“Bryan Catanzaro had spent many years trying to get his colleagues at Nvidia to care about deep learning, but nobody seemed too interested in something that seemed like alchemy—at least not when it was coming from a research scientist who fell squarely in the middle of the company’s totem pole. It was only in 2013 that Catanzaro’s old CEO at Nvidia started to really pay attention to what he had to say. Catanzaro explained why he thought deep learning—a process by which computers’ data-processing mimics, to an extent, the human brain—was so key to the future of AI. Jensen, to Cantanzato’s surprise, was all in, he tells me. “All of a sudden Jensen started caring a lot,” Catanzaro says via video chat from his California home. “It seemed too good to be true.”
As a result, one whole decade before this year’s AI boom, a graphics company mostly known to gaming nerds and PC enthusiasts decided to bet the house on something that, at the time, was as unreal as Mark Zuckerberg’s metaverse and offered no immediate bottom-line appeal. Initially, Catanzaro was suspicious of whether Jensen’s interest wasn’t fleeting or perfunctory.
“I didn’t think that it was actually even possible to focus Nvidia on something like this,” he says, “because, at the time, it was a different company.” But those doubts were misplaced: Almost instantly Nvidia made a hard pivot, kicking off a mad dash to transform itself into an AI powerhouse.
These days, Catanzaro is the vice president of Nvidia’s Applied Deep Learning Research group, the AI research lab that paved the way to the current state of generative AI, thanks to its massively parallel chips that power data centers all over the world. Catanzaro is involved in all sorts of interesting projects: improving graphics quality via generative AI with its Deep Learning Super Sampling technologies; creating the Megatron transformer, which can build and train natural language processing models with billions and trillions of parameters; and helping to better integrate deep learning ideas into a number of user-friendly products and applications. Who knows, maybe one day his team will even develop a more helpful generative AI tutor for struggling students. “We are willing to do what it takes to see our ideas all the way through to the customer,” he says.
And the company’s coffers have only swelled alongside its ambitions. Last quarter, Nvidia earned $13.5 billion, up 101% over the prior year, way well over its $11 billion estimation. Still, Catanzaro says he doesn’t forget the company’s gaming origins. “I think that gaming is going to become the output mechanism for generative AI”, he says. After all, he says, it’s unlikely that eight billion people are going to interact with AI primarily by typing into text boxes; people simply don’t write and they don’t read that much.
“It’s much more natural to interact like the way we’re interacting now,” he says, “which is leading to generative AI as a new form of media, one that is going to be bigger than any other form of media.”
This story is part of AI 20, our monthlong series of profiles spotlighting the most influential people building, designing, regulating, and litigating AI today.”
