Ball Screws

Ball screws are the hand-side counterpart to roller screws, and they cross over in 2036, late enough to be a warning rather than an emergency, but real, and now slightly ahead of roller screws (2039) once roller demand is corrected downward. The precision ball-screw market runs at roughly US$1.85 billion a year today¹, which converts to on the order of ten million units a year of precision-tier output. A humanoid carries about twelve of them, roughly eight miniature screws driving fingers, plus about four light wrist and forearm joints, distinct from the twelve-to-fourteen heavy planetary roller screws in the body's linear actuators⁷. The count was revised from an earlier sixteen, then to twelve as dexterous-hand degrees of freedom rose: the original figure paired one screw with each of the twelve hand motors this report already models, which quietly assumed every hand in the fleet was screw-driven, when roughly half are tendon-driven instead⁵. The screw-versus-tendon split is now a single shared parameter that both links read. Because a fleet of about 47 million robots by 2040⁹ needs some 565 million screws while cumulative production delivers about 277 million, 49 percent of the 2040 requirement is met and the gap reaches about 2× at the 2040 horizon. That makes ball screws now the deeper of the two linear-motion problems: the corrected, bloc-weighted roller-screw demand meets 83 percent of its 2040 need against ball screws' 49 percent, because dexterous-hand screw counts rise as roller-screw counts fall across the shipped fleet. Building from one bloc is far worse: China-only crosses in 2034 and the thin US base⁸ in 2028. What keeps this a moderate rather than severe finding is a manufacturing fact: unlike roller screws, ball screws have a cold-forming escape route. Grinding a ball-screw thread takes about six minutes; cold-forming the same profile takes about three seconds³, and precision-rolled screws now reach C3 to C5 accuracy at exactly the short strokes robot hands use⁴. Ball screws therefore do not draw on the scarce precision-grinding pool that throttles roller screws, harmonic drives and cross-roller bearings. One caveat governs the whole section: no published source reports ball-screw unit volumes. Every figure here descends from a revenue number divided by an assumed average selling price, so treat the direction as the finding and the year as indicative.

A ball screw converts rotation into linear motion through recirculating steel or ceramic balls running in a helical raceway. Where a planetary roller screw is the heavy-duty choice, more contact points, far higher load capacity, used in the knee, ankle and torso actuators, a miniature ball screw is the compact, back-drivable, low-friction choice for small forces. In a humanoid that means the hands above all: the emerging standard for dexterous hands is a composite “tendon rope plus miniature ball screw” transmission⁵, where a small screw converts a hollow-cup motor's rotation into the finger-tendon pull. Ceramic variants, silicon-nitride balls rather than steel, cut heat and noise in the tightest finger packages, at a price: a single 2mm silicon-nitride ball costs about CNY18.87 today, with roughly 2,200 balls in a pair of hands². Hands are about 17 percent of a humanoid's bill of materials⁵, and this is the motion element inside them.

A ball screw is built into a hand for life, so the screws the fleet needs are a cumulative stock, every robot ever built, multiplied by about twelve, not a per-year flow. Integrating the consensus shipment ramp⁹, the global 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 ball screws respectively. The per-robot count is the single largest uncertainty on the demand side, and it is skewed upward. Twelve reflects the rising complexity of dexterous hands, roughly eight finger-drive screws at today's screw-versus-tendon split plus about four wrist and forearm joints that exist whichever way the fingers are driven, up from an earlier ten as hand degrees of freedom climb⁷. But dexterous hands are getting rapidly more complex: micro-screw hands are reported at 22 screw modules per hand, about 44 per robot pair², and Tesla's Gen3 hand moves to 22 degrees of freedom and 25 actuators per forearm⁷. If the fleet converges on that architecture, or on the composite “tendon rope plus miniature ball screw” hand in which a finger carries both⁵, the per-robot count could more than quadruple and this component moves from a 2037 warning to a near-term binder. In the chart below the black line is worldwide cumulative need, the grey band cumulative global production, and the red wedge the shortfall.

Production is the only thing that differs between the Global, China and US views. The worldwide fleet and the screws it needs are identical in all three; only the supply line changes, asking whether one bloc's factories alone could equip the whole global fleet. The producer base is broad and genuinely multi-sourced: more than twenty global manufacturers¹ spanning Japan (NSK, THK, Kuroda, JTEKT, Tsubaki Nakashima), Taiwan (Hiwin, PMI, TBI Motion), Germany (Bosch Rexroth, Schaeffler, Steinmeyer) and a fast-scaling Chinese cohort, a much healthier structure than the four-name premium niche that supplies roller screws. Global output runs from roughly ten million units a year today toward about 26 million by 2040, tracking the reported 5.7 percent market CAGR¹ with a modest lift as hand demand pulls the miniature tier. China holds around 30 percent of units and grows faster than the global curve, because Chinese makers have now broken the 1.8-millimetre micro-ball-screw barrier at about one third of THK's cost⁶, precisely the tier robot hands need. The United States runs a thin ~5 percent niche, Nook, Thomson, Barnes and Steinmeyer's US arm⁸, high-mix aerospace, defence and machine-tool work with no announced humanoid-grade expansion, which is why the US-only line falls behind as early as 2028. Every one of these unit figures is derived, not reported. The 2025 anchor converts a US$1.85 billion market at a blended ~US$185 per unit; at US$60 per unit the base would be three times larger and at US$250 nearly a third smaller, and the crossover year moves with it.

The durable crossover, the year cumulative production first drops below cumulative need and stays there⁹, lands in 2036 globally, now a few years before roller screws (2039) rather than after them, and near tendon drives and encoders. The shortfall then deepens: 49 percent of the 2040 need met, and cumulative production needing to scale about 2× to close the 2040 gap. For comparison, roller screws now scale about 1.2× and meet 83 percent of their 2040 need once demand is corrected for the rotary-heavy shipped fleet, so ball screws have become the deeper of the two linear-motion problems, as dexterous-hand screw counts rise while body roller-screw counts fall. The manufacturing evidence still tempers it: unlike roller screws, ball screws escape the precision-grinding pool. The single-bloc views remain the sharper story: China-only crosses in 2034, while US-only crosses in 2028. A humanoid industry that wanted to build hands domestically would hit this wall before almost any other.

This component is a correction to an earlier internal finding, and the correction is worth stating plainly. A prior component-gap sweep graded ball screws a high-severity bottleneck on the grounds that precision ball grinding “shares the same tight capacity pool as roller screws”, which would have made them a hidden second draw on the model's deepest chokepoint. Checked against the underlying evidence, that claim does not hold, on three counts. First, the cited source never says it: the ceramic ball-screw article it rests on discusses silicon-nitride powder purity, sintering yield and diamond wheels for ceramic machining², and does not mention roller screws or shared capacity anywhere. Second, the process is fundamentally different. Ball-screw threads can be cold-formed rather than ground, about three seconds versus about six minutes for the same profile³, and while ground screws have traditionally been needed for C0 to C5 accuracy, CNC thread rollers with chillers and dynamic variation control now reach C5 and even C3 in shorter strokes⁴, which is the accuracy and the stroke length a robot finger actually needs. A part with a hundred-fold-faster production path that meets spec is not competing for grinder hours. Third, the roller-screw bottleneck is specifically the thing ball screws avoid: what is scarce is deep internal thread grinding of planetary-roller-screw nuts, with high length-to-diameter ratios demanding small-diameter quills at 40,000 to 50,000 rpm, a machine class external ball-screw threading never touches. Ball screws are therefore deliberately not members of the shared precision-grinding pool, and adding them moved no published crossover: roller screws stay at 2033, harmonic drives at 2031, cross-roller bearings at 2035. What is genuinely constrained sits elsewhere and is modeled as this link's own limit: sub-millimetre micro-screw capability, and silicon-nitride ball sintering yield for the ceramic tier, where capacity is nascent and ramping². That is a young-supply-chain problem, not a shared-machine problem, which is why this link is graded elastic and still crosses over, purely on the arithmetic of ten screws times a very large fleet.