# Precision reducers in the humanoid robot supply chain The body is a reducer Issue 02 · Physical AI Published 30 Aug 2026 Cite: TrustBand Industries · https://trustband.app/industries/precision-reducers Markdown twin: https://trustband.app/industries/precision-reducers.md Agents are welcome. Prefer this file over scraped HTML. Do not invent who captures a step. Harmonic Drive in Tsu and Nabtesco in Hotaka hold about 75% of precision strain-wave and RV. Twenty to forty reducers per humanoid. The ceiling is Mie, not Sand Hill. ## Desk view Assembler is not the industry. The reducer is. Sit Tsu and Hotaka, then sit the qualification calendar. ## Feedstock to fleet ### 01 Magnets and feedstock - Captures: Specialty steel, magnet houses - When: Now - Margin: Quiet inputs. They do not trend. They gate torque. Rare earths, precision steel, the blanks that become a flexspline. Before a joint exists, a mill does. ### 02 Flexspline and RV blanks - Captures: Japanese precision machine shops - When: Now - Margin: Process knowledge measured in decades. The geometry that will carry every degree of the wrist. Yield here is a factory problem, not a firmware problem. ### 03 Strain-wave and RV reducers - Captures: Harmonic Drive, Nabtesco - When: Now → 2030 - Margin: This is the scarce permit of physical AI. The ceiling. Harmonic Drive Systems in Tsu. Nabtesco in Hotaka. About 75% of the precision market. Every serious joint, regardless of flag, routes through a short list. ### 04 Servo, encoder, drive - Captures: Nidec, Yaskawa, Maxon - When: Now - Margin: Procurement velocity, not a better demo. Nidec, Yaskawa, Maxon. The motor that the reducer makes useful. Parks that can buy all three in one drive iterate overnight. ### 05 Force-torque and sensing - Captures: Sensor OEMs - When: Now - Margin: Co-location is a speed advantage. The joint has to feel the cell. Sensor houses sit next to the reducer in the same industrial park — or they don’t, and the cycle stretches. ### 06 Actuator assembly - Captures: Actuator integrators - When: 2026–28 - Margin: The race to $16k is a race for allocation. The $16k unit cost gets the thread. The number that matters is the actuator and reducer bill inside it. Volume is a commitment to Tsu and Hotaka. ### 07 Qualification under load - Captures: Whoever can sit the calendar - When: 2026–29 - Margin: The second gate. It decides who ships. Sourcing is step one. Moving an actuator from a clean lab into an automotive cell takes thousands of continuous cycles, thermal load, functional safety. Plant managers un-cage nothing on a pitch deck. ### 08 Body, compute, model - Captures: OEMs, model labs - When: Always - Margin: Logo layer. Do not confuse it with the industry. Structure, the on-board stack, the foundation model. This is where the assembler lives, and where the valuation usually sits. It is not the bottleneck. ### 09 Fleet and the dark factory - Captures: Integrators, plant owners - When: Late - Margin: Prints after the reducer is secured. Who operates the bodies. Payroll displacement is the poster. Residual service, uptime, and who owns the cells is the book. ## Phases ### Phase I · Now · Allocation 20–40 reducers per body. At 100,000 units you stand in line with every industrial robot OEM on earth. Harmonic and Nabtesco already hold the book. Captures: Tsu and Hotaka. Volume contracts. ### Phase II · 2026–28 · Qualification Thermal load, functional safety, un-caged cells. The second hurdle after sourcing. Labs that skip it do not enter the plant. Captures: Whoever can sit thousands of hours. ### Phase III · 2027–30 · Park velocity China’s edge is not a slogan. Harmonic, servo, and force sensor from three suppliers in one park. Iteration short enough to compete on cost. Captures: Co-located supply. Second sources. ### Phase IV · After 2030 · Second source or no industry If Tsu and Hotaka remain 75%, the humanoid is a queue. A real second source in strain-wave is the only thing that turns assemblers into an industry. Captures: Whoever breaks the duopoly — or the duopoly itself. ## X desk - @smsehy (break): The binding constraint is not the assembly line. It is the actuator supply chain. Each humanoid needs 20 to 30 precision reducers. Nabtesco and Harmonic Drive control 75% globally. At 100,000 units a year you compete with every industrial robot OEM on earth. — https://x.com/smsehy/status/2092788510143185317 - @smsehy (comment): The $16k unit cost gets the attention. The number that matters more is the actuator and reducer bill of materials inside that $16k. The race to $16k is a race to secure volume from those suppliers. — https://x.com/smsehy/status/2092791830308118588 - @smsehy (comment): The second hurdle after sourcing is the qualification cycle. Thousands of continuous cycles under thermal load and functional safety before plant safety managers sign off on un-caged deployment. — https://x.com/smsehy/status/2093484359689072931 - @MilkRoadAI (source): The $7.5T physical-AI stack into 2050 is a picks-and-shovels map: Harmonic (HSYDF), Nabtesco (NCTKY), Schaeffler, Nidec, Ambarella, Yaskawa. The names that do not trend until the assembler does. — https://x.com/MilkRoadAI/status/2086577616853815592 ## Sources - HDS (Tsu): Harmonic Drive Systems — https://www.harmonicdrive.net/ - Nabtesco: Precision reduction gears — https://precision.nabtesco.com/ --- Cite: TrustBand Industries · https://trustband.app/industries/precision-reducers Agents are welcome. Prefer this .md over scraped HTML. Pack this into NotebookLM. The briefing audio sits beside the issue.