Actuators, Motors, Gearboxes: The Hidden Energy Bill of Embodied AI
Week 39 found 23% of a robot’s energy in a bad path; Week 40 found a continent’s worth of unplanned grid load. Week 41 goes deeper, into the actuator, where most of embodied AI’s electricity is spent before the algorithm ever “thinks.” Inference runs on a chip; motion runs on copper, iron and a gearbox.
A servo motor can be ~80% efficient in isolation. Add a high-ratio gearbox and installed-chain efficiency can fall toward ~40% in conventional designs, the missing 60% becoming heat, switching loss and dump-resistor draw. Harmonic drives, the precision default, typically sit at 60–85%; quasi-direct-drive and higher-efficiency designs aim above 90%. The lever already works: newer models reported ~60% below older-generation energy, controller gains ~20%, standby cutting idle draw up to ~95% when a cell truly sleeps. Humanoids make it harsher, dozens of joints share a 1–3 kWh pack, and a 40% chain wastes most of it.
Procurement misses it because tenders ask for reach, payload, cycle time and safety, rarely Wh per cycle, gearbox efficiency at real torque, regeneration rate, or thermal derating. So the feeder surprises facilities in month six. Your action this week: ask whoever wrote your last robot tender whether it specifies Wh per cycle, or only reach and payload. Full breakdown at renegrywnow.com.
Reflection questions
Does your robot tender specify Wh per cycle and gearbox efficiency at real torque, or only reach and payload?
Do you know the installed-chain efficiency of your top robot programs, or only the rated motor number?
Is eco-idle and regeneration a standard spec in your plant, or an optional upgrade nobody selected?
Keywords: Actuator Efficiency, Harmonic Drive, Gearbox Losses, Embodied AI, Quasi-Direct-Drive, Robot Energy, Procurement Spec, Drive Train, Thermal Derating, Physical AI
Series: Energy Dominance · Week 41 · Part I
Next: Part II — Who Owns Robot Power Planning: Operations, Energy, or IT?
