Encoders

Encoders are the mildest binder of the humanoid spine, and a subtle one, because the bind falls on a thin precision slice rather than the encoder market as a whole, and because the global bind itself is marginal. Of the roughly ninety-eight million rotary encoders the world builds each year, the overwhelming majority are commodity incremental and low-end industrial units¹; what a humanoid joint needs is the robot-grade precision absolute encoder, seventeen-to-twenty-three-bit servo grade, either a mixed-signal magnetic ASIC or a glass-optical scale², and that sub-tier runs at only about twenty million a year today, roughly half of all absolute encoders³. Worldwide output of the precision slice is modelled to climb from that twenty million toward three hundred million by 2040³, a high-end, low-confidence extrapolation, and on that path it keeps pace with the combined Physical AI fleet⁴ until around 2038 (the figures here track the whole worldwide fleet, not one country’s). After that the arithmetic tips: because every robot keeps its encoders for life, the units the fleet needs accumulate, and by 2040 a fleet of roughly forty-seven million humanoids needs about 1.98 billion precision encoders, about 2.13 billion once the cobot, industrial-arm and quadruped fleets are added (see below), against cumulative output of about 1.79 billion, roughly eighty-four percent met, the gentlest shortfall on the spine. That global slice is genuinely marginal: at the high end of the per-robot count (around fifty encoders) only about three-quarters of the 2040 need is met, and at roughly forty per robot it is near parity, so the global bind is sensitive to a single load-bearing assumption. The robust story is single-bloc self-sufficiency, not the global slice: building from one bloc binds regardless, China-only output falls behind around 2035, and US-only output, a high-value niche, around 2030, because each bloc’s precision-absolute output is small against the worldwide fleet either way.

A joint encoder reports the absolute angular position of a joint so the controller can close the loop and command the motor with sub-degree precision thousands of times a second. A humanoid carries roughly one precision absolute encoder on the output side of each actuator, about twenty-eight of them, plus a second, input-side sensor on each of the fourteen rotary precision joints that run dual encoding (an input sensor on the motor and an output sensor on the joint shaft)⁵, which puts the defensible per-robot count near forty-two (twenty-eight output-side plus dual-encoding on the fourteen rotary precision joints; field estimates run as high as the low fifties). Two technology paths serve this precision tier: glass-optical scales (Heidenhain, Renishaw, Nikon and Tamagawa⁶) give the highest resolution but are slow and capital-heavy to make, while mixed-signal magnetic ASICs (ams, Melexis, MPS) are cheaper, largely fabless, and scale with wafer capacity. Both, crucially, are the precision absolute sub-tier, not the commodity incremental encoder that dominates the headline market¹, which is why the binding question is about the precision slice and not total units.

Because a precision absolute encoder is built into a robot for life, the encoders the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by about forty-two⁵, not a per-year flow. Integrating the consensus shipment ramp⁴, the global installed fleet reaches about 585,000 robots by 2030, ten million by 2035 and forty-seven million by 2040, requiring roughly twenty-five million, four hundred twenty-four million and 1.98 billion precision encoders respectively. In the chart below, the black line is that worldwide cumulative need, the grey band is cumulative global production of the precision slice, 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 encoders 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 of the precision absolute slice follows an S-curve anchored at about twenty million units a year worldwide in 2025³; from there the figures are modelled extrapolation, the high end of a low-confidence band, since no published forecast carries the precision-absolute slice to 2040, climbing to roughly forty-five million by 2030 and one hundred twenty million by 2035 as new magnetic-ASIC and glass-grating lines are assumed to come on⁶, and reaching about three hundred million by 2040. China climbs steepest, from about four million units in 2025 to twenty-five million by 2030 and two hundred million by 2040 as humanoid assembly localizes, Chinese leaders already report better than ninety-percent core-component localization, and CT-Unite has unveiled the first gallium-nitride magnetic-encoder ASIC built for humanoid joints, due 2026⁷, with a domestic encoder magnet-ring supply chain spinning up behind it⁸. The United States holds a smaller, high-value position, strong encoder-ASIC and IC design (Broadcom) and precision houses (BEI, Dynapar), rising from about two million units in 2025 to fifteen million by 2040, with much of its wafer fabrication offshore⁹.

The crossover, the year cumulative production of the precision slice drops below cumulative need⁴, lands around 2038 globally for the combined Physical AI fleet (about 2039 for humanoids alone), the latest of any link on the spine. Before then, the capacity sprint banks enough output to cover the still-small fleet; after it, the durable need climbs as the fleet adds close to ten million robots a year, each humanoid demanding about forty-two precision encoders⁵, while precision output cannot quite follow. But the gap stays mild: roughly 340 million units short by 2040 against a 2.13-billion combined need, about eighty-four percent met, far gentler than the roller-screw or bearing walls. That said, the global bind is marginal and sensitive to the per-robot count: at the high end (around fifty encoders) only about three-quarters of the 2040 need is met, while at roughly forty per robot it is near parity, so the global shortfall could tighten or all but vanish on a single load-bearing assumption. The robust binds are the single-bloc ones, which hold regardless of that count because each bloc’s precision-absolute output is small against the worldwide fleet either way. Building from a single bloc brings the crossover forward and makes it hard: China-only output crosses around 2035, it expands fastest but starts from a small base⁷, and US-only output crosses around 2030, since a high-value niche of about fifteen million units a year⁹ never keeps up with the worldwide fleet.

Humanoids are not the only draw on the precision-absolute slice. By 2040 the collaborative-arm fleet reaches about 3.7 million units running dual encoders, about twelve per six-axis arm versus a humanoid’s forty-two, adding roughly 44 million; the ~6-million-strong industrial-arm fleet, with a single motor-side encoder per axis (~six per arm), another ~36 million; and the ~5-million-strong quadruped fleet, at about thirteen absolute encoders each (one per quasi-direct-drive joint, dual on premium legged units), a further ~67 million, the largest single non-humanoid encoder draw, because a quadruped carries roughly one encoder per joint on a fleet that is scaling fast. Together they lift the 1.98 billion humanoids need to about 2.13 billion. Because the encoder wall is the gentlest on the spine, they only nudge it: the global crossover moves from about 2039 to about 2038 and coverage from about ninety to about eighty-four percent. Drones add nothing here at all, their motor speed controllers are sensorless, using no absolute encoders, even though they are the largest fleet in the report by units, the sharpest reminder that raw volume does not touch this tier.

The ceiling is not set by the magnet or the sensor die alone but by two scarce capabilities stacked on top of each other. The high-volume tier, the mixed-signal magnetic-encoder ASIC, is gated by analog/mixed-signal wafer and packaging allocation¹⁰, the same constrained capacity every motor-control and position-sensing chip competes for. The top tier, glass-optical absolute encoders, needs precision glass-grating micromachining plus per-unit optical calibration², a slow, labour-intensive step that no amount of wafer capacity relieves. China is localizing the magnetic-ASIC tier fast, domestic GaN encoder ASICs and magnet-ring supply now stand up behind Chinese leaders’ reported ninety-percent-plus core-component localization⁷, but the top optical glass-grating tier stays import-reliant⁶. It is that two-layer constraint, mixed-signal ASIC throughput below, glass-grating micromachining and per-unit calibration above, that caps how fast the world can build precision absolute encoders, and it is why even the mildest binder of the spine still slips behind a durable, accumulating fleet.