IMU / Magnetic Sensors

IMU and magnetic sensors are a non-binder, one of the clearest cases in this report where the world already makes vastly more than the worldwide humanoid fleet will ever need. The part is two commodity sensor families a humanoid carries: a MEMS inertial unit (a six-axis accelerometer-plus-gyroscope IMU that tells the robot which way is up¹) and a handful of magnetic-sensor ICs (Hall, TMR or AMR chips). A humanoid uses only about one and a half such units per robot¹, so even a worldwide fleet of forty-seven million robots by 2040 needs only about seventy million and a half of these sensors in total. Set that against an industry that ships billions of units every year: roughly three point eight billion MEMS inertial units alone in 2023², inside a total MEMS market of about thirty-one billion units a year³, plus multi-billion-unit magnetic-sensor output on top⁶. The entire cumulative 2040 fleet need is a few days of current world production. Run the four-series gap and cumulative output runs more than a thousand times the fleet’s cumulative need globally, so the global crossover year is null⁹. The same holds for every bloc: China-only output banks roughly two hundred times the 2040 need, and even the thin US-located slice banks about fifty times, so the China and US crossovers are null as well. The one honest caveat is concentration, not volume: the supplier base is Western and Japanese-led (Bosch, STMicroelectronics and TDK for the MEMS IMU³; Allegro, AKM and Infineon for magnetic⁶) and the true upstream node is Taiwan’s eight-inch foundries. But the base is so vast and so multi-sourced that no single-bloc view ever binds on supply.

This link bundles the two solid-state sensor families that give a humanoid its sense of motion and position. The inertial measurement unit (IMU) is a MEMS chip combining a three-axis accelerometer and a three-axis gyroscope, it tracks the robot’s body pose, balance and orientation, the way the balance organ in your inner ear does¹. The magnetic sensors are Hall-effect, TMR or AMR ICs that detect magnetic fields for contactless position and current sensing⁶. We model about 0.0015 units per robot in thousands, roughly one and a half sensors per robot¹: a humanoid runs essentially a single body IMU (the Figure 02 and Unitree designs each carry one six-axis IMU, magnetometer integrated), plus a small allowance for board-level magnetic ICs. Crucially, this count does not include the per-joint magnetic position sensors, those rotary Hall and encoder ICs belong to the encoder link, and folding them in here would double-count them. These are consumer- and automotive-grade parts: the IMUs robots use are distributor-stocked chips costing about two to ten dollars⁸, not the tactical-grade inertial units that cost orders of magnitude more. One real qualitative caveat: robot-grade high-precision, low-drift IMUs are a thinner sub-tier than the commodity phone IMU, but commodity MEMS volume is so large it swamps the robot need regardless.

Because an IMU or magnetic sensor is soldered into a robot for life, the sensors the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by about one and a half sensors each¹, not a per-year flow. Integrating the consensus shipment ramp⁹, the worldwide installed fleet reaches about five hundred and eighty-five thousand robots by 2030, ten million by 2035 and forty-seven million by 2040, requiring only roughly nine hundred thousand, fifteen million, and seventy million and a half of these sensors respectively. Hold those figures against an industry that ships billions of inertial and magnetic units every single year²: even the entire cumulative 2040 fleet need is a few days of current world output. In the chart below, the black line is that worldwide cumulative need and the grey band is cumulative global production, and the band sits so far above the line that no red shortfall wedge ever opens. That absence of a gap is the finding, not an error.

Production is the only thing that differs between the Global, China and US views. The worldwide humanoid fleet and the sensors it needs stay identical in all three; only the supply line changes, asking whether one bloc’s factories alone could equip the whole global fleet. Here every bloc wins that contest by orders of magnitude. Global combined output of MEMS inertial plus magnetic-sensor ICs runs on the order of five billion units a year today, anchored conservatively on the roughly three point eight billion MEMS inertial units shipped in 2023², within a thirty-one-billion-unit total MEMS market³, plus multi-billion-unit magnetic-sensor volume⁶, growing at about six percent a year toward perhaps eleven and a half billion by 2040, a pace between the roughly three-point-seven-percent all-MEMS unit growth rate⁴ and the faster eleven-point-nine-percent inertial revenue growth² (units grow slower than revenue as the mix shifts to higher-value automotive and robotic parts⁵). The post-2030 figures are a modeled extrapolation of this mature commodity base, not a reported unit forecast. China makes only about eight percent of global inertial-sensor production today, Greater China is roughly a tenth of all-MEMS vendor revenue, and IMUs are even more Western-concentrated³, so about four hundred million units a year now, modeled rising toward a fifteen-percent share by 2040 as China scales domestic sensor fabs. The US is a thin slice of genuinely US-located fab, about three percent: Analog Devices is a real US inertial-and-magnetic fab, joined by Honeywell’s tactical niche, while TDK InvenSense is US-headquartered but fabless (it builds at TSMC in Taiwan⁸) and the consumer volume sits with Bosch, ST and Murata³. Yet even that hundred-and-fifty-million-a-year US niche dwarfs the fleet need.

For every view, the answer is never. Cumulative global production stays more than a thousand times above the fleet’s cumulative need at every year through the 2040 horizon, so there is no global crossover⁹: by 2040 the world has produced something on the order of a hundred and twenty-eight billion of these sensors against a cumulative fleet need of about seventy million and a half², coverage of roughly eighteen hundred times. China-only never falls behind either, running on its own about two hundred times the 2040 need; and, as with steppers, but unlike almost every structural part in this report, even the thin US-only slice never crosses: a niche output banks roughly fifty times the cumulative 2040 fleet need on its own⁶. So all three crossover years are null. This is a link where the “build it from a single bloc” question has no sting on volume. The one thing worth recording is the direction of supply, not its scale: it is Western and Japanese-led, Bosch sits at number one with about two billion dollars of MEMS revenue, ST a distant sixth (and now acquiring NXP’s MEMS business⁵), TDK among the leaders³, and the magnetic side is led by Allegro, AKM, Infineon and Honeywell⁶. The real upstream chokepoint, to the extent one exists, is the Taiwan eight-inch foundry base every fabless sensor house relies on. But with the base this large and this multi-sourced, that is a concentration label, not a capacity constraint.

IMU is the one link where the mobile embodiments, not the arms, are the claimants: an arm bolted to the floor needs no inertial sensing, but anything that flies or walks lives or falls on it. Quadrupeds were the first, carrying one MEMS IMU per unit, about five million added to the humanoids’ ~70 million cumulative need by 2040. Drones then pile in hardest of all: flight control is IMU-critical (one to three per airframe), and drones are the largest fleet by units, so cumulative drone builds add on the order of 700 million IMUs by 2040, roughly ten times the entire humanoid draw, the single biggest volume on this link. And yet it still does not matter: the world turns out well over a hundred billion MEMS inertial sensors a year for phones and cars, so even ten-times-humanoid drone volume is a rounding error on a rounding error. This is the sharpest illustration in the whole report that volume is not what binds, the biggest fleet on the planet lands on a commodity link with a thousandfold headroom and moves the crossover not at all (there isn’t one). Note too that a MEMS IMU is silicon, not a magnet, so a drone’s inertial sensing draws essentially nothing on the NdFeB spine.

Most links in this report bind because some deep node, precision grinding, suspended-coil winding, heavy rare earths, cannot scale fast enough. IMU and magnetic sensors have no such node. They are commodity CMOS-plus-MEMS chips made by the billion on standard semiconductor lines, none of it grinding-gated, so they are deliberately not placed in the shared precision-grinding pool that throttles roller screws, harmonic flexsplines and cross-roller raceways, their effective output equals their standalone output, with no pool throttle. Bosch Sensortec alone ships more than a billion intelligent sensors a year⁷; AKM has shipped over fifty billion Hall sensors in its history⁶; the total MEMS industry turns out about thirty-one billion units a year and is forecast to keep growing³. Stack that against a humanoid pull of about one and a half sensors per robot¹ and the conclusion is unavoidable: there is no binding input to map. The two caveats are real but secondary. First, concentration: the supplier base is Western and Japanese-led and the upstream foundry node is Taiwan-centric, a resilience and geopolitics question, not a volume one, because the parts are second-sourced across many vendors and fabs. Second, grade: robot-grade low-drift IMUs are a thinner tier than the cheapest phone parts, so a humanoid maker pays up for precision, but the commodity MEMS base is so deep that even a demanding sub-tier has ample headroom. On the question this report exists to answer, can the world build enough, IMU and magnetic sensors are among the clearest yeses in the set.