Somewhere in the Netherlands sits the only machine on Earth that can print the cutting edge of computing. Every advanced chip — your phone's processor, NVIDIA's GPUs, the next AI accelerator — passes through an ASML lithography system to be born. If chips are oil in the modern economy, ASML is the wellhead, and it's a wellhead nobody else knows how to build.
What lithography actually does
Chipmaking is photography taken to a deranged extreme. A photomask carrying the circuit pattern is projected onto a silicon wafer coated in light-sensitive resist, and the image is printed, developed and etched into the wafer. Shrink the features and everything downstream — transistor density, power, cost — follows. For four decades the industry chased shorter wavelengths, from mercury lamps to argon-fluoride deep ultraviolet (193nm), stretching 193nm immersion lithography absurdly far with multi-patterning tricks. Eventually physics demanded a new light source.
Enter EUV: 13.5 nanometers of fury
Extreme ultraviolet light at 13.5nm can print features no 193nm system can match in a single exposure, collapsing many patterning steps into one. The catch: EUV photons are absorbed by almost everything, including air and regular glass. So an EUV machine is a vacuum system, lit by hydrogen plasma droplets hit by a laser 50,000 times a second, with mirrors ground so perfectly that scaling one up to the size of Germany would keep the bumps below a millimeter. These are facts, not flourishes. No refractive lens can focus 13.5nm light, so the entire optical path is reflective, and every mirror coat (alternating molybdenum and silicon) is a marvel of materials science.
The current workhorse is the NXE series — NXE:3600D and successors — delivering around 170 to 200+ wafers per hour, depending on the generation, at a list price somewhere north of $150–200 million each. High-NA systems (EXE:5000 and onward), with a 0.55 numerical aperture instead of 0.33, push resolution further and cost reportedly around $350–400 million per machine. Each one ships in dozens of containers and takes weeks to install, level and calibrate.
High-NA: the next expensive miracle
Numerical aperture is how much light the lens can gather; more NA means finer resolution. Jumping from 0.33 to 0.55 sounds modest and is anything but: the optics, the stage mechanics and the mask handling all had to be rethought. High-NA EUV lets fabs print smaller features with fewer multi-patterning passes, which offsets some of its eye-watering cost. Intel is the early adopter, planning High-NA for its 18A/14A-class nodes and future processors, while TSMC is expected to phase it in later, likely around its A16/A14 era. Samsung is also in the conversation.
High-NA's economics only work if throughput and yield hold up. Early systems print well but not yet at full NXE productivity, and every fab boss on Earth is watching the ramp specs like a hawk before committing billions to the transition.
Why ASML has no competition
ASML is the only supplier of EUV lithography systems, full stop. The machine aggregates technology that doesn't exist anywhere else: Zeiss SMT builds the optics (ASML owns a majority stake), the light source comes from a decades-long iteration on laser-produced plasma, and the whole system stitches thousands of suppliers into a single product. Reproducing that stack would take a competitor $20+ billion and a decade with no guarantee. This isn't a market; it's an institution — like the Dutch patiently cornering high-end microscopes, then high-end scanners, for a century.
The natural monopoly is also a geopolitical fault line. Export controls bar the sale of EUV systems to China, and even deep-UV immersion tools face tightening restrictions. Every leading-edge chip in a US cloud or an allied fab exists partly because of a Dutch export license decision. Policymakers understand this better than almost anything else in tech.
What ASML's pace means for chips
ASML ships on the order of a few hundred EUV systems per year, increasingly High-NA. Each one placed is a long-term commitment from a fab: years of depreciation, service contracts, spare parts and process lock-in. The machine you buy today dictates your process economics for a decade. That's why TSMC, Intel, Samsung, Micron (for memory patterning) and even some ecosystem players queue up for systems years in advance. If you want to know when 2nm really gets cheap, don't ask the chip designers — ask ASML's order book.
It's also worth remembering the lithography roadmap doesn't end at High-NA. Research into hyper-NA beyond 0.55, new resist materials, and even alternative patterning approaches continues in parallel. But nothing credible replaces EUV at the leading edge this decade, and probably not next decade either.
The 50,000 droplets per second trick
The detail that keeps physicists amused: an EUV source works by blasting tiny molten tin droplets with a laser. Each droplet is hit twice — first flattened, then vaporized into a plasma that spits out EUV photons — and this happens around 50,000 times a second. Miss the droplet and you miss billions of photons per second; the timing tolerance is on the order of nanoseconds. Collecting those photons is the next miracle: a multilayer mirror stack, alternating molybdenum and silicon, with each layer smooth to well under a nanometer, because roughness scatters EUV light into uselessness. Combine a tin fountain, a stadium-scale laser and German-precision optics, and you get... 13.5-nanometer light, still fragile, still hungry for vacuum, always on the edge of failure.
Service is where the real money hides. ASML systems are sold with long-term service agreements; every swap of a source collector, every recalibration of a mirror, goes through trained ASML engineers. Uptime guarantees are contractually meaningful because a single day of downtime at a leading-edge fab costs more than the service contract does.
The supply chain inside the supply chain
An EUV machine is really a symphony of monopolies. ASML makes the system; Zeiss makes the optics; Trumpf supplies the laser technology; Sanmina handles complex subassemblies; and hundreds of smaller suppliers contribute vacuum components, wafer stages and sensors. Disrupt any one of them and the line stops. This deep-tier fragility is why fabs hold strategic spares and why ASML's installed base is effectively a permanent service franchise in Taiwan, Korea, the US, Japan and Germany.
It's also why export controls work so decisively. You can block the machine, and even the spare parts and service know-how — meaning a fab that lost its ASML support could hobble within quarters. Policymakers didn't just restrict exports; they restricted physics.
What the High-NA order book reveals
Follow the early customers and you can see the roadmap in public. Intel has taken early High-NA systems (EXE:5000 class) to Hillsboro for development, framing 14A around the technology. TSMC has shown High-NA interest but will deploy later, when throughput and economics mature — its A14 timeframe is the likely window. Micron is evaluating High-NA for advanced memory patterning, where smaller pitches buy real density. Samsung Foundry, perpetually squeezed, sees High-NA as a chance to leapfrog TSMC on paper and is racing to operationalize it.
Each High-NA machine prints fewer wafers per hour than its predecessor in early life, so the cost per wafer starts brutal. The industry's bet is that throughput doubles or triples over a few years, the way NXE systems did, turning an exotic early system into the default tool of the next process generation.
Resist chemistry: the quiet crisis inside EUV
EUV prints with photons that each carry enough energy to damage the resist polymer itself. Traditional chemical amplification breaks down under that bombardment, so the industry has chased metal-oxide and other so-called EUV resists to keep pattern fidelity at tiny dimensions. Resist suppliers — a niche oligopoly of Japanese and Korean chemical giants — iterate in lockstep with ASML's optics roadmap. No resist breakthrough, no shrink; no shrink, no node. It's one more reason lithography progress is a convoy, never a solo sprint, and why every EUV forecast is hedged about materials as much as machines.
The installed base effect
Every EUV system shipped is a decade-long annuity in disguise — parts, upgrades, training and constant recalibration. That installed base is why ASML's revenue is unusually predictable for a capital-equipment maker, and why its gross margins hover far above traditional chip-equipment rivals. It also means the leading-edge fabs of the 2030s are already partially locked in: switching patterning philosophy would strand billions in depreciated machines. The industry doesn't just buy ASML tools; it marries them, with maintenance included.
The people problem inside a $400 million machine
For all the physics, an EUV system is run by humans, and there aren't nearly enough of them. Each machine needs a team of factory-trained ASML engineers for install, and the fabs keep their own small cast of super-specialists for daily babysitting: source collectors swapped on schedule, mirrors checked for contamination, tin debris cleaned out of the chamber. Training a qualified EUV technician takes years, and the pool is concentrated in a few fab clusters — Hsinchu, Hwaseong, Hillsboro, Columbus.
This is the quiet fragility nobody prices in. When a leading-edge fab says it's “running at capacity,” what it often means is that its technicians are the binding constraint on how many machines can run simultaneously, not the machines themselves. Labor scarcity is why ASML's service backlog matters as much as its order book, and why even fabs that can afford the hardware still wait months for qualified install days.
It also explains the industry's fixation on uptime contracts. A fab that loses a senior EUV engineer to a competitor loses weeks of effective capacity while a replacement ramps. Chips are printed by people who happen to operate lasers the size of garages, and anyone planning around leading-edge supply would do well to remember that.
The bottom line
ASML doesn't make chips; it makes chips possible. That distinction is why it's arguably the most important single company in the AI supply chain — more than any one chip designer, more than any memory maker, more than most cloud providers combined in strategic terms. The machine that prints the future is Dutch, expensive, irreplaceable, and quietly running behind everything you love or fear about modern AI. And if you ever get the chance to stand next to one of these machines in person, do it — nobody's jaw stays on.