Roller Screws

Roller screws were this report’s original headline bottleneck, but a corrected, bloc-weighted read of humanoid demand makes them a real yet far milder constraint. The Tesla-Optimus design carries 14 planetary roller screws¹, and work-humanoids (Figure, Apptronik, Atlas) sit in the same 10-to-14 linear-actuator class, but that is the archetype, not the fleet. The shipped fleet is dominated by rotary quasi-direct-drive designs (Unitree G1/H1) that use few or zero roller screws, so the per-robot draw varies by production bloc and rises over time as work-humanoids proliferate (China about half a screw rising to two per robot, the US about three rising to eight, the rest of the world about one rising to three). Fleet-weighted, the world needs on the order of two to four screws per robot, not fourteen. Today the world produces roughly 2.25 million sets a year², and announced global output is set to climb toward about 8 million by 2030 as China’s lines, Beite, Xinjian and Wuzhou³, come online⁴. Roller screws do not get those grinders to themselves: they share one finite precision-grinding pool with harmonic drives and cross-roller bearings⁶, so effective output falls to about 63 percent of the announced curve around 2030 and eases back toward about 80 percent by 2040. Even so the announced curve reaches about 15 million sets a year by 2040 (effective output about 12 million). The durable crossover, the year cumulative production drops below cumulative need⁸, now lands around 2039 globally rather than 2033 under the old all-fourteen assumption (the figures here track the whole global humanoid fleet, not one country’s). By 2040 a fleet of roughly 47 million robots needs about 133 million screws, a fleet average near three per robot, and cumulative effective output has delivered about 111 million, meeting roughly 83 percent of the requirement, a genuine but far shallower gap than the earlier 17 percent. Building from a single bloc is worse still: US-only output, a flat aerospace niche⁷, falls behind around 2031, and China-only output around 2035. Under the shared grinder pool it is now harmonic drives, not roller screws, that are the deepest precision-spine bottleneck.

A planetary roller screw converts a motor’s rotation into linear force using several threaded rollers planet-geared around a central screw, giving far higher load capacity and longer life than a recirculating-ball screw of the same size. A Tesla-Optimus-class work-humanoid uses about 12 to 14 of them¹ in its body linear actuators, across the elbows, wrists and legs, where push forces and shock loads are highest, while lighter ball screws handle the hands and forearms. Most of the shipped fleet, however, runs rotary quasi-direct-drive joints and carries far fewer, which is why fleet demand is modeled bloc-by-bloc rather than at a flat 14.

Because a roller screw is built into a robot for life, the screws the fleet needs are a cumulative stock, not a per-year flow. But because the per-robot count varies by production bloc and rises over time, that stock is the sum over each year’s builds of that year’s bloc-weighted screw count, not simply the robot total multiplied by fourteen¹. Integrating the consensus shipment ramp⁸, the global installed fleet reaches about 585,000 robots by 2030, 10 million by 2035 and 47 million by 2040, and the bloc-weighted screw stock it needs reaches roughly 0.9 million, 23 million and 133 million respectively, a fleet average that climbs from under one screw per robot early to about three by 2040 as linear-actuator work-humanoids take share. 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 screws 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: it is nearly flat in 2025, climbs steeply through a Chinese build-out from 2026 to 2030 (Beite at 2.6 million, Xinjian at about 1 million and Wuzhou at about 0.98 million sets a year of announced nameplate³, heavily discounted for low early utilisation⁴), and then plateaus against the limit of precision-grinder supply⁶. These are the announced output figures, nameplate before the shared-grinder-pool throttle described below is applied. Global announced output runs from about 2.25 million sets a year today² to roughly 8 million by 2030 and 15 million by 2040; China climbs from about 0.1 million to 3 million by 2030 and 5.5 million by 2040 as its lines ramp⁴; and the US holds flat at about 0.27 million, an aerospace niche served by Moog, Creative Motion Control, Nook and Exlar, with no announced humanoid lines, Tesla sources Optimus’s screws from Switzerland’s GSA⁷. The effective curve that feeds the gap below is each of these throttled by the shared precision-grinding pool, hardest in the late-2020s ramp, where effective output falls to about two-thirds of the announced curve around 2030, then easing back toward about four-fifths by 2040.

The crossover, the year cumulative production drops below cumulative need⁸, lands around 2039 globally once the shared-pool throttle is applied. Before then the Chinese build-out banks more than enough output to cover the bloc-weighted fleet need; after it, the durable need finally overtakes grinder-limited production, because the fleet is adding close to 10 million robots a year by 2040 and a rising share of them are linear-actuator work-humanoids¹. The gap widens steadily rather than explosively: cumulative production still meets about 83 percent of need at 2040, about 22 million sets short at the 2040 horizon. Building from a single bloc brings the wall forward: China-only output crosses around 2035, throttled by the same shared grinders⁴, and US-only output, a flat aerospace niche of about 0.3 million sets a year⁷, crosses around 2031.

The plateau is not set by steel but by ultra-precision, multi-start roller-thread grinding. Every roller and nut thread has to be ground to micron-class tolerances on scarce Reishauer- and Matrix-class CNC grinders⁵ that carry long procurement lead times and depend on a thin bench of skilled operators. The entire global market for these precision-grinding machines was only about $251 million in 2024⁶, a few hundred new grinders a year, each able to finish only a few thousand complete screws. Crucially, roller screws, harmonic drives and cross-roller bearings do not each get their own grinders: they draw on one finite precision-grinding pool, a narrow oligopoly of grinder OEMs, a thin bench of skilled operators, and the metrology to certify micron tolerances. Summed across all three, the announced curves demand more of that pool than it can deliver during the late-2020s ramp: as the Chinese lines come on between about 2028 and 2034 the three parts collectively over-subscribe the shared grinder base by roughly 1.6 times, so the throttle bites hardest there, effective output drops to about 63 percent of the announced curve around 2030. It then eases back toward about 80 percent by 2040 as the announced curves self-limit at their own grinder ceilings and stop racing ahead of the pool. That is why roller screws’ effective crossover is around 2039 and about 83 percent of the 2040 need is met, both worse than the standalone curve alone implies, though the corrected bloc-weighted demand makes the shortfall far shallower than the earlier all-fourteen read. And roller screws bear the most of the three: their thread grinding is the most grinder-intensive, with a single roller-screw set drawing roughly ten times the pool that one bearing does. Three honesty caveats temper the figure. First, the grinder ceiling is applied only once: it already sits inside each part’s standalone plateau, and the pool layers on the collective ramp-window coupling between the three parts, not a second copy of that ceiling, counting it twice would overstate the squeeze. Second, the throttle is a band, not a point: it is acutely sensitive to the modeled roller-screw grinder-weight of about ten (milder if the true weight is nearer seven, harsher nearer thirteen). Third, it is an upper-bound approximation, it bundles three distinct machine types (thread, gear and raceway grinders) into one grinder-hour budget, which overstates how fungible those machines really are, since you cannot grind a roller-screw nut on an idle bearing finisher. With those caveats, it is shared grinding throughput, not the screw barrel itself, that caps how fast the world can build roller screws, and it is why production plateaus so far below the fleet’s need.