Where the money sits in the robot

Start at the top. Morgan Stanley’s teardown-style estimate for a Tesla Optimus Gen 2-class machine puts the full bill of materials at roughly $55,000, with actuators taking about 56 percent of it. The geography of that spend follows the load path: the locomotion hardware — thighs, calves and feet — accounts for around $21,000 on its own, the hands about $9,500, and the waist and pelvis another $7,800. The battery pack, for perspective, is a few hundred dollars. The joint, not the AI and not the battery, is the cost problem.

Chinese supply-chain analysis slices the same pie by component instead of by limb, and lands in the same place. In X.PIN’s 2026 breakdown of the Chinese humanoid BOM, precision reducers take about 21 percent of component cost, motors 19 percent, lead screws 19 percent, sensors 14 percent, controllers 12 percent and structural parts 10 percent. Add the first three together and roughly six yuan in every ten spent on parts goes into making joints move.

Anatomy of a joint’s price tag

Open up one rotary joint module and the price tag splits into five or six parts. The reducer is usually the single most expensive item: a high-end strain-wave (harmonic) reducer runs $800 to $1,500 per unit at the quality grade humanoid wrists and arms demand, and Chinese equivalents are estimated 30 to 50 percent cheaper than the Japanese and Western incumbents. A planetary gearset does the same job at roughly one fifth of the harmonic’s cost — which is exactly why quasi-direct-drive robots standardised on planetary stages everywhere precision allows it.

The motor tells the same East-versus-West story in miniature. TechBuzz China’s component tracker quotes coreless cup motors from Chinese maker Aoyi Technology at around ¥100 (≈$14) per unit, against ¥4,000+ (≈$560) for imported German equivalents from Maxon or Faulhaber, with Moons’ Industries in between at ¥1,200–2,300. Frameless torque motors — the ring-shaped motors inside every rotary joint — are the second-biggest component category in the robot at roughly 16 percent of total component value; motors of all kinds together are close to a quarter of the BOM. Then come the sensors: a six-axis force-torque sensor still costs ¥5,000 to ¥30,000 (≈$700–4,200), which is why most builders reserve them for wrists and ankles and let motor current do the sensing elsewhere.

The collapse in progress

The reason this page will be out of date within a year is that joint costs are falling faster than almost any other robot subsystem. The cleanest public evidence is Unitree’s own average selling price: about ¥593,000 per humanoid in 2023, down to ¥168,000 across January–September 2025 — a 72 percent drop in two years, driven by vertical integration of its actuators. At retail, the same collapse shows up as a price ladder: UBTech’s Walker S at roughly $74,000, the Chinese fleet average around $52,000, Zhiyuan’s Expedition A2 near $30,000, Unitree’s G1 from about $16,000, the R1 at $5,900, and Berkeley Humanoid Lite — a research platform with 3D-printed cycloidal actuators — under $5,000 in parts. Tesla, for its part, is targeting 50,000 Optimus units in 2026 at a $20,000–30,000 price point, which is only arithmetic if the actuator BOM keeps shrinking.

Where does it end? X.PIN’s supply-chain sources put the 2030 target for a capable Chinese humanoid at ¥100,000–150,000 (about $14,000–22,000) — meaning the whole robot converges on what a single arm’s worth of joints cost in 2023. One widely cited caveat cuts the other way: Chinese content is doing much of the work. One analysis reviewed by GrabaRobot estimates that excluding Chinese parts from an Optimus-class build would roughly triple its cost, which is why US tariff and supply-chain policy has become an actuator-pricing story too.

Who supplies the joint — and who is scaling

Capacity is the other half of the price story, because reducer prices fall when reducer factories multiply. Japan’s Harmonic Drive Systems remains the volume leader, moving from about 1.98 million units of annual strain-wave capacity in 2022 toward roughly 2.14 million in 2025. The Chinese challengers are growing far faster from a smaller base: Leaderdrive from 330,000 to about 790,000 units per year over the same period, and Laifu from 200,000 to about 500,000. Around them, the industrialisation signals keep stacking up: Schaeffler says its formed — rather than machined — strain-wave gearboxes for humanoid joints have completed validation, with scaled manufacturing scheduled for 2027, and China’s Eyou has opened what it calls the first fully automated production line for humanoid robot joints.

FAQ

Sources