Tactile Sensors (hand)

Tactile sensors are the most uncertain link in this batch, and the numbers below should be read as a sketch, not a measurement: there is still no published figure for how many robot tactile modules the world actually makes³, the whole production curve here is derived and modeled. Each robot carries about sixteen sensing modules on a fleet average¹, fingertip-only hands like Tesla’s and Figure’s run near ten, fully instrumented ones like Unitree’s Dex5 reach twenty-four. Tactile is also the report’s second consumable (with batteries): fingertips are the highest-wear, field-replaceable surface on the robot, so the fleet re-buys modules on a roughly four-year clock⁵, and cumulative need, initial fitment plus that replacement flow⁸, climbs to about 1.4 billion modules by 2040 (against roughly 750 million on a build-once count). Against that, the best estimate of today’s output is only about four hundred thousand modules a year³¹¹, a number backed out of a small, unsettled dollar market, now bracketed between a low camp near US$185 million³ and a high camp near US$584 million¹⁰, not counted. The honest tension is this: the bottleneck is almost certainly not factory capacity. There is an elastic flexible-electronics and MEMS base that can pour module volume once a design freezes, so on the optimistic read global supply only falls behind the fleet’s need around 2035, meeting about 43 percent of the 2040 requirement. But the real wall sits outside any tonnage model, per-sensor calibration and yield, specs that have not converged (designs span ten to over a thousand sensing points²), and a small gel cohort that wears out in hundreds to a few thousand contact cycles⁵. If those bind, the curve is far lower and the crossover far sooner. Neither single bloc can supply the world alone: China leads production but at roughly forty-five percent of global output crosses around 2028², and the thin United States niche⁴ is short from about 2025.

A tactile sensor module gives a robot hand a sense of touch, it reads pressure, shear and slip across an array of taxels (touch pixels), in the fingertips or as full-finger and palm pads, so the hand can grasp, regrip and manipulate delicate objects without crushing or dropping them. The part procured and counted here is the module, an integrated sensing array with its readout, not the individual taxel. A humanoid hand carries on the order of five to twelve such modules, so a two-handed robot integrates roughly sixteen procured modules on a fleet average¹, near ten on fingertip-only hands (Tesla, Figure, Sanctuary), about twenty-four on a fully instrumented hand like Unitree’s Dex5, up to thirty-six on the densest arrays (XELA uSkin on an Allegro hand); on a finer taxel basis a single hand holds far more (Unitree’s Dex5 up to about ninety-four sensing points, and high-density hands like PaXini’s DexH13 run to well over a thousand sensing points across a four-finger hand²), but the supply chain ships modules. Designs range widely, from ten on a fingertip-only hand to thirty on a full array¹, which is itself part of the problem. This is distinct from the six-axis force/torque sensor at the wrist: that measures the whole-hand load vector, while tactile resolves the fine, distributed contact map across the skin of the fingers. It is an emerging, fast-moving supply base, many startups plus a hard Chinese push⁴, with low current volumes ramping quickly.

Tactile is one of only two components in the report modeled as a consumable (batteries is the other): a module is bolted into a hand, but fingertips are the highest-wear surface on the robot and are built to be field-replaced⁵, so the fleet re-buys them on a roughly four-year clock rather than fitting them once for life. The need is therefore initial fitment, the worldwide fleet multiplied by about sixteen modules¹, plus a replacement flow on top of it. Integrating the consensus shipment ramp⁸, the installed fleet reaches about 585,000 robots by 2030, ten million by 2035 and forty-seven million by 2040; with the replacement flow the cumulative module need climbs to roughly fourteen million by 2030, two hundred and fifty million by 2035 and 1.4 billion by 2040, close to double the 750 million a build-once count would imply. 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 modules it needs stay identical in all three; only the supply line changes, asking whether one bloc’s factories alone could equip the whole global fleet. Two honesty flags first. One, this is the robot-grade tactile-module line, not the broad touch-sensor market, the roughly two-and-a-half-billion-unit-a-year general tactile market is touchscreens, automotive and consumer electronics⁶, a different category. Two, no one publishes robot tactile-module unit output; the figures here are derived and modeled, low confidence throughout. The best estimate of today’s global output is about four hundred thousand modules a year, backed out of the humanoid-specific tactile dollar market, now bracketed between a low camp near one hundred and eighty-five million dollars in 2025³ and a high camp near five hundred and eighty-four million¹⁰, at a few-hundred-dollars-a-module price¹¹; the two camps sit roughly threefold apart, so the base stays derived, not counted. From there the modeled S-curve climbs steeply, to about nine million modules a year by 2030 and roughly fifty million by 2035, plateauing near one hundred and ten million by 2040, as Chinese array lines scale and the flexible-electronics base absorbs the volume. China is the aggressive ramper: Tashan’s electronic skin has entered mass production⁴ and PaXini is a leading domestic tactile-module maker², climbing from a derived one hundred and seventy thousand modules a year to about twenty-one million by 2035 and forty-six million by 2040 at roughly forty-five percent of global supply. The United States holds a far thinner niche, GelSight, Tekscan, Pressure Profile Systems and SynTouch⁴, from roughly forty thousand modules a year to about eleven million by 2040, with no announced robot-grade reshoring. Read every one of these post-2025 numbers as a modeled, optimistic ‘fab-is-elastic’ extrapolation, not a reported forecast.

On the optimistic curve, the crossover, the year cumulative global production drops below cumulative need⁸, lands around 2035. Before then the steep early ramp banks enough output to cover the still-small fleet; after it the need explodes, because the fleet is adding close to ten million robots a year by 2040, each demands about sixteen modules¹, and, as a consumable, keeps re-buying them every few years. By 2040 cumulative output meets only about 43 percent of the fleet’s 1.4-billion-module need, a shortfall on the order of eight hundred million modules, even on the elastic read. Building from a single bloc is far worse: neither China nor the United States can unilaterally supply the worldwide fleet at any point. China leads production but at roughly forty-five percent of global output² its cumulative supply crosses around 2028, and the thin United States niche⁴ is short from about 2025. The honest headline is not the precise crossover year but its fragility: push calibration and yield even modestly and the whole curve drops, pulling the wall years closer.

Tactile is the odd one out: its wall is not factory tonnage. Unlike a precision-ground bearing or screw, there is an ample flexible-electronics and MEMS fabrication base, the same base that makes the broad multi-billion-unit touch market⁶, that can scale module volume fast once a design is frozen. What is missing is technology-readiness, and it binds in three ways a capacity model cannot see. First, per-sensor calibration and yield: every array’s taxel map must be individually characterized and must hold over temperature and wear, so output is gated by metrology and test throughput, not wafer starts, a semiconductor-like elasticity only if calibration and yield scale with it. Second, spec non-convergence: working designs span from ten to over one thousand sensing points per hand², so there is no settled module to mass-produce against, the category is young enough that ‘mass production’ only began recently⁴. Third, durability: the fleet is about ninety percent solid-state (piezoresistive, magnetic and barometric arrays rated a hundred thousand contact cycles or more), with only a small, largely-academic gel and visuotactile cohort that wears out in five hundred to five thousand cycles⁵. Even solid-state fingertips do not last the robot’s full life, abrasion, cover delamination and calibration drift make them field-replaceable parts, so we now model tactile as a consumable on a roughly four-year clock, which is what pushes the 2040 need past 1.4 billion modules; a gel-heavy fleet would collapse that interval to months and pull the wall years closer still. This is exactly why bank and bottleneck analyses flag tactile (hand) for low domestic technology readiness rather than score it as a tonnage bottleneck⁷. The production S-curve above is therefore best read as a maturity projection: if calibration, yield, spec convergence and durability all come good, the elastic fab base keeps the fleet roughly supplied; if any of them stall, the more likely case for so emerging a supply base⁹, the real curve sits well below the modeled one and the crossover arrives years sooner. The uncertainty, not the point estimate, is the finding.