Cycloidal Reducers

Cycloidal is the link that does not bind on volume, and that is the honest headline. Unlike the grinder-bound parts, cycloidal’s exposure is supplier concentration, not a wall the world cannot build past. Two facts drive it. First, humanoids barely use cycloidal at all: mainstream designs (Optimus, Figure, Apptronik, Unitree) run harmonic gears plus roller screws and carry close to zero cycloidal reducers, Tesla’s Optimus drives its rotary joints with harmonic reducers and uses no cycloidal at all⁶, so we model only about 0.3 reducers per robot, and most designs use none¹. Second, the RV/cycloidal industry is already large and over-built: it ships roughly four-point-six million units a year against about eight-point-seven-five million units of nameplate², structural overcapacity before a single line is added for robotics. Multiply the tiny per-robot count by even a forty-seven-million-robot worldwide fleet and the cumulative need, about fourteen million reducers by 2040, is a fraction of what the industry produces over the same years. So global production never falls behind, and China-only never falls behind either: both run several times the fleet’s need through the 2040 horizon. The one view that crosses is the thin US niche, about one percent of world output⁴, which falls behind around 2032 and meets only about eight percent of the 2040 need. The real risk lives elsewhere: Nabtesco holds roughly sixty percent of the precision-reducer market by value³, so the question is pricing and qualification at the high-torque end, not whether enough reducers can be built.

A cycloidal, or rotate-vector (RV), reducer is the heavy-torque workhorse of robot joints: an eccentric cam drives cycloidal discs against ring pins, delivering very high torque density plus the shock-load and overload tolerance that strain-wave (harmonic) gears lack⁸. In an industrial arm it sits on the base axes, carrying the highest-torque rotary joints. Humanoids mostly do not use it: harmonic gears plus roller screws cover the same joints lighter and more compactly, so cycloidal appears only in a few niche designs (for example Fourier’s GR-1) rather than the mainstream bill of materials¹. That is why we model only about 0.3 RV reducers per robot, with most designs at zero¹, the lowest per-robot count of any link in this batch, and the reason cycloidal is a concentration story rather than a volume one.

Because a cycloidal reducer is built into a robot for life, the reducers the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by about 0.3¹, not a per-year flow. Integrating the consensus shipment ramp⁵, the worldwide installed humanoid fleet reaches about 585,000 robots by 2030, 10 million by 2035 and 47 million by 2040, requiring only roughly 176,000, 3 million and 14 million cycloidal reducers respectively. Hold those need figures against an industry already shipping about four-point-six million reducers every year²: even the 2040 cumulative stock of fourteen million is barely three years of current output. In the chart below, the black line is that worldwide cumulative need and the grey band is cumulative global production, and the band stays far above the line at every point, so no red shortfall wedge ever opens.

Production is the only thing that differs between the Global, China and US views. The worldwide humanoid fleet and the reducers 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. Real output follows an S-curve rather than a smooth ramp. Global RV/cycloidal output runs from about four-point-six million units a year today² to roughly six-point-five million by 2030 and a modeled twelve million by 2040, growing off the industrial-robot baseline⁵ as Nabtesco doubles RV capacity by 2026 and a third Hamamatsu plant lifts it toward two million precision gears a year by 2030⁷, alongside the broad Chinese ramp. The post-2030 figures are a modeled extrapolation above the roughly six-percent industrial baseline, not a reported forecast. China climbs from about one-point-seven million units to six million by 2040, led by a deep domestic base, Zhongda Leader, Zhejiang Zhenkang, Shuanghuan and Qinchuan², that already holds roughly thirty-seven percent of global RV production by value, with unit share higher still because Chinese reducers price well below Nabtesco⁴. The US holds far flatter, from about fifty thousand units a year to ninety thousand, a one-percent niche (Onvio and a thin defense/medical base) with no announced robot-grade reshoring⁴; most reducers are imported from Japan’s incumbents.

For the worldwide view, the answer is never: cumulative global production stays above the fleet’s cumulative need at every year through the 2040 horizon, so there is no crossover. By 2040 the world has produced on the order of a hundred and twenty-seven million reducers against a combined fleet need of about thirty-two million², still roughly four times cover. Industrial arms, now modeled as an embodiment, are cycloidal’s real consumer, about three RV reducers each on the base joints, ~18 million by 2040, alongside the humanoids’ thin ~14 million (cobots, quadrupeds and drones use none, a QDD quadruped is all planetary, and a drone has no gearbox at all); yet even together they draw only a slice off a large, over-built base. China-only never falls behind either: its cumulative output runs about four times the 2040 need on its own. The single view that crosses is the US-only line: a roughly one-percent niche⁴ that produces about fifty thousand reducers a year cannot keep up with a durable stock that reaches three million by 2035, so US-only output falls behind around 2032 and meets only about eight percent of the 2040 fleet need. That contrast is the whole point of this link: the wall is not how many reducers the world can build, it can build far more than the fleet needs, it is who builds them.

Cycloidal is the comfortable link on volume, so its real constraint is concentration. The binding input is genuine, precision cycloidal-disc and cam grinding, needle-roller bearings and zero-backlash assembly carry long leads and demand a thin bench of skilled grinders, and that skill base overlaps qualitatively with the precision-grinding pool that throttles roller screws, harmonic flexsplines and cross-roller raceways. But cycloidal is not modeled as a member of that shared pool: because even with industrial arms the robot draw stays well under the over-built RV base, it is treated standalone, and its production curve is its own announced output, not a throttled slice. What concentrates the risk is ownership, not capacity. Nabtesco holds roughly sixty percent of the precision-reduction-gear market by value³ and about thirty-five percent by unit¹, its grip tightest at the high-torque, high-margin end, looser across the broad unit base. Sumitomo and Spinea fill the Western tier; a broad Chinese base (Zhongda Leader, Zhenkang, Shuanghuan, Qinchuan) now holds about thirty-seven percent of global production value⁴ and is the reason the unit market does not choke even as Nabtesco defends the value crown. So the exposure here is pricing and qualification at the premium, high-torque end, and the long requalification cycle to move a heavy joint off Nabtesco, not a physical ceiling on how many reducers the world can build. On raw volume, cycloidal scales ahead of the build ramp at every point.