On December 17, He Xiaopeng, Chairman of XPeng Inc., shared his thoughts on robotics and the AI bubble in a social media post. He stated that humanoid robots will become a battleground for tech giants, while specialized robots will see fierce competition across various fields, presenting abundant opportunities for success. He believes there is currently no AI bubble and that the future AI market holds immense potential.

Key takeaways from his post:

**On U.S. Startups and Robotics** During his recent U.S. trip, He engaged with nearly 30 professionals in the AI sector. He observed that U.S. startups are thriving in AI, biotech, and finance. In Silicon Valley, AI ventures—particularly in SaaS and physical AI robotics—are highly concentrated, with soaring valuations. Half of the entrepreneurs he spoke with were focused on robotics. He noted that Chinese robotics firms often start with joint and control technologies, while U.S. companies prioritize model development. He predicts humanoid robots will be dominated by industry giants, whereas specialized robots will attract diverse players with ample room for success.

**On Physical AI** Quoting Wittgenstein, "The limits of my language mean the limits of my world," He emphasized that language, while essential for knowledge and communication, can also constrain imagination. Many skills—like walking or swimming—are learned through observation, imitation, and practice, not just language. Thus, the AI field is shifting from language models to multimodal or even world models. Over the next three years, He expects breakthroughs not in the digital realm (where OpenAI focuses on AGI commercialization) but in physical AI. For instance, autonomous driving could leap to near-L4 or full L4, and humanoid robots might rapidly advance from L1 to early L4 capabilities. Though progress in physical AI may be slower than digital AI, its impact on daily life will be profound.

**On the AI Bubble** He acknowledged that every tech era sees localized bubbles, from the internet to new energy vehicles, as markets evolve from chaos to order. Overall, he believes AI will drive transformative societal changes and remains in its infancy (0 to 0.1 phase). While U.S. valuations seem inflated, China’s are more reasonable, with the former prioritizing cutting-edge research and the latter emphasizing practical applications. He concluded there is no current AI bubble, only vast future opportunities.

**On AGI’s Arrival** Today’s AI, He noted, relies on imitation learning (e.g., absorbing textbook knowledge) and reinforcement learning (e.g., iterative problem-solving). Like autonomous driving—which improves rapidly by learning from millions of drivers—AI exhibits emergent capabilities but lacks true creativity. Achieving AGI requires advancements in multimodal-to-world models (akin to understanding causality), continuous learning, and long-term planning, which may take years and depend on foundational breakthroughs.