). For two decades the industry was pinned on the second and third because the first hit a wall.

02 — The 193 nm Wall and the Multi-Patterning Tax

Twenty Years Stuck at One Wavelength

Light sourceWavelengthEra / nodeHow resolution was pushed
i-line (mercury)365 nm~1990s, >350 nmShorter λ each generation
KrF excimer248 nm~250–180 nmShorter λ
ArF excimer (DUV)193 nm~130–65 nmShorter λ
ArF immersion (193i)193 nm~45 nm and belowWater (n≈1.44) between lens and wafer raises effective NA to ~1.35
193i + multi-patterning193 nmdown to ~7 nm (no EUV)Print one layer with 2–4 masks & exposures — costly & yield-eating
EUV13.5 nm7 nm → 2 nm and beyondA 14× shorter wavelength — one exposure replaces many
High-NA EUV13.5 nm~2 nm / sub-2 nmNA 0.33 → 0.55 for finer resolution still

Why the wall mattered. When you can't shorten λ, the only way to keep shrinking is to split one layer into multiple masks and exposures — multi-patterning. Quadruple-patterning means 4× the masks, 4× the exposure passes, more deposition/etch steps, and more chances for misalignment. Cost and cycle time balloon; yield suffers. EUV's whole economic justification is collapsing several patterning steps back into a single exposure.

Connect to L3 (China): SMIC can still reach ~7 nm using 193i multi-patterning — which is why "China made a 7 nm chip" headlines appear. But without EUV it is slow, low-yield, and economically punishing, and it does not scale to 3 nm / 2 nm. The wall is exactly where the chokepoint bites.

03 — Why 13.5 nm Light Is Almost Impossibly Hard

What ASML Actually Had to Solve

Going from 193 nm to 13.5 nm isn't an incremental step — it crosses into a regime where the laws of optics flip. At 13.5 nm, everything absorbs the light: air, glass lenses, even the mask substrate. So nearly every assumption of conventional lithography had to be discarded.

EUV light generation — laser-produced tin plasma
Tin droplet generator ~50,000 droplets / second CO₂ laser plasma ~500,000 °C emits 13.5 nm EUV collector All-reflective optics ~10 Zeiss Mo/Si mirrors ~70% reflective each → ~few % of light survives Entire path is in vacuum — air absorbs EUV. No lenses anywhere; glass absorbs EUV too.
A CO₂ laser vaporizes ~50,000 tin droplets per second; each flash emits 13.5 nm EUV. A collector mirror gathers it, ~10 multilayer Zeiss mirrors steer and shrink the pattern, and only a few percent of the original light reaches the wafer — which is why source power (250–500 W) is the perennial throughput bottleneck.
The source
Tin-plasma light, not a lamp
You can't buy a 13.5 nm bulb. ASML's source (built on its acquired Cymer + TRUMPF lasers) blasts molten tin droplets with a high-power CO₂ laser ~50,000×/sec, creating a plasma hotter than the sun's surface that radiates EUV. Wall-plug-to-EUV efficiency is well under 1%.
The optics
Mirrors, not lenses
EUV passes through no material, so there are no lenses — only mirrors, made by Zeiss from ~50 alternating molybdenum/silicon layers tuned to reflect 13.5 nm. Each reflects ~70%; after a dozen bounces only a sliver of light remains. The mirrors are the flattest objects humans manufacture.
The mask
Reflective reticle
Because glass absorbs EUV, the mask is reflective, not transmissive — itself a multilayer mirror with the pattern etched on top. Defect-free EUV mask blanks and EUV pellicles are their own hard sub-industries.
The system
A vacuum the size of a bus
Air absorbs EUV, so the whole light path runs in vacuum. A single machine has ~100,000 parts from ~5,000 suppliers, weighs ~180 tons, ships in 40+ freight containers, and prints ~150–200 wafers/hour.
04 — Why There Is Exactly One Supplier

The Monopoly Is the Supply Chain

ASML's moat isn't a patent — it's that productizing EUV took ~30 years and tens of billions of cumulative R&D, plus the simultaneous solution of source, optics, mask, and system, each dependent on a sole-source partner ASML either owns or has locked up:

Intel and Nikon/Canon (the old DUV competitors) tried EUV and effectively gave up; the capital and integration risk were too high. The result: ASML has ~100% share of EUV and the large majority of leading-edge DUV. A leading-edge EUV machine runs ~$150–200M; a High-NA system is ~$350–400M. Only TSMC, Samsung, Intel, and the memory makers (SK Hynix, Micron, Samsung) can buy and run them at all.

This is what "picks-and-shovels at the foot of the chain" means: ASML gets paid whether TSMC, Samsung, or Intel wins the foundry war — and (via the GPU bill of materials) whether NVIDIA or AMD or a custom ASIC wins the accelerator socket. Its falsifier is independent of every downstream battle.

05 — High-NA EUV: The Next Lever

NA 0.33 → 0.55

With wavelength frozen at 13.5 nm, the next resolution gain comes from the other term in the Rayleigh equation: numerical aperture. High-NA EUV raises NA from 0.33 to 0.55, shrinking the smallest printable feature by ~1.7× and again replacing multi-patterned EUV steps with single exposures at the 2 nm-and-below frontier.

The trade-offs are real: High-NA uses anamorphic optics (different magnification in x and y), which halves the usable field — so large dies must be stitched from two exposures. Each machine is ~$380M+. Intel took first delivery (Twinscan EXE series) as part of its 18A/14A bet, ahead of TSMC — a rare case of Intel leading on a tool, and a key variable in whether Intel Foundry can re-enter the leading edge.

06 — The Geopolitical Lever

One Machine, One Chokepoint

Because ASML is the sole gate to leading-edge fabrication, it is the most precise instrument of chip-export policy that exists. The Netherlands (under US coordination) has never allowed EUV sales to China, and from 2023–2024 progressively restricted advanced DUV immersion tools (e.g. the NXT:2000i class) as well.

07 — Investment Implications

Monopoly Economics, Cyclical Orders

Pricing power + recurring revenue
Monopoly on the gating tool → ~50%+ gross margin and pricing power on each system. Critically, a huge and growing installed-base / service & upgrade business smooths the cycle: ASML keeps earning on every machine already in the field. Watch service revenue as the recurring-quality signal.
The order book is the lead indicator
EUV orders are lumpy and tied to the foundry/memory capex cycle — they swing with TSMC/Samsung/Intel buildout decisions and the AI capex wave. Net bookings (especially EUV + High-NA) are the cleanest forward read on leading-edge expansion. A bookings air-pocket can precede a foundry capex pause.
High-NA = the next upgrade super-cycle
At ~$380M/system, High-NA adoption (Intel first, then TSMC/Samsung) is a multi-year revenue ramp and an ASP step-up. Its pace depends on whether sub-2 nm scaling stays economically worthwhile — the key swing factor in the bull case.
China is the two-sided wildcard
Tighter controls cut near-term China sales but reinforce the structural chokepoint that makes ASML strategically irreplaceable. The tail risk is a credible domestic EUV (SMEE) — currently remote. Track Dutch/US control changes and any SMEE EUV milestone.
Thesis Update

ASML in THESIS.md — Now Grounded (★★★)

Promote the ASML entry from stub to grounded. Key technical assumption: EUV remains a single-source monopoly (Zeiss optics, Cymer/TRUMPF source, 5,000-supplier integration) that no rival can replicate this decade. Falsifier — the signals that would change your mind:

Primary Source

Go Deeper

Watch first: Asianometry — "How ASML Won Lithography" / EUV explainers — the clearest technical history of why EUV is so hard and how ASML cornered it. High trust, engineer-level but accessible.

Read: ASML — EUV Lithography (official technology pages) for the primary-source mechanics, and Chip War (Chris Miller, in RESOURCES.md), chapters on lithography and the Dutch chokepoint, for the geopolitical framing.

Comprehension Check

Quiz — 5 Questions

Select the best answer for each.

1. According to the Rayleigh criterion CD = k₁·λ/NA, EUV's primary advantage over 193i comes from changing which term?

A much larger k₁ process factor
A ~14× shorter wavelength λ (193 nm → 13.5 nm)
A water immersion layer that raises NA above 1.0
A larger mask that exposes more die area per pass

2. Why does an EUV scanner use mirrors instead of glass lenses?

Mirrors are cheaper to manufacture than precision lenses
13.5 nm light is absorbed by glass (and air), so refractive optics are impossible
Mirrors allow the wafer stage to move faster during exposure
Lenses cannot be made flat enough for any lithography

3. China's SMIC can reach ~7 nm without EUV. Why is this not a refutation of the EUV chokepoint?

SMIC actually uses smuggled EUV tools for those chips
7 nm via 193i multi-patterning is low-yield, costly, and doesn't scale to 3/2 nm
7 nm is already the smallest node any chip will ever need
SMIC licenses TSMC's process to reach 7 nm

4. Which best explains why ASML has ~100% EUV market share — i.e., the real moat?

A single broad patent that blocks all competitors until 2040
~30 years and tens of billions to co-solve source, optics, mask, and a 5,000-supplier integration with sole-source partners
Government regulation that bans any other firm from making scanners
EUV machines are simply cheaper than competitors' DUV tools

5. For an investor, which is the cleanest forward-looking indicator of leading-edge expansion at ASML?

Trailing reported revenue from last quarter
Net new system bookings, especially EUV and High-NA
The spot price of tin used in the light source
The number of patents filed during the year
From your instructor: The single sentence to walk away with — resolution is set by wavelength, and ASML is the only company that productized 13.5 nm light. Everything else (the monopoly, the pricing power, the China lever) follows from that. Ask me anything: the tin-plasma physics, why High-NA halves the field, or how to read ASML's bookings against the foundry capex cycle.