A.I. agents may be all the rage right now, but Nvidia CEO Jensen Huang is already looking beyond the hype. A.I.’s next phase will be “physical A.I.,” he said during Nvidia’s fourth-quarter earnings call yesterday (Feb. 26). Physical A.I. is a form of A.I. systems that can interact directly with the real world and encompasses products like autonomous vehicles and robots.
“Now is the beginning of the agentic A.I. era, and you hear a lot of people talking about it. Then, there’s physical A.I. after that,” Huang told analysts. He went on to explain, “They can’t just understand the meaning of words and languages, but they have to understand the meaning of the world, friction and inertia, object permanence and cause and effect.”
To this effect, Nvidia in January unveiled a platform of “world models,” which attempt to simulate real-world environments. Known as Cosmos, the new platform promises to advance the training of physical A.I. systems by providing highly realistic virtual worlds. Self-driving cars, for example, could be tested across various simulated weather conditions, while robotics companies could examine the safety of their products in different scenarios.
Cosmos has already been adopted by robotics companies like 1X and Neura Robotics and autonomous vehicle makers like Waabi and Wayve. The platform is also working with Uber (UBER) as it explores robotaxi products, Nvidia’s chief financial officer Colette Kress said yesterday. “Just as language foundation models have revolutionized language A.I., Cosmos is a physical A.I. to revolutionize robotics,” she added.
“World models” is a concept picking up steam in A.I. Last year, Fei-Fei Li, a renowned computer scientist and leading figure in A.I. research, founded World Labs, which looks to infuse A.I. with “spatial intelligence” that understands and interacts with the real world. The startup raised $230 million last year and unveiled an early glimpse of its progress in generating virtual 3D scenes from 2D images.
For the time being, A.I. developers are heavily focusing on agents, with leaders like OpenAI and Microsoft (MSFT) emphasizing autonomous, action-taking features in recent A.I. releases. Nvidia is doing just fine in the agentic A.I. era. The company’s quarterly revenue surged 78 percent year-over-year to $39.3 billion, the vast majority of which came from its data center business. Profit during the quarter jumped 80 percent to $22 billion.
Nvidia’s physical A.I. dreams remain a minuscule part of its business, with its automotive and robotics division accounting for less than 2 percent of the company’s total revenue during the November-January quarter. This share will rise in the future. “Someday, there will be 1 billion cars on the road, and every single one of those cars will be robotic cars,” Huang told analysts, adding that Nvidia will “be improving them using an A.I. factory” powered by physical A.I.
Despite exceeding Wall Street’s expectations on both revenue and net income, some investors were spooked by Nvidia’s underwhelming gross margin outlook. Nvidia’s stock is down by more than 3 percent today.
