TDK invests JPY100B in physical AI sensors, targets 2029 production

TDK's 100-billion-yen sensor expansion is a positioning play, not a capacity story. The physical AI bottleneck is not compute—it is the sensing-actuation loop. Every joint encoder and current-sense node in a humanoid robot depends on magnetic sensing. Retrofitting Ojiya rather than greenfield-building compresses the ramp to 18-24 months but locks the process window into an existing line architecture, a vulnerability if TMR or spin-torque oscillator topologies iterate faster than planned. Upstream, doubled output pulls hard on neodymium-iron-boron feedstock and precision winding, pressuring magnet suppliers in mainland China and Taiwan, China. Downstream, the precision jump from ±1° industrial grade to ±0.1° robotic grade accelerates the Hall-effect-to-AMR/TMR generational shift, and this capex almost certainly funds that architectural migration. Infineon and Microchip (post-Allegro) dominate automotive-grade magnetic sensing, but humanoid robots carry no AEC-Q100 moat. TDK's design-win window is roughly 12-18 months before Tesla, Figure, or leading Chinese OEMs lock their BOMs. The real risk is demand timing. If humanoid volume slips past 2027, the depreciation burden hits TDK's P&L in FY2026-27, forcing a price war in automotive and industrial segments that compresses industry ASPs across the board.