--- title: "A Hand, Sold to the Lab First" date: 2026-11-05 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-hand-volume-13-1105b language: en --- # A Hand, Sold to the Lab First > Robots can run, jump and do flips on a show stand, yet they often stall the moment they must actually pick something up or drive a screw. The trouble usually lies not in the legs or the voice but in Robots can run, jump and do flips on a show stand, yet they often stall the moment they must actually pick something up or drive a screw. The trouble usually lies not in the legs or the voice but in that short stretch below the wrist, which the industry calls the last ten centimeters — the dexterous hand, an end-of-arm part. This article is not about how clever the hand is, nor does it guess who will come out ahead; it looks at only two plain things: how many such hands are sold in a year, and who buys them. The numbers look lively, but the destination is often overlooked, and it is the destination that decides how far the hand still is from a real factory. ## Nearly 20,000 Units in One Year According to statistics from Gaogong Industry Research Institute, China's dexterous-hand market sold about 19,200 units in 2025, up 236.84% year on year. That is well over double the previous year, yet set against manufacturing as a whole it remains a very small number. The growth looks steep only because the base is small. 19,200 units is not 190,000, still less 1.9 million; it describes a market just getting started, where buyers mostly order a few or a few dozen units to try out, rather than placing an order for thousands at once. The next two figures need even more care. The same institute expects 2026 sales to reach 70,200 units, up another 265.63%, and projects that sales could exceed 430,000 units by 2030. Note the wording: one is a forecast, the other a projection; both are estimates, not goods already sitting in a warehouse. Mixing what has happened with what is projected is the easiest way to misread this kind of reporting. The 19,200 units is what has actually been sold; the two later figures are curves the institute drew on the assumption that scenarios widen and costs fall. A curve can be steep, but it sits on paper, not in today's ledger. ## The Hands Reached the Labs First Selling a lot and working on a factory line are two different things. Where these hands go once unpacked is the better question. According to media reports, an executive at a company that entered the field from auto parts says the units actually moving today are still mostly simpler designs, such as three-finger and six-active-degree-of-freedom models. The same executive added that demand comes mainly from research, data collection and model training, and that the share reaching real production stations remains very low. Buyers mostly do not expect it to cover a shift right away; they use it to validate, gather data and feed models. It is a research tool first. Another company puts the shift more vividly. An executive at Lingxin Qiaoshou told China Business Journal that the dexterous hand has gone from "buy one for the lab to try" to "ordered on a production line to a set rhythm," and its use has spread from universities and research institutes to industrial clients and scenario clients. That is one company's own observation, not an industry-wide ledger. Put the remarks together and the picture clears: the buyer list is lengthening while the number of seats on production stations has barely grown. The two do not conflict; they simply run at different speeds. According to media reports, a researcher at a think tank also cautioned that many products perform fine on a demo stand but fail after thousands of continuous cycles on a line, and that factory customers first weigh the economics. If the sums do not work, they will not replace existing equipment at scale. ## Two Companies' Delivery Ledgers Many media outlets have called 2026 the first year of mass production for dexterous hands, largely because leading firms have pushed deliveries to the 10,000-unit threshold. Delivery and sales are not the same either: the former is a company's own reported result, the latter an institute's estimate of the whole market. According to media reports, Yinshi Robotics delivered more than 10,000 dexterous hands in 2025, with a global share above 60%, and set a 2026 target of 30,000 to 50,000 units. The official website of the Changshu municipal government, describing the company's new base, gives the same 10,000-unit delivery figure and the same next-year target. Only because the two accounts agree does the figure stand. Another often-mentioned company is Linjiedian; China Business Journal reported that by the end of September 2026, cumulative deliveries of its OmniHand series had passed 15,000 units, with cumulative deliveries across all product lines above 20,000 units. Note the wording: cumulative deliveries, with the date fixed at the end of September 2026 — how many units had been handed over up to that day. The date is spelled out so carefully because another report uses a different caliber, saying that as of October, full-year cumulative production had passed 15,000 units. One is cumulative deliveries at the end of September, the other full-year production as of October; the start and end points and the wording differ, so they cannot be cited as the same number, and this article uses only the first. 10,000 units sounds large, yet spread across a whole year and the national market it still suggests a business only just moving from samples to batches. A delivery milestone disclosed by a company says its own line is running; it does not mean the whole industry has shifted gears. Keep a company's scorecard separate from an institute's market estimate, and one firm's progress will not be misread as everyone's. ## Money Moved Faster Than Hands While hands are still gathering experience in labs, money has already rushed in. According to incomplete media tallies, total financing in China's dexterous-hand sector was 16.877 billion yuan for the whole of 2025; in the first half of 2026 alone it passed 25 billion yuan — more in half a year than in a full year. The phrase "incomplete tallies" cannot be dropped. It is not an official ledger but a figure the media assembled item by item from public disclosures, where omissions, double counts and differing calibers are all possible. Read it as a thermometer for a trend and it works; read it as an exact invoice to check every entry and you will misread it. Nor are the entrants only hand makers. According to incomplete media tallies, since 2025 nearly 30 A-share auto-parts companies have set up robotics subsidiaries or moved directly into the field. Their established trades — motors, transmissions, sensors — overlap heavily with the parts inside a single hand. The logic behind early bets is not complicated: however the full robot changes, its end still needs a hand to touch the physical world. But money tends to run ahead of orders. Financing figures describe expectations, delivery figures describe the present, and the gap between them has to be closed by round after round of tuning. ## Shipments Are Not Factory Floors Stack these figures up and the conclusion is plain: sales are rising, the forecasts are steeper, money is arriving faster, yet the bulk of buyers are still research, data collection and model training, and the share of hands on real production stations remains low. All of these can be true at once without contradicting one another. For a general reader, the two words worth separating are "happened" and "projected." The former covers the 19,200 units and the cumulative deliveries at the end of September 2026; the latter covers the expected 70,200 units and the projected 430,000 units. One is an invoice, the other a hope. Only when a hand is ordered by a production station rather than a lab bench has it truly entered the factory. Looking back at today's numbers then, it will be clear they recorded a period of preparation, not a race already finished. **Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.** --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-hand-volume-13-1105b Free to quote with attribution and a link to the original.