AINEWS
The $400 million machine powering the future of chipmaking
ASML, a Dutch company, has a near monopoly on the lithography machines that pattern the world's most advanced microchips. The question is whether anyone can catch up.
ASML, a Dutch company, has a near monopoly on the lithography machines that pattern the world's most advanced microchips. The question is whether anyone can catch up.
Every advanced chip ever made, from the iPhone processor in your pocket to the data-center GPUs training the latest AI models, passes through a machine built by a single company in the Netherlands. ASML's Twinscan EXE:5200 High-NA EUV lithography system costs about $400 million per unit, weighs 150 tons, and occupies more than 200 cubic meters of floor space. It sits at the absolute chokepoint of the AI compute stack. The question nobody in the industry wants to answer out loud is what happens to the AI buildout when the only company that can build the next generation of chips is a Dutch monopoly with a 30-year head start.
High-NA EUV is the successor to standard EUV, the extreme-ultraviolet light technology ASML pioneered and now dominates. EUV lithography works by firing light at wavelengths of 13.5 nanometers onto a silicon wafer to imprint the fine circuit patterns that define a modern chip. High-NA pushes that further, using higher-numerical-aperture optics to print even smaller features, which is what lets chipmakers push past the 3-nanometer node. ASML spent roughly 16 years and around $10 billion getting EUV out of the lab and into high-volume manufacturing, and the High-NA generation is now the platform for the next round of logic and memory scaling.
The physical scale of the tool is part of the moat. MIT Technology Review's reporting describes the High-NA system as the size of a double-decker bus, with 200 or more cubic meters of precision-machined aluminum, tubes, and pressurized tanks. Jos Benschop, ASML's executive vice president of technology, has led the design of this generation for more than a decade. The system holds a few mirrors at near-atomic precision using mechatronic positioning, a stack of actuators, sensors, and control loops that effectively cancels vibration and gravity well enough to print features measured in atoms. Building one requires a supply chain that runs through Germany, Japan, and the United States; the light source comes from Cymer in San Diego, the optics from Carl Zeiss in Oberkochen, and a long tail of precision mechanics from partners across Europe and Asia.
That is why no one has caught up. Canon and Nikon, the Japanese incumbents in lithography, have not shipped a competitive EUV system. The capital, the optical engineering, the light-source physics, and the decades of integration with chipmakers' roadmaps form a barrier that is more about accumulated know-how than about a single patent. The dominant model in the industry is that ASML is the only company on Earth that can build these tools at scale, and that every advanced fab, whether in Taiwan, South Korea, the United States, or Japan, depends on its delivery schedule.
Two credible challengers are trying to break that. The first is Substrate, a US-based startup that is positioning itself as an alternative-lithography platform. The second is China's parallel domestic effort, documented in a Reuters investigation as a multi-decade state-backed attempt to build an indigenous chipmaking stack after successive rounds of US export controls cut China off from the most advanced ASML systems. Both efforts face structural disadvantages that go beyond funding: Substrate is racing a fully integrated incumbent with a working supply chain, and China's program is constrained by access to the highest-end EUV optics and metrology.
The real risk is not that someone catches ASML next year. It is that the AI buildout of the late 2020s depends on a single supplier for its most critical input, with no second source. A fire, a sanctions escalation, a yield problem at Zeiss, or a delivery slip at ASML can ripple through the entire AI roadmap. The bottleneck of the AI economy is not compute, model weights, or even power, but the lithography tools that print the chips everything else runs on. The next test is whether Substrate can ship a working alternative at any scale, whether China's domestic program breaks through the optics barrier, and whether ASML's own throughput can keep up with the demand curve for the most expensive manufacturing tool ever sold.