China’s humanoid robots are being tested by reality, and reality is winning.
A manager at one robotics firm, whom we will call Wang Hao, told Chinese business outlet Yicai that the dexterous hand his company sources from Inspire Robots (Yinshi) started failing after gripping a five-kilogram object roughly 50 times. Inspire is one of China’s leading dexterous-hand makers. The episode lays bare a problem the industry would rather not discuss: as robots move toward deployment, the reliability of their core components is nowhere near ready.
Chinese firms are pushing humanoids into mass production, but the supply chain underneath them is immature and the critical parts have not been battle-tested. Hardware lifespan is now the bottleneck.
The hands that keep breaking
Wang’s company has already shipped robots to partners. They are being trialled on car assembly lines and in garment-factory loading and unloading, where moving heavy bins by hand is punishing work. “We want robots to replace this kind of heavy, dangerous manual labour,” he said.
2026 is the pivotal year in which Chinese companies are moving humanoids into real industrial settings. Before this, the machines were fine as university research platforms, stage performers, or museum guides. Factory floors are far less forgiving.
Loading bins and handling materials on an assembly line tests a robot’s hands above all. Almost everything humans do, we do with our hands. If humanoids are to matter economically, they need hands that are both dexterous and durable.
Robot makers rarely produce every component themselves. For the hand, Wang’s company runs a dual track: developing its own four- and five-fingered hands while also buying from outside suppliers such as Inspire, a firm now ten years old. In Wang’s view, today’s dexterous hands are simply not good enough. When a robot moves five-kilogram loads in a warehouse, some hands break after about 50 cycles. Given the hand costs under 10,000 yuan, he calls that “forgivable, for a low-cost part.”
In industrial use, a hand needs to survive tens of thousands of cycles to even pass the bar. Factories run fast; a single failure can stall a line. And once a hand is on a robot inside a plant, “you can’t go in and swap the hand every time — that just doesn’t work industrially.”
An Inspire executive countered that the company shipped more than 10,000 hands last year to around 700 customers, and that its load-tested units reach up to a million cycles. He called Wang’s experience an isolated case, “not very representative,” and said the hands are constantly improving — for instance, switching some models from hard-plastic casings, fine for lab use, to all-metal for the rigidity demanded when mounted on a robot.
Others echo the durability concern. “In a warehouse the grip frequency is extremely high — one robot needs to grab eight to ten thousand times a day. The top dexterous hands on the market today last only 300,000 to 500,000 cycles. And that’s a low-degree-of-freedom hand, not a high-DOF one; it might last about a month before breaking,” said Dong Qinpeng, who leads embodied-intelligence robot R&D at warehouse-robotics firm Geek+.
A supply chain still maturing
The core tension is that a dexterous hand cannot yet be both nimble and tough. Complex mechanisms buy flexibility but limit load capacity; rugged hands struggle with delicate tasks. At showcases like the World AI Conference, firms display beautiful, pricey five-fingered hands; for actual simple work, many quietly fall back on two-fingered grippers.
“Two-fingered hands basically have no lifespan problem, but their use is limited — fine for simple grabs in controlled settings, but they can’t cover the millions of SKUs in a warehouse,” Dong said. Geek+’s latest robot uses a three-fingered hand as a compromise: dexterous enough to handle huge SKU variety, yet with enough volume to engineer for reliability, cabling and motor selection. It is, he said, an attempt to fill the last durability gap in real-world deployment.
The hand problem reflects the immaturity of the whole robotics supply chain. AgiBot (Zhiyuan) put its Genie G2 robot into a Longcheer tablet factory this year; it only began screening and qualifying some core parts in early 2025. “Early on it was very hard to find parts that fully met our standards — from performance to capacity to quality systems and delivery,” said AgiBot partner Yao Maoqing. One of the firm’s quality engineers never once sat in the head office after joining; he was sent straight to a core-component supplier and camped there for three months.
The Genie G2 was tested hard once inside the plant. When it arrived last December, the team was confident it would breeze through. Then, under heavy loads and high line speeds, dozens of software, hardware, communication and integration issues surfaced and had to be fixed. Customers want a “plug-and-play” workforce, like buying a car — “put it in and it just works reliably,” Yao said — not something that keeps failing and needs restarts and repairs.
Supply-chain maturity is tightly linked to price. Chinese humanoids typically ship at several hundred thousand yuan, which works against mass adoption. As volume production ramps, it will gradually harden the supply chain — and only a mature chain can drive costs and prices down far enough to put robots into the real world.
Source: Sohu IT. Translated and adapted from Sohu IT (it.sohu.com).