Cross-Roller Bearings

Cross-roller bearings are comfortable through the early-2030s, and then they tighten. The numbers here track the robot-grade precision crossed-roller tier, the raceway-ground, P4/P2-class rings that go into actuator outputs. The entire crossed-roller bearing market is itself small: only about US$549 million, or roughly three-to-four million rings a year across every grade², a distinct and far smaller category than the seventy-to-ninety-million-unit general cylindrical roller-bearing market⁴ it is often confused with. The robot-grade precision sub-tier is essentially the bulk of that small crossed-roller market today, roughly four million rings a year², and is set to climb toward about twelve million by 2030 as China’s robot-bearing lines ramp⁶, which keeps pace with the worldwide humanoid fleet’s need¹⁰ until around 2035 (the figures here track the whole global humanoid fleet, not one country’s). After that the arithmetic turns: because every robot keeps its dozen cross-roller bearings for life¹, the rings the fleet needs accumulate, and that growing stock outruns production, and production here is not the announced curve but the throttled one, because cross-roller bearings now share one finite precision-grinding pool with planetary roller screws and harmonic reducers⁷. The announced curve tops out near thirty-five million rings a year, but once the pool is shared each part is throttled below its announced output, hardest in the late-2020s ramp, when the announced lines over-subscribe the shared grinder base and hold each part to only about three-fifths of its curve, easing back toward four-fifths by 2040 as the curves self-limit at their own grinder ceiling, so cross-roller’s effective rate by 2040 is closer to twenty-eight million rings a year. The result is still a worse picture than the standalone numbers imply: by 2040 a fleet of roughly forty-seven million robots needs about five hundred and sixty-five million rings, yet pool-throttled cumulative output has delivered only about two hundred and twenty million, just under two fifths of the need met. Building from a single bloc is worse still: China-only output falls behind around 2034, and US-only output, a thin aero-defense niche, falls behind as early as 2027.

A cross-roller bearing is the high-rigidity output bearing of a rotary actuator: cylindrical rollers alternated at ninety degrees in a single race carry moment, thrust and radial load in one compact ring¹, replacing what would otherwise be a paired bearing arrangement. It sits at each rotary-actuator output, where the harmonic reducer hands torque to the limb. Crucially, crossed-roller bearings are a small, distinct market, only about US$549 million, or three-to-four million rings a year across all grades², not to be confused with the much larger general roller-bearing category⁴ that has nothing to do with humanoids. The part that matters here is the robot-grade precision sub-tier, ground and superfinished to micron tolerance, which is essentially the bulk of that small crossed-roller market, the slice (shared with robotics, semiconductor and medical motion²) precise enough to serve an actuator joint. A humanoid carries up to about twelve of them¹, roughly one per rotary actuator output, though twelve is an upper-ish, premium count: not every rotary joint takes a full crossed-roller ring, and distal joints (wrist, hand) often use angular-contact or thin-section bearings instead, which are a separate, commodity-elastic link and are not counted here.

Because a cross-roller bearing is built into a robot for life, the rings the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by twelve¹, not a per-year flow. Integrating the consensus shipment ramp¹⁰, the global installed fleet reaches about 585,000 robots by 2030, 10 million by 2035 and 47 million by 2040, requiring roughly 7 million, 121 million and 565 million precision cross-roller rings respectively. 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 rings 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. This is the robot-grade precision line, essentially the bulk of the small crossed-roller market, not the broad commodity roller-bearing total: real output follows an S-curve rather than a smooth ramp. Global precision output runs from about four million rings a year today² to roughly twelve million by 2030 and a modeled thirty-five million by 2040, climbing steeply as China’s robot-bearing capacity comes online and then plateauing against the limit of precision raceway-grinding supply⁷. These are announced output figures, the curve each maker projects for its own lines, before the shared precision-grinding throttle described below; the effective rate the fleet actually sees is lower, because cross-roller bearings draw on the same finite grinding pool as roller screws and harmonic drives. The post-2030 figures are a modeled extrapolation, a humanoid-driven step-change well above the crossed-roller market’s cited ~5% baseline CAGR³, not a reported forecast. China climbs from about 1.5 million rings to nine million by 2030 and twenty-six million by 2040 as its lines ramp⁶, led by Luoyang Hongyuan, which holds more than ninety percent of China’s robot-grade bearing segment and booked a roughly 120-million-yuan humanoid order in early 2025⁶, alongside Guoyang Precision and HONB⁵. The US holds far flatter, from about 0.3 million to 1.5 million rings, a niche precision tier (RBC, Kaydon/SKF, Schaeffler USA) serving aero, defense, medical and semicon, with no announced robot-grade reshoring; most humanoid cross-roller bearings are imported from Japan’s incumbents⁹.

The crossover, the year cumulative effective production drops below cumulative need¹⁰, lands around 2035 globally, a year earlier than the standalone curve alone would put it, because the shared precision-grinding pool throttles output once all three sibling parts draw on it. Before then, the early pool slack and the Chinese build-out bank enough output to cover the still-small fleet; after it, the durable need explodes, because the fleet is adding close to ten million robots a year by 2040 and each one demands twelve rings¹, while pool-throttled grinding cannot follow. By 2040 the effective rate is roughly twenty-eight million rings a year against an announced thirty-five million, the throttle has eased to about four-fifths by then as the curves self-limit, so cumulative output meets only about thirty-nine percent of the fleet’s five-hundred-and-sixty-five-million-ring need, a shortfall of roughly three hundred and forty-five million rings, still worse than the just-over-half its standalone curve implied. Building from a single bloc brings the wall forward. China-only output crosses around 2034, it expands quickly but starts from a small base and plateaus near twenty-six million rings a year⁶, still importing the high-end raceway grinders it needs to go further⁸, and US-only output crosses as early as 2027, since a thin aero-defense niche never keeps up with even the early fleet.

Cobots and industrial arms draw on this pool too, but less than a naive count suggests. Each articulated-arm joint has a crossed-roller output bearing, yet on many designs that bearing is integrated into the harmonic-drive unit itself (e.g. Harmonic Drive’s HFUS-2SH), and Universal Robots’ joint patent replaces it with a pair of cheaper angular-contact bearings, so we model about three discrete crossed-rollers per arm, not six. At about 3.7 million cobots and ~6 million industrial arms by 2040 that adds roughly 11 million (cobot) plus ~18 million (industrial) bearings to the humanoids’ ~565 million need; the global crossover stays about 2035 and coverage about 37 percent. They deepen the shared raceway-grinding draw without moving the wall. Quadrupeds, by contrast, do not draw on this link at all: their quasi-direct-drive leg joints ride on ordinary deep-groove and angular-contact bearings, not crossed-rollers, so the ~5-million quadruped fleet adds nothing here (its bearing draw lands on the standard support-bearing link instead). Drones are the same story writ larger: the biggest fleet in the report rides on tiny commodity motor bearings, not a single robot-grade crossed-roller, so it adds nothing to this wall.

The plateau is not set by steel, bearing-grade steel is abundant⁷, but by precision cross-roller raceway and roller grinding: the ground-and-superfinished V-races that let alternating ninety-degree rollers carry moment, thrust and radial load in one ring to actuator tolerances. New grinder lines carry long procurement lead times, the metrology scales just as slowly, and the work depends on a thin bench of skilled grinding technicians. The decisive constraint is that this grinding pool is shared with planetary roller screws and harmonic flexsplines, all three components compete for the same superfinishing lines and the same scarce technicians⁷, so the real ceiling on cross-roller bearings sits below what its standalone supply suggests once the sibling links draw their share. China’s build-out runs straight into this gate: high-end grinder localization remains below half, and the raceway grinders themselves are still largely imported⁸. It is that grinding throughput, not the ring blank itself, that caps how fast the world can build precision cross-roller bearings, and it is why production plateaus so far below the fleet’s need. This shared-pool constraint is now operationalized in the model rather than left as a caveat, though deliberately as an upper-bound approximation, not a second hard ceiling. The roughly thirty-five-million-a-year figure is cross-roller’s announced standalone curve⁷, and that curve already carries its own raceway-grinding ceiling inside its plateau; the pool layer does not apply that ceiling a second time, which would double-count it. What the pool adds is the collective coupling across the ramp window: the three precision parts, planetary roller screws, harmonic flexsplines and cross-roller raceways, actually draw on one finite precision-grinding base, set by the same OEM grinder oligopoly (Reishauer, Studer, Kapp, Matrix), the same thin bench of skilled grinding operators, and the same metrology. Through the late-2020s ramp the three announced lines together over-subscribe that shared base by roughly one-and-a-half to one-point-six times, so the throttle bites hardest there, holding each part to only about three-fifths of its announced curve around 2030, and then eases back toward four-fifths by 2040 as the individual curves flatten into their own plateaus and stop competing so hard for the same machines. For cross-roller that pulls the effective crossover forward to about 2035 and leaves only about thirty-nine percent of the 2040 need met, still worse than its standalone curve implies. The throttle is best read as a band, not a point: it is sensitive to the modeled grinder-hour weights, and it bundles three distinct machine types, thread grinding, gear or flexspline grinding, and raceway grinding, into one budget, which overstates how directly they trade against one another. Cross-roller raceway grinding is in fact the lightest draw of the three, about one times the pool’s unit weight, against roughly three times for a harmonic flexspline and ten times for a roller-screw set⁷; it is throttled at all only because it competes for the same scarce machines and people, not because it is the binding sub-bottleneck, the roller-screw thread grinding is.