ASML

You cannot prompt-engineer your way to extreme ultraviolet lithography — Claude doesn't know how to aim 13.5-nanometer wavelength light through mirrors polished to 1/10,000th the width of a human hair. The business model is 'sell irreplaceable physical infrastructure to make the chips that run AI,' which turns out to be the most AI-resistant revenue stream ever conceived.

Their engineers design machines that etch patterns 1,000 times smaller than a human hair using mirrors that must be perfect to the atomic level — automation helps with simulation and design optimization, but the core expertise remains stubbornly analog. You still need humans who understand both quantum physics and mechanical engineering well enough to make light behave like a really expensive pencil.

ASML's moat is forty years of accumulated precision manufacturing knowledge, $20 billion in R&D sunk costs, and a global supply chain so specialized that their mirrors come from one German company and their lasers from one American lab. Competitors aren't trying to build a better mousetrap — they're trying to reinvent physics, optics, and materials science simultaneously.

Their 'computational lithography software' and AI-enhanced process optimization show they're absorbing machine learning where it helps, but the core value proposition remains wonderfully immune — somebody still has to manufacture the chips that run ChatGPT, and ASML owns the only printing press.

AI doesn't eliminate the need for semiconductors — it multiplies it exponentially. Every ChatGPT query, every autonomous vehicle, every smart device requires chips that can only be made with ASML's machines. They're not just surviving the AI revolution — they're the landlord collecting rent on every AI breakthrough ever made.