Lingxin, a dexterous-hand maker just three years old, has seen its valuation climb to around 20 billion yuan at its B-plus round and near 40 billion at the next, dragging the ceiling of the entire dexterous-hand field upward. It is one of the fastest-rising names in the embodied-AI gold rush, and its bet is unusual: do not hunt for the gold, sell shovels to every miner.
The bottleneck everyone argues about
The hand is the most visible choke point of embodied intelligence. As early as 2018 OpenAI admitted that contact, friction and damping are hard to simulate accurately, and Musk has said the forearm and hand are harder to build than the rest of the robot combined. Every gain in degrees of freedom forces motors, reducers and transmissions to upgrade together, amplifying heat, failure rate and weight. That is why the field still argues over whether a robot even needs a five-fingered hand at all.
Sell the shovel, own the data
Lingxin laid three wide nets. First, it builds all three technical routes, direct-drive, tendon and linkage, refusing to bet on a single ‘mine’. Second, its product matrix spans from a subsidised 3,999 yuan unit to a 42-degree-of-freedom research flagship, covering teleoperation and a skill library. Third, it integrates vertically to push cost down. The founder, Zhou Yong, ranks companies like trees: a few hundred million is a seed, billions a sapling, hundreds of billions a mature tree, trillions a rainforest.
Lingxin has two routes to the top. One is a CATL-style super-supplier so cheap that every builder buys rather than make, though customers from Apple to Unitree tend to insource the core eventually. The other is an Nvidia-style ecosystem lock, and Lingxin’s LinkerSkillNet already holds more than 500 standardised, reusable hand skills. The true moat may not be the hardware but the data: the more hands it sells, the better the model, the stronger the product. The flywheel only turns if real-robot data stays central to training.
Read the original report (OFweek Robotics)
Translated and adapted from OFweek Robotics (robot.ofweek.com).