Nvidia's Les Karpas: robots lack a data moment

The comparison the article draws is between language AI and physical AI. ChatGPT's November 2022 debut showed what became possible once large language models were connected to a broad, internet-scale body of text. Robotics, by contrast, has been discussed, built upon, and in some ways adopted for a long time, but it has not had that kind of breakthrough.

The article points to a specific structural reason rather than a lack of effort. Self-driving fleets from Waymo and its competitors accumulated datasets over years of road miles, and those datasets are still expanding market by market. That route is not available to general-purpose robots, so a growing ecosystem of startups is attempting to manufacture scale artificially through simulation, synthetic data, and foundation models trained across many robot forms at once.

Karpas's position at Nvidia Inception means he regularly interfaces with the same companies trying to solve that data gap, and his background spans architect, manufacturing engineer, startup CEO, venture studio executive, and corporate venture capitalist roles. His session sits alongside founders from Shield AI, Colossal Biosciences, FieldAI, and Foxglove, suggesting the conversation will be less about a single fix and more about whether artificial data can substitute for real-world scale.