More than three years after ChatGPT transformed artificial intelligence from a theoretical concept into a daily utility, the robotics sector remains searching for its equivalent breakthrough. Les Karpas, Nvidia’s Global Head of Physical AI for Inception, addressed this disparity during a session at TechCrunch Disrupt 2026 titled “Robots Are Waiting for Their ChatGPT Moment. Here Is What Is Standing in the Way,” held on the Real World AI Stage.
Karpas identified a fundamental obstacle: the absence of a comprehensive, internet-scale dataset for physical AI. While language model developers like OpenAI and Anthropic leveraged vast text corpora, and autonomous vehicle companies such as Waymo accumulated extensive road mileage data over years, robotics lacks a comparable universal training ground. To bridge this gap, a growing network of startups is currently developing synthetic data, simulation environments, and foundation models designed to operate across multiple robot architectures simultaneously.
The challenge of connecting digital models with physical execution is the driving force behind the creation of the Real World AI Stage at the event. Alongside Karpas, founders from Shield AI, Colossal Biosciences, FieldAI, and Foxglove will examine the distinct hurdles and prospects within the industry.
Karpas occupies a unique vantage point to analyze these limitations. In his role at Nvidia, he engages regularly with startups spanning robotics, automotive, manufacturing, mobility, and smart city sectors. His career trajectory reflects the interdisciplinary expertise required in this field, encompassing positions as an architect, manufacturing engineer, startup CEO, and venture executive at organizations including iRobot, Stanley Black & Decker, Herman Miller, and Cirque du Soleil.
TechCrunch Disrupt 2026 takes place from October 13 to 15 at Moscone West in San Francisco, attracting over 10,000 technology leaders, founders, and investors. Early registration offers savings of up to $200 for tickets purchased before September 25, with additional discounts of up to 30% available for group bookings.
Sim-to-real transfer is improving fast. I reckon we’ll see a viable breakthrough within the next two years.
What about safety regulations? Even with perfect data, getting robots into homes will take decades of legal battles.
Is this just hype to justify Nvidia’s new robot chips? I hope they actually solve the physics simulation problem.
Finally, someone admits the data gap is the real bottleneck. Simulation isn’t enough; we need real-world diversity.