Robot hands: which startup is ahead?

Which robot-hand startups belong in this comparison?

The real robot-hand startup field currently contains four scaled manufacturers, several fast-moving technical challengers, and a smaller group of specialists leading in touch, research, or prosthetic-derived hardware.

We include companies that sell dexterous hands independently or are clearly building them for external robot manufacturers. That gives us LinkerBot, AGILINK, Inspire Robots, Xynova, Wuji Technology, Sharpa, PaXini, BrainCo, Tesollo, DexRobot, and Psyonic. Shadow Robot also belongs in the analysis as a technical benchmark, although a company founded in 1987 stretches the usual definition of a startup.

We exclude Tesla, Figure, Unitree, Agility Robotics, and other humanoid manufacturers whose hands mainly serve their own robots. Their technology can still be excellent, but an external customer generally cannot choose those hands as an independent component. We also leave out conventional industrial-gripper companies such as Schunk and Robotiq because their products solve a narrower manipulation problem.

The boundaries remain slightly messy. PaXini began with tactile sensors and now sells hands and humanoids. BrainCo and Psyonic came from prosthetics. AGILINK inherited technology and manufacturing experience from AgiBot. Sharpa sells both a hand and a broader manipulation platform. Those differences make the comparison richer, although they also make funding and shipment numbers harder to line up.

Funding figures are especially uneven. Chinese companies often announce the size of one round without publishing their lifetime total. Some databases also count grants, secondary transactions, or financing raised for wider businesses that include products beyond robot hands. We therefore use disclosed minimums and label uncertain totals instead of manufacturing a clean but misleading dataset.

Company	What it currently sells or develops	Publicly disclosed funding
LinkerBot	A wide range of linkage, tendon-driven, and direct-drive hands for humanoids, research, and industry	At least $150 million by late 2025, followed by an undisclosed B+ round at a reported $3 billion valuation
AGILINK	OmniHand dexterous hands, industrial grippers, tactile hardware, and manipulation software	Nearly RMB1 billion across four rounds
Xynova	Flex dexterous hands, micro electric cylinders, motors, and integrated joints	Nearly RMB1 billion
Inspire Robots	Commercial dexterous hands and the miniature linear actuators that power them	C1 and C2 rounds totaling several hundred million RMB; full lifetime total undisclosed
PaXini	Multidimensional tactile sensors, sensor-rich hands, data systems, and humanoids	More than RMB1 billion in its latest disclosed round alone
BrainCo	Prosthetic hands, dexterous robot hands, and brain-computer-interface products	Third-party totals range widely because funding covers the whole company rather than only robot hands
Wuji Technology	Twenty-DoF hands, miniature motors, teleoperation gloves, and data-collection tools	Undisclosed
Sharpa	The 22-DoF Wave hand, tactile sensing, manipulation models, and humanoid systems	Undisclosed
Tesollo	Multi-jointed grippers and fully actuated humanoid hands	Series B completed; amount undisclosed
DexRobot	A family of tactile dexterous hands and the DexTele data-collection system	Undisclosed
Psyonic	A touch-sensing prosthetic hand also sold to robotics researchers	At least $2.19 million tracked publicly, although databases disagree on the complete total
Shadow Robot	Research-grade dexterous hands and robust machine-learning hardware	Undisclosed

Is there a clear robot-hand leader today?

LinkerBot is the clearest overall leader today, with AGILINK and Inspire Robots forming the main commercial challengers.

LinkerBot combines a broad product catalog, aggressive prices, vertically integrated components, thousands of monthly units, and access to external customers. Reuters recently described it as the global leader in highly dexterous robot hands and reported production of almost 5,000 units per month. The company was preparing to double that output while raising capital at a valuation above most complete humanoid-robot startups.

AGILINK has emerged as the most dangerous challenger. It inherited a running business from AgiBot rather than beginning with an empty factory, then quickly expanded the OmniHand range, raised four rounds, and reported early profitability. Inspire Robots has a longer history, more than 700 reported customers, proprietary actuators, and years of manufacturing experience. Both have already moved well beyond laboratory quantities.

The technical gap is much smaller. Xynova, Sharpa, BrainCo, Wuji, PaXini, and Shadow Robot each look stronger on at least one important dimension. LinkerBot leads overall because it has industrialized the product, not because it wins every technical comparison.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who is shipping the most robot hands?

LinkerBot appears to have the highest current shipment capacity, while AGILINK and Inspire Robots have reached the same broad commercial tier at lower disclosed monthly rates.

LinkerBot reported shipping 10,000 hands during 2025. Production then rose to almost 5,000 per month, according to Reuters and a recent Guardian report. Holding that rate for a full year would produce nearly 60,000 hands, around six times the company’s previous annual total. Actual sales may differ from factory output, but no independent hand supplier has publicly disclosed a comparable current run rate.

Inspire Robots delivered more than 10,000 hands in 2025, up from around 2,000 in the previous year. That works out to roughly 830 units per month on average and represents about fivefold annual growth. The company now targets between 30,000 and 50,000 deliveries, although that remains guidance rather than completed demand.

AGILINK reported more than 8,000 hand deliveries within roughly five months of becoming independent. That implies an average near 1,600 per month. Some products, customers, inventory, and production systems came from AgiBot, so the figure does not represent volume built entirely from scratch.

Xynova is preparing annual capacity for 10,000 hands. That would equal only about two months of LinkerBot’s current reported output. That gap is pretty brutal.

Company	Latest useful volume evidence	What the number tells us
LinkerBot	Almost 5,000 units of monthly production	Clear lead in current manufacturing rate
AGILINK	More than 8,000 hands delivered within roughly five months	Already operating at commercial scale, partly inherited from AgiBot
Inspire Robots	More than 10,000 deliveries during 2025	Established volume producer with rapid annual growth
Xynova	Building capacity for 10,000 hands per year	Entering scaled production, still well behind the top three
Other specialists	No comparable public total	Technical progress remains easier to verify than commercial scale

Who has built the strongest robot-hand product lineup?

LinkerBot has the strongest commercial lineup today, while Xynova, AGILINK, Sharpa, and BrainCo are pushing harder on technical performance.

LinkerBot covers several mechanical architectures and complexity levels. Its catalog ranges from simpler hands with fewer active joints to 20-DoF systems such as the L20 and more advanced research configurations. Prices reportedly run from roughly $600 to $15,000. A university, humanoid developer, data-collection company, or factory can therefore buy different levels of dexterity from the same supplier.

Inspire Robots has spent roughly a decade refining commercial hands and the miniature linear actuators inside them. The company reports more than 20 product models, although its best-known volume products use six actively controlled degrees of freedom and coupled finger joints. They offer less independent movement than newer 20-DoF designs, but their lower complexity can reduce weight, cost, and integration work.

AGILINK has built a smaller but well-organized range. OmniHand 3 Ultra-T targets advanced dexterity and tactile manipulation. OmniHand 3 Lite targets compact humanoids and higher-volume use. OmniPicker 3 focuses on industrial grasping. The newer Ultra-M uses 20 direct-drive joints for dense-contact research.

The strongest individual specifications currently come from Xynova, AGILINK, Sharpa, and BrainCo. Xynova’s Flex 2 weighs around 400 grams and combines 23 degrees of freedom with a 12-kilogram maximum grasp load and a 4-kilogram continuous operating load. AGILINK’s Ultra-T weighs less than 500 grams, carries 5 kilograms, and provides more than 25 degrees of freedom, including passive movement.

Sharpa’s Wave has 22 active joints and company-reported tactile sensitivity down to 0.02 newtons. BrainCo’s Revo 3 uses 21 independent motors, full-palm sensing at a reported 0.01-newton resolution, and optional visuotactile fingertips. Wuji Hand 2 adds 20 active degrees of freedom, a 1,000-hertz control rate, ROS 2 support, simulation files, and matching teleoperation tools.

LinkerBot’s L20 is heavier at around 1.1 kilograms, with 20 active degrees of freedom, a 10-kilogram maximum load, 100 newtons of grasping force, and ±0.2-millimeter repeat positioning. It gives up some compactness in exchange for force, availability, and proven manufacturability.

These specifications remain difficult to compare directly. Companies use different definitions of degrees of freedom, payload, continuous load, force resolution, and tested lifespan.

Hand	Main technical strengths	Main limitation today
Xynova Flex 2	23 DoF, roughly 400 g, 12 kg maximum load, 4 kg continuous load	Scaled reliability still needs customer proof
AGILINK OmniHand 3 Ultra-T	More than 25 DoF, under 500 g, tactile sensing, 5 kg payload	Many performance claims remain company-reported
Sharpa Wave	22 active DoF and very sensitive tactile arrays	Limited public shipment and durability data
BrainCo Revo 3	21 active DoF, full-palm tactile sensing, visuotactile options	Wider company focus makes hand traction harder to isolate
LinkerBot L20	20 active DoF, 10 kg load, 100 N grasp, strong availability	Heavy compared with the newest high-DoF hands
Wuji Hand 2	20 active DoF, 1,000 Hz control, integrated software and teleoperation	Few public customer or production figures
Shadow DEX-EE	Built for repeated machine-learning experiments and fast repair	Three-finger research design rather than a human-like production hand

Who leads robot-hand tactile sensing and training data?

PaXini currently leads the tactile-sensor layer, while AGILINK, Wuji, Sharpa, and BrainCo have built the strongest combinations of touch, hardware, and training tools.

PaXini started with multidimensional tactile sensing and expanded into hands, data systems, and complete humanoids. Its DexH13 reportedly carries 1,140 sensing points and an integrated vision module. That allows the hand to detect contact across a much larger surface than systems with simple pressure sensors only at the fingertips.

AGILINK has a stronger route from touch to repeated commercial behavior. Its hands combine tactile arrays, palm vision, force control, and several mechanical architectures. The company’s SDK also supports a compact model with more than 400 tactile points. Thousands of delivered units could generate useful contact and failure data if customers permit the company to collect it.

Wuji has built a coherent data-collection stack around its hand and glove. The glove captures movement, pressure, and touch while its software retargets human motion to the robot. Wuji was also identified as the manufacturing partner behind the Genesis Hand used in Genesis AI demonstrations.

Sharpa places roughly 1,000 tactile sensors in each fingertip of its most advanced system. BrainCo combines full-palm sensing with visuotactile fingertips, while DexRobot has introduced DexTele as a dedicated teleoperation and data-acquisition product.

PaXini leads the sensor technology itself. AGILINK and Wuji currently have the clearest paths for turning touch data into repeatable robot skills.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Which robot hand offers the best value for money?

LinkerBot currently offers the best overall value because buyers can choose between genuinely cheap entry products and advanced hands from a supplier already manufacturing at scale.

Wired reported a LinkerBot price range of roughly $600 to $15,000. The cheapest products will not match a 20-DoF research hand, but they let developers begin collecting data or testing humanoid designs without spending tens of thousands of dollars on each hand. LinkerBot also supports common interfaces such as CAN and RS-485, and several models have ROS-compatible tools and distributor availability.

AGILINK also looks competitive. The earlier OmniHand 2025 sells for about $4,420, with tactile and non-tactile configurations available. The new Lite model is aimed at mass deployment, while the Ultra-T’s quick-release tendon system is designed to reduce repair time. Ten-minute component replacement could save more money than a lower purchase price in a fleet running every day.

Xynova may deliver the best high-end capability per dollar, although public prices still come mainly from distributors rather than a stable global catalog. The combination of low weight, strong load capacity, tactile sensing, and replaceable external parts makes Flex 2 attractive for humanoid builders.

Tesollo’s DG-5F has been listed at around $21,000, while SharpaWave has appeared around the $50,000 level in specialist databases. Psyonic’s Ability Hand also costs around $20,000 in robotics channels. These products can make sense for advanced research or specialized work, but they are harder to justify across a large humanoid fleet.

Who has the strongest real-world robot-hand deployments?

AGILINK currently provides the clearest evidence of hands working through full shifts in commercial environments.

The company reports nearly 1,000 units operating for eight-hour shifts in warehouses, pharmacies, factories, and other service settings. Those deployments sit inside a larger base of more than 8,000 delivered hands and 10,000 grippers. The figures come from the company rather than audited customer disclosures, but they reveal more than a staged manipulation video.

LinkerBot has more manufacturing volume and names research laboratories, manufacturers, and humanoid developers as customers. Wired also reported that the company uses its own hands on assembly lines that manufacture additional hands. We still lack a clean breakdown between paid external deployment, distributor inventory, internal use, research equipment, and hands attached to prototype humanoids.

Inspire Robots has the broadest disclosed customer base, with more than 700 customers across research, education, healthcare, industry, and commercial robotics. Its longer history suggests that many buyers have moved past initial evaluation, although the company publishes little information about renewals, fleet expansion, or operating hours.

Psyonic offers a different type of proof. Its Ability Hand is used daily by amputees, is covered by Medicare and other American insurers, and contributed to a business with $5.35 million in 2025 revenue. Its recent collaboration with ABB now uses human-generated prosthetic-hand data to improve robotic dexterity.

Which startup can manufacture robot hands at real scale?

LinkerBot is currently the only independent robot-hand startup producing at a rate that could supply several large humanoid programs at once.

The company operates multiple factories, makes key motors and reducers internally, and currently produces thousands of hands each month. Vertical integration gives it tighter control over cost and supply than a startup assembling motors, tendons, sensors, and reducers purchased from several outside vendors. China’s electric-vehicle supply chain also provides access to miniature motors, batteries, electronics, and precision manufacturing.

Inspire Robots has already passed the 10,000-unit annual threshold and manufactures its own micro linear actuators. Its next target would place output somewhere between 2.5 and five times its previous annual deliveries, although that depends on real orders arriving.

AGILINK started with supply-chain and production knowledge from AgiBot, which explains how it reached volume so quickly. Replaceable tendons, compact modules, and a separate industrial gripper also reduce the need for heavily customized customer hardware.

Xynova is building a 5,400-square-meter factory with targeted annual capacity for 10,000 hands and 200,000 micro electric cylinders. The cylinder target may eventually matter more than the hand target because Xynova could sell components to other manufacturers. For now, LinkerBot remains far ahead on completed production.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Which robot hand is the most reliable?

Shadow Robot has the strongest public evidence of surviving punishing machine-learning experiments, while commercial lifetime claims from the newer manufacturers remain difficult to compare.

Google DeepMind helped Shadow develop DEX-EE after older anthropomorphic hands repeatedly broke tendons during reinforcement-learning work. DEX-EE uses three large fingers, replaceable modules, active compliance, and a mechanical layout built to tolerate impacts and bad control commands. DeepMind researchers reported that the system could operate reliably for at least 300 hours, a meaningful result for experiments involving thousands of collisions and failed grasps.

Xynova claims millions of open and close cycles for Flex 2. DexRobot describes DexHand021 as a mass-production platform designed for maintenance and compatibility. AGILINK gives its OmniPicker industrial gripper a one-million-cycle lifespan. Those numbers are encouraging, though a controlled cycle test does not reproduce side impacts, twisted tendons, dust, heat, overloaded fingertips, and unpredictable contact with real objects.

An independent study of the Inspire RH56DFX shows why outside testing remains valuable. Researchers measured force-limit overshoot of up to 1,618% at high speed with the original interface. After calibration, modeling, and a new controller, the same hardware reached an 87% success rate across 300 grasps and 65% on peg insertion, compared with 10% for one baseline. The hand was useful, but getting there required more engineering than the product page suggested.

Psyonic also deserves attention. Its compliant fingers are built to withstand blunt impacts because prosthetic users cannot treat their hand like fragile laboratory equipment. Daily human use provides a demanding kind of durability testing, even though the six-motor architecture is simpler than the leading humanoid hands.

For now, Shadow wins the evidence contest. LinkerBot, Inspire, AGILINK, and Xynova may already deliver longer industrial lifetimes, but independent tests have not yet shown it.

Which robot-hand startup is growing fastest now?

AGILINK is growing fastest relative to its short independent history, while LinkerBot is adding the most absolute manufacturing capacity.

AGILINK became independent at the start of the year, completed four rounds, passed a $1 billion reported valuation, expanded its product range, and announced first-quarter profitability. Its 8,000-plus hand figure includes a business inherited from AgiBot, yet the company has still compressed several years of normal startup development into a few months.

LinkerBot’s acceleration is larger in physical terms. Its latest monthly production rate would generate roughly six times the hands it shipped throughout 2025. The company also closed a round at a reported $3 billion valuation and is already discussing another at twice that level. Investors are valuing it more like a future category standard than a niche component maker.

Inspire Robots grew annual deliveries by about 400% between 2024 and 2025 and recently raised several hundred million RMB from industrial and state-backed investors. It has less hype than the newer unicorns, but its combination of growth, customer count, actuator production, and operating history makes it one of the more credible challengers.

Several smaller companies have gained momentum lately without yet matching the leaders commercially. DexRobot introduced a full hand family and DexTele at Automate. Tesollo completed a Series B and began preparing for a Korean listing after 2027. Sharpa connected its tactile manipulation stack to Nvidia’s Isaac GR00T ecosystem. Psyonic began working with ABB on human-generated manipulation data.

BrainCo delivered the freshest major product update. Its Revo 3 appeared at the World Artificial Intelligence Conference with 21 active degrees of freedom, full-palm tactile sensing, and a 70-newton reported grip force. BrainCo has enough capital and prosthetic experience to rise quickly, although we still need customer volume for the robot version.

Is robot-hand funding producing real leadership?

LinkerBot and AGILINK are converting capital into commercial progress most convincingly, while several heavily funded challengers are still investing ahead of demand.

LinkerBot had disclosed at least $150 million before completing its latest round. That money has produced factories, internal component manufacturing, a broad catalog, lower prices, and high output. Its funding therefore supports visible operating assets rather than a long sequence of laboratory prototypes.

AGILINK has also moved quickly from capital to products and deployment. Nearly RMB1 billion is substantial, but the company gained a head start through AgiBot’s team, designs, suppliers, and customers. We should therefore credit both capital efficiency and the value of the business it inherited.

Xynova has raised a similar amount while shipping far fewer hands. Its spending has gone into a new factory, micro cylinders, motors, integrated joints, and two generations of hand hardware. That looks slower when measured against current revenue, but the component platform could eventually support many products and external customers.

PaXini raised more than RMB1 billion in one recent round and may pursue a Hong Kong listing. Its capital supports tactile sensors, hands, data systems, and a full humanoid. That breadth creates a larger opportunity and a larger execution burden. PaXini’s leadership in tactile sensing is clearer today than its leadership in finished hands.

BrainCo has raised hundreds of millions across a wider neurotechnology company. The Revo 3 benefits from those resources, but attributing the whole funding total to the robot-hand race would exaggerate its position. The same caution applies when comparing Psyonic’s modest tracked funding with Chinese companies financed to build factories. Psyonic generated millions in revenue from a regulated prosthetic product; that is a different capital model rather than a direct manufacturing comparison.

If you want more recent data on this point, please see our latest humanoid robotics market report.

Which startup leads each part of the robot-hand market?

Robot hands currently have several specialized leaders, although LinkerBot covers more important subsegments than any other independent supplier.

LinkerBot leads mass-produced external hands. It has the broadest combination of prices, architectures, and output. Inspire Robots leads the established affordable segment, especially for customers comfortable with fewer independently controlled joints. AGILINK has the strongest evidence of operational deployment and the best chance of connecting a large installed base with tactile manipulation software.

Xynova leads the high-performance weight-to-load contest on published specifications. Sharpa is a leading candidate for high-resolution tactile research. BrainCo has built one of the strongest direct-drive and visuotactile platforms. PaXini remains the most credible pure tactile-infrastructure company.

Wuji stands out in teleoperation and data collection. Its hand, glove, software, simulation resources, and manufacturing relationship with Genesis AI form a coherent manipulation stack. The company could become more important than its limited funding and shipment disclosures currently suggest.

Shadow Robot leads robust machine-learning research. Tesollo is one of the strongest independent suppliers outside China, supported by a 20-DoF product, international ambitions, and a newly launched IPO process. Psyonic leads prosthetic-derived validation, while DexRobot is building a broad open platform around production hands and teleoperation data.

The category may eventually separate into two businesses. One group will manufacture affordable hands by the tens of thousands. Another will sell sensing, data, simulation, control software, and learned manipulation skills. LinkerBot is best positioned in the first group. AGILINK, Wuji, PaXini, and Sharpa are trying to compete across both.

Can we trust the robot-hand numbers?

We can trust the broad competitive hierarchy more than the exact market shares, shipment totals, or performance rankings.

LinkerBot says its 2025 shipments represented 80% of global high-DoF demand. Inspire Robots separately reported more than 10,000 deliveries during the same year. Some Chinese industry estimates placed total domestic dexterous-hand sales near 19,200 units. As seen above, LinkerBot and Inspire alone would exceed that estimate before counting AGILINK, PaXini, Unitree, BrainCo, or other suppliers. The definitions clearly differ. So, yes, the dataset is messy.

One company may count hands shipped to distributors, while another counts end-customer deliveries. A pair of hands may appear as two units. Lower-DoF prosthetic-style hands may be included in one dataset and excluded from another. Internal transfers to an affiliated humanoid company can also look like external demand unless the company publishes a customer breakdown.

Performance claims have the same problem. Degrees of freedom may include passive joints. Maximum load can describe a brief static test rather than continuous operation. Cycle-life tests may run without payload. Tactile resolution can refer to the smallest laboratory measurement rather than useful accuracy across the whole palm.

We therefore put more weight on recurring operation, repairability, customer access, manufacturing rate, and independently observed tests. LinkerBot, AGILINK, and Inspire have clearly reached commercial scale. The precise order among the newer technical challengers remains less certain.

Which robot-hand startups are actually ahead?

LinkerBot is ahead overall today, AGILINK is the closest challenger, and Inspire Robots remains the strongest established alternative.

LinkerBot wins because it already behaves like an industrial supplier. Buyers can choose several types of hands, purchase them at different price points, and order them from a company manufacturing thousands per month. Its lead is widest in production, affordability, and product coverage. Independently verified reliability and detailed customer evidence remain weaker.

AGILINK ranks second. It has rapidly built a strong product family around tactile sensing, direct-drive and tendon architectures, maintainability, and deployment. As pointed out above, the company reports thousands of delivered hands and nearly 1,000 units working full shifts. Its main caveat is that much of its starting position came from AgiBot.

Inspire Robots ranks third. It has years of experience, proprietary actuators, hundreds of customers, and more than 10,000 reported deliveries in 2025. Some newer products offer greater independent finger control and richer tactile sensing, but Inspire has already proved that customers will buy its hands in volume.

Xynova takes fourth place through hardware strength and vertical integration. Flex 2 may be the most balanced hand on paper, and the micro-cylinder business could become a valuable supply-chain position. Its factory ramp and customer deployments will determine whether it can close the gap.

Wuji ranks fifth because its hand, glove, software, simulation tools, and work with Genesis AI form one of the strongest emerging manipulation stacks. PaXini follows through its tactile-sensing lead. Sharpa and BrainCo have advanced hardware but still need clearer shipment and customer evidence.

LinkerBot stays first for now because it has already crossed the gap between building an impressive hand and supplying a market.

Rank	Startup	Why it currently holds this position
1	LinkerBot	Clear lead in manufacturing scale, pricing range, product breadth, and availability
2	AGILINK	Strong deployments, rapid growth, tactile systems, maintainability, and a credible software strategy
3	Inspire Robots	Proven volume, hundreds of customers, proprietary actuators, and long manufacturing experience
4	Xynova	Excellent hardware specifications and vertical integration, with production still ramping
5	Wuji Technology	Strong hand, glove, software, simulation, and data-collection ecosystem
6	PaXini	Leading tactile-sensor position and substantial funding, with less transparent hand volume
7	Sharpa	One of the most capable tactile hands, but limited commercial evidence
8	BrainCo	Advanced 21-DoF platform backed by deep prosthetic experience and substantial company resources
9	Tesollo	Credible 20-DoF hardware, international reach, and improving financial maturity
10	DexRobot	Broad hand lineup and promising teleoperation stack, without enough disclosed customer traction
11	Psyonic	Proven durability, real revenue, open interfaces, and useful touch sensing, but lower dexterity

If you want more recent data on this point, please see our latest humanoid robotics market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.