Morgan Stanley: Gap Widens Among Leaders in China's Humanoid Robot Supply Chain, Shipments to Multiply in 2026
nashnova research
After visiting ten Chinese humanoid-robot companies on the ground, Morgan Stanley concludes that the gap between leaders and laggards is widening — every integrator targets a multi-fold shipment jump in 2026, and the first industry shakeout may arrive soon.
How many units are integrators targeting for 2026?
One leading Chinese integrator expects a sharp increase from its 5,000+ unit base in 2025. Fourier targets 2,000 (up from 400–500); MagicBot targets over 1,000; Kepler targets 300 (up from 70–80).
This means → the volume gap between the top and bottom already exceeds ten-fold, and it will only widen once mass production begins.
Morgan Stanley notes the industry is moving past video demos in 2025. With more customer feedback in 2026, the first shakeout could come quickly.
Who is buying these robots — and for what?
Early shipments are still driven by R&D and government-backed procurement; large-scale commercialisation has not begun.
One leading integrator expects its 2026 shipments to split roughly into thirds: entertainment / commercial services, industrial / data collection, and R&D. Fourier sees 60% of 2026 revenue still from R&D, education, and data collection.
In plain terms = the industry is on a staircase path — improving one use-case at a time through human-in-the-loop data flywheels, not achieving a single breakthrough in general-purpose manipulation.
Which component suppliers are pulling ahead?
Leaderdrive (绿的谐波): shipments of its harmonic reducers — gear units that let robot joints rotate precisely — are expected to grow exponentially, potentially contributing half of total company revenue in 2026. Monthly capacity will rise from 50,000 units now to roughly 80,000 by mid-year and 120,000 by year-end, building toward a 1 million-unit target for 2027.
Hengli Hydraulic (恒立液压): widely regarded as the primary ball-screw supplier to a North American integrator, holding roughly 70% share. Hengli is building capacity to supply 2,000–3,000 robots per week.
This means → Morgan Stanley maintains its Overweight rating on both, naming them top picks. Both are shifting from single-component vendors to module-level suppliers — covering actuators, motors, and bearings — to capture a larger share of per-robot value.
Why are suppliers building factories overseas?
Morgan Stanley's research shows a North American integrator prefers non-China capacity to hedge geopolitical risk; once mass production begins — projected for July 2026 — this may become a hard requirement.
Hengli will supply ball screws from its Mexico plant. Sling and Shuanglin plan factories in Thailand. Leaderdrive is evaluating both the U.S. and Thailand.
This reflects a forced decentralisation of the supply chain — even when technology and cost advantages sit in China, capacity must be dispersed to lower tariff and policy risk.
Are compute power and tactile sensors settled?
On compute, Fourier runs chips from both NVIDIA and Horizon Robotics. One leading integrator also uses Chinese chips in lower-end models. Both say current compute power is still insufficient.
On tactile sensors — components that let a robot "feel" contact force and shape — approaches diverge sharply: Fulai backs piezoresistive; Tashan emphasises data-consistency advantages of capacitive; Paxini uses a Hall-effect design.
In plain terms = the industry has not converged on a single technical route. Multiple approaches coexist, meaning the competitive window for sensor suppliers remains open — it is too early to call a winner.
市场有风险，内容仅供研究参考，不构成投资建议。