Why 3 AI Luminaries Are Betting on Physical AI Over Trillions in Chatbots

The AI industry is witnessing a significant pivot among some of its brightest minds, moving away from large language models (LLMs) that power chatbots like ChatGPT and Claude toward 'world models' and physical AI. Louis Castricato, a researcher who spent eight years studying LLMs at Brown University, recently abandoned his doctoral studies to found Overworld, a startup focused on AI that can understand and navigate the physical environment. His decision reflects a growing sentiment that fundamental LLM research has reached a plateau. 'We basically have passed the point of doing real fundamental LLM research,' Castricato said. 'Now it's just applications.' This shift is not isolated. Fei-Fei Li, the renowned 'Godmother of AI,' launched World Labs in San Francisco to develop world models that learn the statistical structure of space and time—how light falls, how objects obey physics, and how scenes appear from new angles. In an essay published in June 2026, she argued that language models alone cannot achieve true intelligence because they lack the capacity to 'read the room.' Yann LeCun, another AI pioneer, left his post as Meta's Chief AI Scientist in 2025 to start Advanced Machine Intelligence Labs (AMIL) in Paris, dedicated to similar goals. These departures mark a critical juncture. While investors are still pouring trillions of dollars into chatbot-focused companies like Anthropic and OpenAI, the frontier of innovation is shifting. World models aim to give AI a genuine understanding of the physical world, enabling robots and autonomous systems to interact safely and effectively. This could unlock enormous new markets in manufacturing, logistics, healthcare, and consumer robotics—areas where language-only AI falls short. The implications are profound. First, the talent drain from academia and Big Tech labs could accelerate as more researchers see greater scientific and commercial opportunities in physical AI. Second, it may fragment the AI ecosystem: the current concentration of resources in a handful of LLM behemoths could give way to a more diverse landscape of specialized world model startups. Third, the technical challenges are immense—incorporating sensory input, real-time interaction, and safety constraints—so early movers with deep expertise, like Li and LeCun, may have an outsized advantage. However, the near-term commercial path is less clear than for chatbots; revenue models for world models are still nascent, and the hardware integration adds complexity. Yet if these pioneers succeed, they could build platforms that underpin the next generation of AI-driven robots, drones, and spatial computing. For the broader market, the pivot signals that the era of pure language AI may be peaking, and the next wave of value creation will come from systems that can see, move, and act. Companies that fail to adapt risk being left behind, while startups with strong research pedigrees could become the new giants.