Force/Torque Sensors

Six-axis force/torque sensors are the deepest binder on the whole spine. Demand is unforgiving because each sensor is durable, built into the wrist or ankle for the robot’s life, so every one the worldwide humanoid fleet³ has ever installed stays counted, and the fleet needs four of them per robot¹. Production cannot keep up. The binding pool is the narrow robot-grade six-axis slice, only about 52,000 to 60,000 sets in 2024⁵, not the broad roughly 195,000-unit all-application 6-axis market⁴ that ordinary market reports quote. That robot-grade output runs from only about 70,000 sets a year today⁵ toward a modeled roughly 5 million by 2040, while the fleet’s cumulative need climbs to about 188 million sensors over the same span³. By 2040 cumulative output has delivered only about 34 million of those, about 18 percent of the requirement, under a fifth, the worst coverage ratio of any part in the spine. The choke is not silicon or steel but per-unit multi-axis decoupling calibration⁹, which is slow to scale because it is capital- and metrology-labor-intensive. Building from a single bloc is worse still: China-only output and US-only output both fall behind the global fleet’s need from 2025 onward.

A six-axis force/torque sensor measures all three forces and three torques at a joint in one decoupled package, the feedback that lets a robot push, grasp and balance with compliant force control rather than blind position control. In a humanoid these sit at the load-bearing extremities, the wrists and ankles². We model four per robot¹, two wrists and two ankles, the configuration Tesla’s Optimus Gen2 carries among its thirty-plus onboard sensors¹; the cheap in-actuator single-axis load cells are a separate, abundant class and are not counted here.

Because a six-axis sensor is built into a robot for life, the sensors the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by four¹, not a per-year flow. Integrating the consensus shipment ramp³, the global installed fleet reaches about 585,000 robots by 2030, 10 million by 2035 and 47 million by 2040, requiring roughly 2.3 million, 40 million and 188 million six-axis sensors respectively³. In the chart below, the black line is that worldwide cumulative need, the grey band is cumulative global production, and the red wedge is the shortfall that opens once need outruns supply.

Production is the only thing that differs between the Global, China and US views. The worldwide humanoid fleet and the sensors it needs stay identical in all three; only the supply line changes, asking whether one bloc’s factories alone could equip the whole global humanoid fleet. Every figure here is the robot-grade six-axis slice, the qualification-gated sets that go into robots, not the broad roughly 195,000-unit all-application 6-axis market⁴. Real output follows an S-curve: a small calibration-gated base today, a steep humanoid sprint from 2028 to 2035, then a plateau as per-unit decoupling-calibration throughput saturates⁹. Global robot-grade output runs from about 70,000 sets a year today⁵ to roughly 780,000 by 2030⁵; the later waypoints, about 3 million by 2035 and 5 million by 2040, are a modeled humanoid-pull extrapolation, not sourced shipments. China climbs fastest as the assembly hub, from about 24,000 sets a year today to roughly 280,000 by 2030⁶, then a modeled 1.5 million by 2035 and 3 million by 2040, on the back of calibration-automation disruptors like Kunwei, which prices about 30 percent under ATI, and Link-touch⁷. The US holds a flat high-precision premium niche, from about 10,000 sets a year to a modeled 150,000 by 2040⁸, led by ATI (now Novanta) on NIST-traceable metrology rather than volume.

The crossover, the year cumulative production drops below cumulative need³, lands around 2029 globally for the combined Physical AI fleet (about 2031 for humanoids alone, pulled forward by cobots, see below). Before then, the early build-out banks just enough output to cover the still-small fleet; after it, the durable need explodes, because the fleet is adding millions of robots a year by the late 2030s and each humanoid demands four sensors¹, while calibration-limited production cannot follow. By 2040 the fleet needs about 192 million sensors against cumulative output of only about 34 million⁵, about 18 percent met, the deepest deficit on the spine. Rebasing the supply line onto the narrow robot-grade six-axis slice rather than the broad ~195,000-unit all-application market only pushes the crossover earlier and the deficit deeper. Building from a single bloc brings the wall all the way forward: China-only output crosses as early as 2025⁶, and US-only output crosses in 2025 as well⁸, since neither bloc’s base, even China’s fast-ramping one, ever covers the whole global fleet’s accumulating need on its own.

Humanoids are not the only draw on this narrow pool. By 2040 the collaborative-arm fleet reaches about 3.7 million units, and a fleet-blended cobot carries roughly one six-axis force/torque sensor, Universal Robots ships a wrist sensor as standard, Franka and Doosan wire one into every joint, and the Chinese-majority brands estimate torque from motor current with none. That adds about 4 million sensors to the roughly 188 million humanoids need by 2040, about 192 million in all, and because cobots ship in volume today (~65,000 built in 2024 vs ~18,000 humanoids) it pulls the global crossover forward from about 2031 to about 2029. Traditional industrial arms, by contrast, barely touch this pool: a classic six-axis welding or handling arm senses force from motor current and joint torque rather than a dedicated six-axis wrist sensor, which stays a minority add-on for force-controlled assembly, so the ~6-million-strong industrial-arm fleet adds a negligible draw and does not move the wall. Quadrupeds are the same story: a four-legged robot closes its balance loop with an IMU and joint-torque estimation, and any foot-contact sensing is a simple force switch, not a six-axis wrist sensor, so the ~5-million quadruped fleet also adds essentially nothing to this pool. Coverage stays about 18 percent met: the deficit is so deep that cobots deepen it rather than create it. Drones round it out: a multirotor has no wrist and no contact sensing, so despite being the largest fleet in the report it adds nothing to this pool.

The plateau is not set by silicon or steel but by per-unit multi-axis decoupling calibration. Each finished sensor must be individually characterized on a metrology rig to build the decoupling matrix that cross-cancels six-channel crosstalk and temperature drift⁹, a sequential, per-unit procedure that is slow to scale because it is capital- and metrology-labor-intensive, so throughput rises with rig-hours and skilled metrologists rather than with floor space. That is exactly the axis on which the Chinese entrants are competing: Kunwei and Link-touch are winning on calibration-automation throughput, undercutting ATI by about 30 percent, rather than on the sensing element itself⁷. The market remains concentrated in a thin tier of precision-metrology incumbents, ATI (Novanta) leads at about 14 percent share, with Schunk second⁴, and it is their calibration capacity, not any raw-material node beneath it, that caps how fast the world can make trustworthy six-axis F/T.