NextFin News — The hardest part of a humanoid remains the hand. A human palm and fingers combine bone, joint and soft tissue into grips that range from a firm clamp to a light pinch. Metal and polymer replicas still lag. That gap helps explain why viral routines and timed sprints have not yet produced large numbers of machines doing paid labor on Chinese factory floors.

Industry estimates put global humanoid installations last year in the low tens of thousands. The share performing sustained productive work is widely believed to be much smaller. Without reliable hands, robots are confined to pushing carts, moving large rigid parts or repeating scripted motions. With better hands, the task set expands toward tools, fasteners, cables and the messy variety of assembly and logistics.

Chinese suppliers have moved early into that niche. LinkerBot, a Beijing-based specialist in dexterous manipulators, has become a frequent reference among integrators. Independent research cited in trade coverage attributes to it a large share of high-end robotic hands already deployed, and the company has discussed raising monthly output as orders rise. Customers span consumer electronics, technology platforms and industrial automation.

Hardware strength extends across the full body. Chinese firms accounted for the overwhelming majority of global humanoid unit shipments in the first half of this year, according to industry tallies. The advantage rests on dense clusters of mechanical engineers, component makers shaped by the electric-vehicle boom, and a policy environment that has treated intelligent robots as a priority manufacturing category. Motors, reducers, sensors and structural parts that once relied on imports are increasingly sourced at home. Bill-of-materials costs for domestically assembled humanoids have continued to fall, reinforcing volume ambitions.

Capital markets have amplified both the boom and the caution. High-profile listings drew intense first-day demand and later valuation compression as investors questioned how quickly demos convert into recurring revenue. Planning authorities have also warned against indiscriminate entry into the category. The same enthusiasm, however, has funded rapid iteration inside plants. Automakers, electronics manufacturers and logistics operators are putting early machines on lines not only to cut labor in narrow tasks but to generate the interaction data needed for better control software.

Software remains the harder climb. Embodied AI—the models that map perception to action in unstructured settings—still trails the maturity of the mechanical stack. Newer Chinese firms are racing to close that gap by collecting factory telemetry, training world models and selling integrated packages of hardware plus control software. Training philosophy itself is becoming a competitive axis. Some groups emphasize industrial data from assembly and warehousing. Others capture high-skill human demonstrations—precise knife work, instrument playing, fine tool use—on the theory that rare dexterity transfers better than average factory motion. Neither approach has yet delivered a general-purpose system that replaces broad categories of human labor.

Cost trajectories keep the pressure on. Bank research has projected that component costs for a typical Chinese-made humanoid could fall by roughly half toward the end of the decade if scale and design learning continue. Long-range forecasts of very large markets assume aging workforces and steady gains in autonomy. Those forecasts remain sensitive to one stubborn variable: whether manipulators can handle the small, irregular and force-sensitive objects that define most paid physical work.

For China’s humanoid sector, the near-term contest is therefore practical rather than theatrical. Production capacity for bodies and hands is expanding. Factories are becoming the primary classroom. The companies that turn dexterous hardware and improving software into machines employers keep after the demonstration ends—because they can pick up a screw, a connector or a soft pouch and place it correctly—will set the pace of commercial adoption. The rest will remain impressive on stage and idle on the line.