At TechCrunch Disrupt 2026, Nvidia’s Les Karpas will dissect the fundamental data gap preventing general-purpose robots from achieving the same transformative leap as large language models and what it will take to finally bridge the digital-physical divide.
ByMark LimPublished about an hour ago•4 min read
The debut of ChatGPT in November 2022 fundamentally altered the trajectory of artificial intelligence, transforming large language models from academic curiosities into ubiquitous tools that reshaped industries overnight. Text, images, and code could be generated, summarized, and manipulated with unprecedented fluidity, creating a sense that intelligence itself had been democratized. Yet while AI has permeated daily life through screens and servers, robotics remains stubbornly tethered to niche applications and controlled environments. Despite decades of advancement in actuators, sensors, and control theory, the field has yet to experience its own “ChatGPT moment,” a singular breakthrough that propels general-purpose robots from laboratories, warehouses, and factories into the messy, unpredictable fabric of everyday human existence. At TechCrunch Disrupt 2026, Les Karpas, Global Head of Physical AI at Nvidia Inception, will address this critical inflection point in his session on the Real World AI Stage, exploring why robotics is still waiting for its defining catalyst and what systemic barriers must be dismantled to achieve it. His analysis goes beyond technical specifications to examine the economic, infrastructural, and philosophical challenges that have kept embodied intelligence perpetually “five years away” from mainstream adoption.
Karpas identifies the core bottleneck with stark clarity: unlike language models, which were trained on vast, internet-scale text corpora accumulated over decades, physical AI lacks an equivalent foundational dataset. Language exists as a universal, digitized medium; physical interaction does not. While autonomous vehicle companies like Waymo have painstakingly built proprietary datasets through millions of real-world miles, these efforts remain siloed, geographically limited, and insufficiently diverse to train truly general-purpose embodied intelligence. A robot trained to navigate a warehouse in Arizona may fail catastrophically when placed in a cluttered Tokyo apartment or a rainy Seattle sidewalk. The result is a fragmented landscape where robots excel in narrow domains but lack the adaptive reasoning required for true versatility. Startups are now attempting to manufacture scale artificially through high-fidelity simulation, synthetic data generation, and cross-embodiment foundation models that attempt to distill universal principles of physics and manipulation. However, bridging the sim-to-real gap remains one of the most persistent challenges in the field; virtual environments cannot perfectly replicate the friction, wear, sensor noise, and chaotic unpredictability of the physical world. Without a shared, comprehensive dataset that captures the infinite variability of physical interaction across cultures, architectures, and tasks, robotics will continue to lag behind the exponential progress seen in purely digital AI.
Nvidia’s perspective on this challenge is uniquely informed by its position at the nexus of hardware, software, and ecosystem development. As head of Physical AI within Nvidia Inception, Karpas interfaces directly with the startups and enterprises striving to solve this data deficit across robotics, automotive, manufacturing, mobility, and smart cities. He sees firsthand how computational power alone cannot compensate for the absence of representative data; even the most advanced GPUs are limited by the quality and diversity of their training inputs. His background reflects the interdisciplinary nature of the field itself: spanning roles as an architect, manufacturing engineer, startup CEO, venture studio executive, and corporate investor at organizations ranging from iRobot to Cirque du Soleil, he embodies the convergence of disciplines required to advance embodied intelligence. Architecture teaches spatial reasoning and human-centric design; manufacturing instills precision and scalability; entertainment demands adaptability and audience awareness. At Disrupt, he will be joined by founders from Shield AI, Colossal Biosciences, FieldAI, and Foxglove, each navigating distinct facets of the physical AI frontier. Their collective insights will illuminate not only the technical hurdles but also the business model innovations, regulatory frameworks, workforce implications, and infrastructure investments necessary to catalyze widespread adoption. They will discuss how to build datasets that respect privacy and safety, how to create evaluation benchmarks that matter beyond lab metrics, and how to align robotic capabilities with genuine human needs rather than technological novelty.
For attendees, Karpas’ session represents more than an analysis of current limitations; it is a roadmap for future opportunity and a call to action for the entire ecosystem. Understanding the precise contours of the data gap can inform investment strategies, guide startup pivots, and reveal underserved niches where new solutions might gain traction. Investors can learn to distinguish between ventures that are genuinely solving foundational problems and those merely repackaging existing automation with AI buzzwords. Founders can identify collaboration opportunities to pool data resources and accelerate collective progress. Policymakers and ethicists can engage with the practical realities of deploying intelligent machines in public spaces. The Real World AI Stage is designed precisely for this kind of actionable insight, connecting those building in the physical world with the capital, partnerships, and knowledge needed to accelerate progress. With ticket prices increasing after September 25 and group discounts available, securing access now ensures participation in conversations that could define the next decade of robotics. Just as ChatGPT emerged from years of incremental research before suddenly capturing global attention, robotics’ breakthrough may be closer than it appears—but only if the right minds are in the room to recognize and seize it. The moment will not arrive through hype or speculation; it will emerge from the unglamorous, collaborative work of building the datasets, standards, and trust that make physical AI not just possible, but indispensable.
About the Creator
Mark Lim
Hi I am mark an automotive student and a car, tech and food enthusiast ! Im gonna try and post daily & hope you enjoy what I write and do share my page with people you know. I would gladly appreciate it! Cheers
