A new generation of chipmaking equipment is beginning to reach fabs, and its importance is hard to overstate. ASML’s latest lithography machine is enormous, expensive and technically extreme, but it sits at the center of a simple question: how long can the industry keep making chips denser, faster and more useful for artificial intelligence?
The answer matters because modern AI systems depend on powerful hardware. The companies building and deploying advanced models need chips that can perform more calculations while using energy efficiently. That pressure has made one Dutch company, ASML, an unavoidable force in the semiconductor supply chain.
The machine behind smaller chip features
Lithography is the chipmaking step that transfers tiny patterns onto a silicon wafer. Those patterns define the transistors, wiring and other components that eventually become microchips. In plain terms, the lithography machine helps draw the circuitry that makes a chip work.
ASML’s newest system is the size of a double-decker bus and weighs more than 150 tons. Jos Benschop, ASML’s executive vice president of technology, describes the machine as more than 200 cubic meters of technology built around “mechatronic devices that hold a few mirrors in a position with atomic precision.”
The price is just as striking as the scale. The new machines are shipping to chipmaking factories, or fabs, at $400 million each. That cost is high, but chipmakers are racing to build newer chips every year, and that race depends on tools that can create smaller components and place them more densely together.
The new machine improves on ASML’s earlier extreme-ultraviolet, or EUV, systems. Those first EUV machines could create transistor features with a resolution of 13 nanometers. The latest system reaches a resolution of just eight nanometers, described in the source as the width of about 40 silicon atoms.
Why AI raises the stakes
The demand for better chips is not abstract. Firms such as OpenAI and Anthropic are building server farms to train and deploy more powerful AI models, and those models require increasingly powerful hardware. The source article describes this as a major new source of demand for denser chips.
ASML’s chief technology officer, Marco Pieters, links the machine directly to the computing needs now emerging around AI. “We can allow customers to go to smaller and smaller features, and that opens up the space for whatever we see now today in AI, which is absolutely mind-blowing,” he said. “I think we’ve only seen the tip of the iceberg.”
The logic is straightforward. Smaller chip features make it possible to fit more circuitry into a given space. That has long been part of the recipe for faster and more energy-efficient chips. For years, ASML’s tools have helped keep Moore’s Law alive by allowing chip density to continue improving rather than plateauing.
The source also notes how quickly AI changed the demand picture. After EUV machines went on the market in 2017, some questioned how quickly major chipmaking firms would need them. But after OpenAI released GPT-3 and then ChatGPT, artificial intelligence moved into the mainstream, and companies including OpenAI, Google, Meta and Anthropic became hungry for high-end chips.
How ASML reached this position
ASML did not arrive at EUV lithography quickly. Nine years ago, the company began selling machines based on a new way of patterning chip features. The technology uses extreme-ultraviolet light, radiation outside the visible spectrum, created by firing lasers at tiny molten drops of tin tens of thousands of times a second.
That earlier EUV project took 16 years and about $10 billion in research. The machine works in a vacuum because EUV light is absorbed by regular glass lenses and even by air. Instead of ordinary lenses, it relies on mirrors to direct the light. Zeiss, a German optics company, had to develop new techniques for polishing and inspecting those mirrors.
ASML was not the only company to explore EUV. Nikon and Canon also worked on the approach, but they dropped out while ASML continued. Jeff Koch, a former ASML employee who is now an analyst at SemiAnalysis, described the company’s approach as intensely engineering-driven: “Let’s send thousands of engineers and just have them mow down these problems.”
That long bet paid off. ASML now produces about 90% of all chip-lithography tools worldwide, according to the source. If a company wants to make advanced chips, ASML is difficult to avoid.
A concentrated supply chain
ASML’s dominance is not only a business story. It also has geopolitical weight. The chipmaking field is described as being essentially controlled by two big players: ASML, which makes the lithography machines, and TSMC, which uses ASML’s machines to manufacture the vast majority of microchips.
That concentration has made governments uneasy. In 2019, the US government pressured the Dutch government to impose an embargo that prevents ASML from selling high-end machines to any Chinese firm. Marc Hijink, author of Focus: The ASML Way, sums up the stakes by saying that geopolitically, “chips are the new oil.”
James Proud, cofounder and CEO of the lithography startup Substrate, argues that the current structure is risky. “There’s a huge concentration in a small number of players,” he said. “And the supply chain is just very expensive.”
The source describes two kinds of pressure on ASML’s position:
- China is pouring billions into efforts to replicate ASML’s technology.
- Startups such as Substrate are trying to build lithography machines that are cheaper, smaller and more capable than ASML’s large systems.
For now, ASML’s lead remains substantial. The article says the near future clearly belongs to ASML, even as competitors try to find a way into its market.
The next step in chip shrink
ASML’s latest machine does not use a completely new kind of light. Instead, it raises what chipmakers call the numerical aperture. The source describes this as a move from an NA of 0.33 to an NA of 0.55.
That change matters because it can cut the size of transistors by close to half and nearly triple their density on a chip. The approach is called high-numerical-aperture EUV, or high NA. Compared with the original EUV breakthrough, it is described as evolutionary rather than revolutionary.
Even so, the engineering problems are severe. Higher numerical aperture changes how light reaches and reflects from the reticle, the mask that carries the chip design. Because the pattern on the reticle is three-dimensional, steeper light angles can create shadows and reduce the clarity of the pattern.
The solution required changes to the reticle pattern and to the optical system that shrinks and transfers that pattern onto the wafer. The reticle also moves with acceleration up to 22 g, much faster than in ASML’s original EUV machine.
The result is a tool built for the next phase of advanced chipmaking. It is huge, costly and difficult to copy. It also shows why the future of AI hardware may depend not only on chip designers, but on the machines that make their designs physically possible.