Pick Problems That Take a Decade

At Stripe Sessions in 2024, Huang gave the logic of choosing difficulty: “I like hard things, because it takes a long time to do.” Asked how long CUDA took to pay off, he answered: “Probably 10 years, yeah.”

The same patience, from other stages. At Caltech in 2024 he called CUDA “The programming model that we dedicated over 20 years to invent, and that is revolutionizing computing today.” On Acquired in 2023 he said: “we spent about a decade in $0 billion markets.”

The underlying belief. Tae Kim, in The Nvidia Way, attributes to Huang the conviction that accelerated computing would solve problems that normal computers couldn’t. That belief is what he was waiting out.

Doesn’t transfer to impatient capital. A decade-long bet needs shareholders who will tolerate a depressed stock and a core business that throws off cash meanwhile. Most founders don’t control their board or their runway that tightly, and plenty of ten-year bets turn out to be ten-year mistakes that nobody writes up.