China’s Dexterous Hand Industry, 2026 Part 1 — The Demand Picture

How large the market actually is, which platforms are buying, and how far today’s installed base still sits from production work — the demand half of MIR Databank’s 2026 white paper.

Executive Summary

The dexterous hand is the component on which the humanoid-robot investment thesis ultimately rests. A bipedal platform that can walk, balance and see is impressive; a platform that can pick up an unfamiliar object, reorient it in-hand and seat a connector is commercially useful. MIR Databank’s 2026 white paper is the most detailed public accounting RobotToday has seen of how far China’s supply base has closed that gap — and how far it has not.

Four conclusions matter most to an international reader. These four are RobotToday’s reading of the report rather than MIR’s own summary; figures attributed to MIR are the white paper’s, the framing around them is ours.

• The category is real but pre-commercial. MIR estimates 2026 shipments will pass 30,000 units and that the total China market will exceed RMB 3 billion (roughly USD 420 million) by 2030. Against that, only 3.5% of 2025 demand came from industrial production use, while research, education and data collection together accounted for 74.4% (Figure 15; the body text rounds these to 4% and 74%). The installed base today is overwhelmingly a training and demonstration fleet, not a working one.

• 2026 is the inflection year, not the breakout year. MIR frames 2026 as the first period in which small-batch, industry-grade orders — logistics sorting, automotive assembly, 3C electronics — replace research procurement as the marginal source of growth. The step change in volume is modelled for 2027 onward.

• Hardware differentiation is closing; software differentiation is opening. Low-level motion control is treated by MIR as a commodity capability. The defensible layer is the upper model tier — multimodal perception, task decomposition and skill generalisation — and the proprietary interaction data required to train it. The report codifies this in a five-level dexterity scale (DL1–DL5) in which most shipping product sits at DL1–DL2.

• Third-party supply will remain dominant. In the industry’s early development phase, more than 80% of humanoid OEMs sourced their hands externally rather than building them, and MIR expects third-party procurement to keep the leading share over the long term. MIR expects in-house programmes at leading OEMs to stay small-batch and differentiation-driven, because micro-motor, micro-reducer and tactile-sensor integration is too capital-intensive for any single OEM to cover across a full product range. Specialist hand suppliers therefore keep the volume.

Headline figures

Currency conversion at approximately RMB 7.15 = USD 1.00 is RobotToday’s, for reader convenience only.

1. Market Trajectory

MIR models the China market in four phases. The phase language matters more than the point estimates, because it identifies which buyer is setting the price and the specification at each stage.

Source: MIR Databank, Chapter 2. RobotToday summary.

Value against volume

MIR publishes the market twice — Figure 4 in RMB millions, Figure 5 in units — and annotates each with a compound growth rate to 2030. The two do not match.

Source: MIR Databank, Figures 4 and 5. These are the only numeric annotations on either chart; MIR publishes neither the underlying bar values nor an average selling price.

Everything MIR actually labels

The white paper charts eight years of history and forecast across six figures but publishes almost no bar values. To spare readers checking, this is the complete set of numbers legible on the charts themselves; every other figure in this briefing comes from the body text.

Source: MIR Databank, Figures 4, 5 and 11–14. As an order-of-magnitude guide, the axis ceilings are 300,000 units on Figure 11, 200,000 on Figure 12, 30,000 on Figure 13 and 15,000 on Figure 14, and the 2030E bar approaches the ceiling in each case — but MIR labels none of those bars, so they are not quoted as figures here.

RobotToday reading: units growing materially faster than value, in the history as well as the forecast, implies sustained erosion in average selling price as micro-actuator supply localises. MIR does not publish an ASP series and does not state this conclusion, but the gap between the two annotations is difficult to read any other way, and suppliers whose margin depends on today’s price band should treat it as the working assumption. The second point worth flagging is scale: the 2030 value estimate of RMB 3 billion is modest relative to the attention the category receives — roughly a mid-tier industrial component segment, not a transformative market on its own. Its strategic weight comes from being the gating component for the far larger humanoid and embodied-AI markets above it.

2. Host Platforms: Where the Hands Are Fitted

2.1 Humanoid robots

MIR defines a humanoid as a robot with a human-like torso, head, dual arms (with manipulation mechanisms and joints) and lower limbs (locomotion mechanism), using embodied intelligence for perception, cognition, decision-making and execution. Three form factors are tracked, and their attachment rates diverge sharply.

Source: MIR Databank, Table 5 and Section 5(1)3. RobotToday condensation.

MIR’s 2025 pivot points: China humanoid output reached 18,010 units — the one absolute figure MIR labels on Figure 11 — growing roughly 6.6× year on year as leading manufacturers built production lines and supply chains, following 2024 as the prototype-validation year. The 2025–2030 CAGR is put at approximately 73%, which on RobotToday’s arithmetic implies output approaching 280,000 units by 2030; MIR does not label that bar. Wheeled platforms are expected to dominate near-term volume on cost and deployability; small/mid bipeds are the fastest-growing increment; full-size bipeds grow their share steadily and pull the dexterous-hand attachment rate with them.

2.2 Industrial embodied-AI robots

MIR’s second host category is defined as a robot with a manipulator structure and a vision system that uses embodied intelligence for perception, cognition, decision-making and execution. It splits into mobile units (mobile base plus collaborative arm plus end effector — essentially the intelligent evolution of the composite robot) and fixed units (fixed base plus multi-axis industrial or collaborative arm — the intelligent evolution of the industrial robot, covering intelligent welding, random depalletising and adaptive grinding).

• Fixed units dominate current demand, concentrated in 3C electronics and automotive-component flexible assembly stations. Rigid grippers still serve heavy load/unload work; dexterous hands are appearing in micro-component assembly, flexible material lamination and precision connector insertion. Representative players: Weiyi Intelligent Manufacturing, ROKAE Robotics.

• Mobile units are mostly still in on-site validation. Production configurations run low-cost two-finger rigid grippers; dexterous hands appear only in precision material transfer and laboratory automation. Battery life and BOM cost constrain penetration. Representative players: JAKA Robotics, Siasun (DUCO).

MIR labels one absolute figure on Figure 13: 800 industrial embodied-AI robots produced in China in 2025, against a chart axis running to 30,000 by 2030. That base is two orders of magnitude below the humanoid figure, which is the honest scale of this second demand block today. MIR’s judgement is that industrial embodied AI will nonetheless be the second durable growth block for dexterous hands alongside humanoids — lower-profile than the humanoid narrative, but with clearer near-term procurement logic because the buyer is an existing automation customer with a quantifiable line problem.

3. Application Mix: 2025 Against 2030

The single most useful table in the white paper is the application split. It quantifies precisely how far the category is from production duty today, and where MIR expects the weight to move.

Source: MIR Databank, Figure 15. 2025 industrial demand is charted as a single 3.5% block; the 2030 vertical breakdown is MIR’s, and the 47.0% total is RobotToday’s sum of MIR’s five industrial slices. Chapter 6 of the white paper rounds the 2025 figures to 4% industrial and 74% research-plus-data-collection; the chart values are used here. “n/d” means not disclosed at total-market level for 2025. MIR does publish 2025 splits inside the industrial and commercial blocks — they are in the two tables below — but those are shares of their own block, not of the whole market, and MIR does not restate them against the total.

Inside the industrial block

Source: MIR Databank, Figure 16 and Table 7.

Inside the commercial block

Source: MIR Databank, Figure 17. Percentages are shares of the commercial segment, not of the total market.

Outside industry and commerce: three demand logics MIR treats separately

Chapter 6(3) of the white paper covers a third group that the percentage split largely hides, because only special operations appears in Figure 15 as a discrete 4.9% slice for 2030. For an international reader these matter disproportionately: prosthetics and special operations are demand logics that stand on their own rather than riding on humanoid volume, and two of the three have buyers who are neither factories nor laboratories.

Source: MIR Databank, Chapter 6(3) and Table 9. Condensed by RobotToday. Only special operations is separately quantified in Figure 15 (4.9% of the 2030 market); home service and wearable rehabilitation fall inside the residual “other” slice.

Named pain points by vertical

Source: MIR Databank, Tables 7–9. Condensed by RobotToday.

4. What This Means

The judgements in this section are RobotToday’s reading of the white paper, not MIR conclusions. Figures remain MIR’s and are attributed where used.

For international vendors assessing China

Two asymmetries are worth noting. First, China’s advantage in this category is described by MIR as supply-chain and iteration velocity, not as architectural novelty — the research lineage still runs through DLR, NASA, Shadow Robot and the Japanese and American university programmes. Second, the Chinese market’s 2025 demand base is research and demonstration, which means Chinese suppliers have accumulated far more hardware-hours than production-hours. The gap between a hand that performs on a demonstration stand and one that runs three shifts is the gap the entire industry is now trying to close, and it is not yet closed by anyone.

For anyone sizing the opportunity

The 2030 value estimate of RMB 3 billion is modest against the attention the category attracts; its strategic weight comes from being the gating component for far larger humanoid and embodied-AI markets rather than from its own scale. The forecast also grows in units faster than in value, which encodes an explicit assumption of sustained price erosion. Read together, those two facts argue against treating the dexterous hand as a standalone market and in favour of treating it as the constraint that determines how quickly the markets above it can open.

Prepared by RobotToday. Source: MIR Industry Research / MIR Databank, 2026 China Dexterous Hand Industry White Paper. Translation, condensation, structuring and supplier-directory mapping are RobotToday’s. Forecast figures are MIR estimates as published and have not been independently verified. Company names link to profiles in the RobotToday Supplier Directory at robottoday.com/suppliers-discovery.