Jensen Huang founded Nvidia in 1993 with the idea that graphics processing would eventually need to do things that traditional chip architectures simply could not handle efficiently. He was right, but not quite in the way he expected.
The chips Nvidia built to render video game graphics — GPUs — turn out to be exceptionally well-suited to the mathematical operations that underpin modern AI. Training a large language model is, at its core, a massive exercise in matrix multiplication, the same operation that renders the lighting in a video game.
The company’s H100 chip, launched in 2022, became the defining piece of hardware of the AI boom. Every major AI lab — OpenAI, Google DeepMind, Anthropic, Meta — built their training infrastructure around it. At its peak, H100 servers were so scarce that companies were chartering private jets to transport them from factories.
By 2024, Nvidia’s data centre revenue had grown to more than $90 billion annually, up from $15 billion two years earlier. The company’s market capitalisation surpassed Apple, Microsoft, and for brief moments, every other company ever measured.
Jensen Huang’s response to concerns about future demand has been characteristically direct. “Demand is enormous. Every country, every company, every industry is going to need to build AI infrastructure. We’re at the very beginning.”
Leave a Reply