Future of Tech: Physical AI — Bridging Industrial and Humanoid Robotics
Summaries are public. PDF access requires an active membership; all members have the same access. Sign in
Bernstein examines how physical AI drives a renaissance in industrial robotics and the maturation of humanoid robots, proposing a unified framework encompassing body, brain, skill, and world models. Industrial robots require high precision and speed, making hardware and motion control the primary differentiators. Conversely, humanoid robots demand task generalization and complex physical interactions, making brain and world models critical.
Total humanoid robot shipments in 2025 are estimated at only 20,000 units, with industrial applications comprising just 4%. However, as task generalization improves, industrial applications are expected to reach a 50-60% peak when the market scales to 1-10 million units annually, before eventually normalizing to around 10%. Overlap between industrial and humanoid applications in industrial verticals remains limited.
Success in this space follows two paths: owning the technology frontier and/or controlling an ecosystem. FANUC is expanding its lead over traditional peers like Yaskawa and ABB by leveraging a broader robot portfolio, superior multi-axis motion control, and extensive process packages. Mech-Mind leads the robotic path-planning layer with roughly 30% global market share. In the humanoid segment, Unitree acts as the de facto product platform with about 25% global share, achieving rapid revenue growth. Meanwhile, Figure AI and LimX pursue end-to-end technological breakthroughs, and Nvidia dominates the world foundation-model layer.
Bernstein rates FANUC (6954.JP), Inovance (300124.CH), Keyence (6861.JP), Cognex (CGNX), and Harmonic Drive (6324.JP) as Outperform, while rating Leader Harmonious Drive (688017.CH) as Underperform.