Inside the Chinese startup racing to give every robot a craftsman's hands

Robots can walk. They cannot fold a shirt. The binding constraint on every humanoid program, from Tesla's Optimus to a wave of Chinese factory pilots, is the hand, and the most aggressive bet on closing that gap now sits inside a Beijing startup called Linkerbot.

The company raised a round that values it at roughly $3 billion, according to a Forbes profile of founder and CEO Alex Zhou. Earlier reporting by Reuters and Wired in May pointed to a $6 billion target in a subsequent round; treat the two numbers as separate marks unless reconciled. The round drew Ant Group and HongShan, Tech Funding News reported.

Linkerbot's product is a five-fingered dexterous hand, already deployed in factories across China for grasping and assembly, per Zhou. The hands are line-of-business units, not research prototypes. That distinction separates Linkerbot from the better-funded humanoid labs that ship walking demos.

Dexterity matters more than locomotion because it does the work. Legs move a robot to the task. Hands do the task. At the Los Angeles All-In Summit in September 2025, Elon Musk said Optimus walks well and that human-like hands have proved "far more elusive," a line the Forbes profile anchors. For every humanoid program, the same arithmetic applies: walking is a solved control problem, manipulation is not.

Zhou's strategy, as he laid it out in the Forbes profile, is to combine a proprietary AI skills database with Chinese manufacturing scale. The skills database is the soft layer: a corpus of human demonstrations and task recipes the hand can draw on at runtime, the way a language model draws on its training corpus. Manufacturing scale is the hard layer: the ability to ship hands by the thousand at a price that lets a factory customer treat them as a line item, not a research project.

The bet is that dexterity, like language, is partly a data problem. If a hand can pull on enough examples of "screw in a bolt," "pick up a glass," "fold a t-shirt," the marginal cost of a new task drops, and the customer stops paying for bespoke programming. That is the same logic that took large language models from research project to commodity API, applied to a different modality.

There is a hard caveat the source itself flags. The field is already crowded, and the technology gap is not yet closed. The hands work in factories, which are structured and repeatable. The next goal Zhou names, cooking and cleaning, is unstructured and unforgiving: a kitchen does not reset between attempts, a shirt does not sit still, a child does not. The AI Insider report on the company's ambitions treats the home deployment as the actual test, not the factory pilot.

Zhou sets his own bar: "The core of what we do is letting robotic hands have the skills of a top craftsman," he told Forbes from an office crammed with robotic parts. A craftsman's hand knows what a screw feels like before it bottoms out, knows when a grip is about to slip, and knows the difference between a glass and a cup. None of that is in the spec sheet.

Linkerbot is a useful instance of a larger pattern, not a verdict on it. The hand problem is the same one Optimus is wrestling with, and the same one every humanoid program that wants to ship a useful worker is wrestling with. A $3 billion round, a factory deployment, and a $6 billion target together answer one narrow question: whether scale and data together can buy the dexterity a single research lab cannot. The next twelve months of factory shipments, and the first home pilots, are the test.