Yao Song, 34, has launched his third venture, Striding AI, securing approximately $100 million in angel financing to pioneer what the company defines as Physical Intelligence. Building on his previous exits co-founding Deephi Tech, which was acquired by Xilinx for $300 million in 2018, and launching commercial aerospace firm Oriental Space, Yao now targets the convergence of artificial intelligence and physical environments. The Beijing-headquartered startup positions itself as a Physical Intelligence systems provider rather than a traditional robotics or algorithm firm. Yao distinguishes his focus from the broader Embodied AI trend by emphasizing machines that internalize fundamental physics laws, momentum conservation, and environmental variables rather than merely executing programmed motions. At the core of this approach is the company’s proprietary Latent World Action Model 1.0, a 2.3-billion-parameter architecture that processes environmental data in compressed latent space rather than raw pixels. This design prioritizes real-time decision-making, achieving 98.6 percent task success on the Libero benchmark with inference times of roughly 187 milliseconds, significantly outperforming pixel-level video generation approaches in latency and computational efficiency. The technical stack is further reinforced by RLinf, an open-source reinforcement learning framework co-developed by Tsinghua University assistant professor Yu Chao, which accelerates training throughput for embodied systems. Striding AI’s commercial strategy departs from the industry norm of developing generic models before seeking applications. Instead, the company anchors its initial deployments in high-volume, repetitive environments through strategic backing from major investors. CP Group provides access to global retail and convenience store networks, while Huaqin Technology supplies precision manufacturing and electronics assembly lines. This scenario-first methodology ensures a continuous feed of high-fidelity operational data, which Yao identifies as the critical bottleneck for physical AI development. The company will initially target overseas markets, particularly regions facing acute labor shortages such as Japan, Europe, and North America, where the economic case for automation is most immediate. On the hardware front, Striding AI is pursuing a dual-platform strategy, deploying wheeled robotic arms alongside humanoid configurations. Yao notes that wheeled bases currently offer superior stability, cost efficiency, and sub-millimeter positioning accuracy, making them more viable for near-term commercialization despite humanoids representing the eventual endpoint. He cautions against premature household deployment, citing unresolved safety and standardization hurdles, and expects domestic integration to align with the maturation of next-generation energy storage and regulatory frameworks. Yao frames the company’s trajectory against the evolutionary milestones of large language models, noting that Physical Intelligence has yet to experience its equivalent of a breakthrough demonstration, widespread public recognition, or scalable revenue model. Striding AI aims to catalyze these phases by transitioning robotic systems from experimental showcases to reliable, revenue-generating service providers in commercial and industrial settings. The capital injection will primarily fund model scaling, real-world data acquisition, and international pilot deployments as the company races to define the architecture of the physical AI era.