Tesla's Optimus Gen 3 is heading to full-scale production while Chinese makers race ahead. Is this the "EV moment" for humanoids? Drag the sliders to ramp output and watch Wright's law grind the unit cost down as the fleet grows.
Studying aircraft factories, Theodore Wright found that every time cumulative units produced doubles, unit cost falls by a roughly constant percentage — the learning rate.
cost(N) = C₀ × (N / N₀)log₂(1 − LR)At a 20% learning rate, 10 doublings (1× to 1,024× volume) cut cost by ~89%.
Lithium-ion pack prices fell about 89% from 2010 to 2020 (roughly $1,200 to $132 per kWh, per BloombergNEF) as cumulative production doubled again and again. Global EV sales went from tens of thousands per year in 2010 to over 10 million by 2022. Cost decline unlocked demand, which funded more capacity — a flywheel.
Humanoids share the EV supply chain: motors, battery cells, power electronics, cameras and compute. The bill of materials is dominated by actuators and precision reducers — exactly the components China's motor and harmonic-drive ecosystem is scaling. Shared parts mean humanoids start partway down an existing learning curve.
Tesla targets high-volume Optimus Gen 3 lines; Unitree, UBTech and Agibot are already shipping units and cutting prices aggressively. Analysts have repeatedly revised humanoid market forecasts upward (Goldman Sachs to ~$38B by 2035). This page is an illustrative model, not a forecast — real learning rates vary by component and are earned, not guaranteed.
Cost is not the only gate. Dexterous hands (20+ degrees of freedom with tactile sensing) remain the hardest subsystem; mean-time-between-failure must reach thousands of hours for factory duty; and safety certification for robots working beside humans is still being written. Learning curves apply to each of these separately — the slowest curve sets the pace.
Wright's law assumes demand keeps absorbing output. If early humanoids disappoint on reliability or ROI, volumes stall and the doublings stop. Rare-earth magnet supply, precision reducer capacity, and export controls could also flatten the curve — as could a shift to task-specific (non-humanoid) robots that soak up the same demand.
Start at 50,000 units in 2026 with 80% annual growth. Production compounds; cumulative volume is the running sum.
P(t) = 50,000 × 1.80ⁱBy 2032 cumulative volume is roughly 3.7M units — about 6.2 doublings from the 50k base (2⁶․² ≈ 74×).
doublings = log₂(Cum / Cum₀)At a 20% learning rate each doubling multiplies cost by 0.80. Six doublings: 0.80⁶ ≈ 0.26, so a $65k robot approaches ~$17k.
cost = 65,000 × 0.80^doublings| Year | Annual units | Cumulative fleet | Doublings | Unit cost | vs 2026 |
|---|