Eric Siu breaks down four physical bottlenecks that could slow AI growth through 2030: indium phosphide wafers made by only a handful of non-Chinese companies, limited advanced packaging capacity concentrated in four firms, power constraints with industrial gas turbines sold out, and critical minerals like gallium facing a Chinese export pause ending November 2026.
What is the smallest object, Neil, that if it stopped being made tomorrow would freeze the entire AI industry by Friday? Memory. No, it's a polished piece of indium phosphide the size of a coaster grown in a furnace over 2 weeks, made by exactly 2 companies in the world that are not Chinese. It's not a chip, it's not a GPU, it's not a model. The 2026 to 2030 AI buildup is gated by 4 physical constraints and almost nothing else. Number one is indium phosphide wafers, advanced packaging, 4 companies on Earth that matter, power, Okay, industrial gas turbines sold out into 2030. And you have critical minerals. Chinese pause on gallium, germanium, and antimony export controls expires November 27th, 2026.
