The Physical AI Conundrum
The tech world remains defined by the release of ChatGPT in late 2022, a catalyst that transformed large language models from research curiosities into indispensable daily tools. Yet, while software-based AI has seen a rapid, explosive trajectory, the field of robotics has experienced a more measured evolution. Despite decades of development, the industry is still waiting for a singular, groundbreaking breakthrough—a 'ChatGPT moment'—that can seamlessly bridge the gap between digital intelligence and physical execution.
Les Karpas, Nvidia Inception’s Global Head of Physical AI, is preparing to tackle this challenge head-on during the upcoming TechCrunch Disrupt 2026. As a central figure in coordinating Nvidia’s massive ecosystem of robotics and mobility startups, Karpas is uniquely positioned to identify why general-purpose robots have struggled to achieve the widespread, intuitive utility seen in virtual chatbots.
The Data Bottleneck
The primary obstacle identified by industry leaders is a significant lack of accessible, large-scale training data. In the software domain, companies like OpenAI and Anthropic were able to leverage vast swathes of internet data to train their foundation models. Robotics lacks a comparable repository. Physical movement, spatial reasoning, and real-world interaction require distinct datasets that cannot be scraped from websites or wikis.
While autonomous vehicle companies like Waymo have successfully compiled extensive data through years of on-road operation, most robotics firms lack that luxury. The industry is currently exploring ways to bypass this by generating synthetic data and utilizing sophisticated simulation environments. This push to bridge the gap between digital simulation and physical reality is the defining mission of modern robotics, and it will be a focal point of discussions at the Real World AI Stage at Disrupt.
Why it Matters
- Data Scarcity: Unlike LLMs, robotics requires complex physical-world datasets that do not exist in a readable format on the web.
- Bridging the Gap: Startups are increasingly turning to simulation-based training and synthetic data to accelerate the development of physical AI.
- Ecosystem Collaboration: The shift toward universal foundation models trained across diverse robot forms could prove to be the turning point for the industry.
Expertise Meets Industry Potential
Karpas brings a cross-disciplinary pedigree to this conversation, having worked as a manufacturing engineer, an architect, and a venture capitalist across firms ranging from iRobot to Stanley Black & Decker. This background allows him to see the robotics challenge through multiple lenses: hardware limitations, software scalability, and the pragmatic requirements of manufacturing and mobility.
At TechCrunch Disrupt, held from October 13-15 in San Francisco, attendees will gain access to these insights alongside leaders from innovative companies like Shield AI and Colossal Biosciences. The event aims to foster a bridge between theoretical research and the hard engineering required to make autonomous, general-purpose robots a standard part of our everyday infrastructure.
