Harmonic Reducers

How a harmonic reducer works

A harmonic reducer, or strain wave gear, gets a very large speed reduction out of a single, compact stage. The principle was invented by the American engineer C. Walton Musser, who applied for the basic US patent in 1955. As Harmonic Drive Systems describes it, it has three parts: a wave generator, which is an elliptical cam inside a thin ball bearing on the input shaft; a flexspline, a thin steel cup with teeth on the outside that connects to the output; and a circular spline, a rigid ring with internal teeth fixed to the housing.

The circular spline has two more teeth than the flexspline. As the wave generator turns, it flexes the flexspline into mesh at two points, and each full turn shifts the flexspline back by two teeth. That gives single-stage reduction ratios of 30:1 to 320:1 with very little backlash, in a hollow, coaxial package that is small and light for the torque it carries.

Those properties make it a standard part in robot joints. Chinese makers draw the same line between the two main robot reducers: harmonic reducers for lighter-load joints such as wrists and smaller arms, and RV (cycloidal) reducers for heavy-load joints (Haozhi 2026 interim report, p. 6).

Why humanoid robots use them

Most harmonic reducers still go into industrial and collaborative robots, but humanoids add a second source of demand. At Tesla AI Day 2022, Tesla said the Optimus rotary actuator combines a bespoke permanent magnet motor, an integrated mechanical clutch, a crossed roller bearing on the output side, and a strain wave gear as the gear train. The design has been revised since, but the logic holds: a humanoid needs many compact, precise rotary joints, and a harmonic reducer is the most direct way to get high reduction in a small joint.

Where China stands

The incumbent is Japan's Harmonic Drive Systems, formed in 1970 from the merger of Hasegawa Gear Works and the US company USM, and still the owner of the HarmonicDrive trademark.

China's leader is Leaderdrive (绿的谐波, 688017.SH). It says it was the first Chinese company to industrialise harmonic reducers and broke international brands' monopoly in China's robot reducer market, while acknowledging that the international leaders have long dominated the product (2026 interim report, pp. 8, 22). Its volumes are growing fast: it sold 425,158 harmonic reducers in 2025, up 72.5% on 2024 (2025 annual report, corrected, p. 38). In the first half of 2026, revenue rose 38.6% to ¥348.5m, which the company attributes to share gains in industrial robots and a sharp increase in embodied-AI (humanoid and other) robot business; harmonic reducers and metal parts made up ¥288.1m of it (2026 interim report, pp. 6, 145).

Other listed makers are broader motion-control companies for which harmonic reducers are one product among many:

• Zhongda Leader (中大力德, 002896.SZ) sells harmonic, RV and planetary reducers, gear motors and integrated actuator units, and lists AgiBot among its end customers (2026 interim report, p. 9).
• Haozhi (昊志机电, 300503.SZ), best known for machine-tool spindles, makes harmonic reducers in sizes 11 to 58 with ratios of 30 to 160. It says they are used by leading collaborative robot makers and that it has agreements with some leading humanoid makers, without naming them (2026 interim report, p. 15).

Volumes up, margins down

The filings show a market where volume is rising and prices are under pressure:

• Leaderdrive's gross margin fell from 41.1% in 2023 to 37.5% in 2024 and 36.9% in 2025 (2025 annual report, p. 35). Its harmonic reducer and metal parts revenue and cost for the first half of 2026 imply a margin of about 33% (our calculation from p. 145 of the interim report).
• Zhongda Leader's precision reducer margin was 14.6% in the first half of 2026, down 9.0 points on a year earlier, and the company said competition had left it with rising revenue but falling profit: net profit attributable to shareholders fell 40% (2026 interim report, pp. 9, 11).

Our view: the gap between a roughly one-third margin at the leader and the mid-teens elsewhere suggests that scale, yield and customer qualification matter more than the basic design, which is decades old and widely understood.

What to watch

• Margins as humanoid volume arrives. Rising unit sales with falling margins would mean the humanoid opportunity is being competed away on price.
• Smaller units. Humanoid hands and compact joints push towards miniature reducers, which Leaderdrive names as a development priority (2025 annual report, p. 47).
• Alternative joint designs. Some humanoid joints use planetary gearboxes with high-torque motors instead; Haozhi, for example, lists humanoid planetary reducers alongside its harmonic range (2026 interim report, p. 15). Which approach wins in each joint will set the size of this market.