Light Origins

Light Origins is an artificial intelligence firm established in late 2024, focused on developing foundation models for embodied intelligence. The company's objective is to expand the capabilities of large models from the digital sphere into the physical world, enabling AI to interact with and operate within real-world environments safely and reliably. The company was founded by CEO Roger Jiang (Jiang Xu), a former algorithm expert at OpenAI. Jiang was a core contributor to the development of InstructGPT, ChatGPT, and GPT-4, and is credited multiple times in the GPT-4 technical report. His background includes a Ph.D. in Physics from the University of Maryland and extensive work in large-scale model training and foundational reinforcement learning from human feedback (RLHF) research. The company pursues an "intelligence-first" approach, with its research and development spanning foundation models, multimodal perception, real-world feedback, robotic systems, and continual learning. Light Origins is creating an integrated system where models, compute, data, hardware, and deployment all contribute to a single learning loop to power the next generation of Physical AI. The firm's primary product focus is the development of general-purpose robots, with a bipedal humanoid robot announced as forthcoming. The business operates in the advanced AI and robotics market, targeting the development of underlying intelligence for physical AI systems. In August 2026, Light Origins announced it had secured several hundred million yuan in a Pre-A funding round. The round was led by CASIC Investment (Guoke Investment), with participation from China Merchants Venture Capital and Xianghe Capital. These funds are allocated towards ultra-large-scale embodied AI model training, the buildout of its multimodal data infrastructure, and team expansion. Headquartered in Shenzhen, China, the company also has locations in Beijing and Singapore and has grown to over 100 employees, with a significant majority dedicated to research and development. Keywords: Physical AI, embodied intelligence, foundation models, general-purpose robots, bipedal humanoid, Roger Jiang, Jiang Xu, OpenAI, ChatGPT, GPT-4, artificial intelligence, multimodal perception, continual learning, robotic systems, real-world feedback, CASIC Investment, China Merchants Venture Capital, Xianghe Capital