Nvidia did not predict AI, it built the tools that let others find it
Intel saw little graphics helpers for gamers, while Nvidia kept improving the software that made those chips useful for science, money, and AI.
All-In with Chamath, Jason, Sacks & Friedberg · Former Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe CodingThe secret was not foresight. It was tools that let other people find the future first.
That is how Pat Gelsinger, the former chief executive of Intel, explains the fall of Intel and the rise of Nvidia. When Intel ruled personal computers with its central processor, the chip that does most everyday computing, its engineers looked at Nvidia's graphics chip, the chip that draws video game pictures, and dismissed it. In Gelsinger's telling, the attitude was that Intel made the big central processor and Nvidia made the little graphics helper for gamers.
Nvidia changed the meaning of that little chip by building software around it. The key piece was CUDA, a programming system that let ordinary developers use a graphics chip for math that had nothing to do with pictures. A graphics chip is built to do thousands of simple calculations at once, instead of a few calculations very fast. CUDA made that ability usable, and Nvidia kept improving it release after release.
The turning point came from outside Nvidia. Researchers in supercomputing in Japan, people whose job is running giant science calculations, realized those graphics cards were a more computationally dense platform for their work. Gelsinger calls that a defining moment. It proved the cards were no longer just for graphics, but general purpose computers that could be aimed at many hard problems.
Intel tried a rival idea called Larrabee, which tried to get the same result by starting from Intel's standard PC chip design. Gelsinger says it was killed a week after he first left Intel. Nvidia, by contrast, kept working on the slow software, improving it generation after generation.
So Nvidia did not predict that its chips would become the best way to handle Bitcoin computations or train artificial intelligence. As Gelsinger puts it, quoting writer William Gibson, "the street finds its own use for technology." Nvidia built the software foundation, and the hacker community found the uses, including Bitcoin and AI, that made it valuable.
Questions people ask
Why is a graphics chip so good at AI?
A graphics chip is designed to handle thousands of small math jobs at the same time, because every dot on a game screen can be calculated separately. That same design turns out to be ideal for AI, which also requires huge numbers of parallel calculations to adjust its digital connections.
What exactly is CUDA?
CUDA is Nvidia's system of programming tools, libraries, and instructions that lets developers write programs for its graphics chips. Without it, only graphics experts could use the chips well, and with it, scientists, engineers, and AI researchers could treat them like general purpose computers.
What happened to Intel's Larrabee project?
Larrabee was Intel's late 2000s effort to build a graphics and parallel computing chip starting from standard Intel processor technology. Intel canceled it as a consumer graphics product, and its ideas later lived on in supercomputer chips, but it never produced a rival software community like CUDA.