Vision Sensors (CMOS cameras)

Vision sensors are one of the clearest “the world can build enough” links in this report, and the four-series model says so without hedging: production never falls behind anywhere. A humanoid carries only about four and a half cameras, a fleet average across the OEMs; Tesla’s Optimus runs about three, and the eight-camera figure some specs cite is a Full-Self-Driving carryover rather than the robot’s own count¹, so even a worldwide fleet of about forty-seven million robots by 2040 needs only on the order of two hundred and twelve million image sensors in total¹⁰. The CMOS image-sensor (CIS) industry already makes that many roughly every fortnight: global output runs on the order of seven billion sensors a year, growing from about six and four-fifths billion units in 2023 toward eight and a half billion by 2029², with smartphones alone accounting for more than sixty percent of the market⁴, some four and four-tenths billion smartphone sensors in 2024³. Run the gap and cumulative production banks hundreds of times the fleet’s cumulative need in every view: the global, China and US crossover years are all null¹⁰. China is genuinely strong here by unit volume, Shanghai’s GalaxyCore is the world’s number-two image-sensor maker by shipments, alongside SmartSens and the China-owned OmniVision³, so even the single-bloc “build it from China alone” view stays comfortably elastic, and even the thin US fab niche around onsemi never crosses⁷. The one honest nuance is that high-end global-shutter and depth sensors are a thinner premium tier, but the commodity RGB CIS base swamps the robot need many times over. There is no shortfall here to worry about.

A vision sensor is a CMOS image sensor (CIS), the silicon chip that converts light into a digital image, packaged into a camera module with a lens and interface. It is the same commodity device that sits in every smartphone, car and security camera, which is exactly why it is so abundant: smartphones are more than sixty percent of the CIS market on their own⁴. A humanoid uses a handful of them for head stereo vision, body and hand cameras and peripheral coverage; Tesla’s Optimus runs about three cameras and deliberately uses no LiDAR, on the argument that cameras are roughly a hundred times cheaper¹ (early spec sheets cited eight, but that was a carryover from Tesla’s Full-Self-Driving camera suite, not the robot’s own count). Most humanoid designs run two to six RGB cameras plus one or two depth sensors, so the model uses about four and a half cameras per robot as a fleet average¹, which keeps the demand math honest rather than letting the old eight-camera anchor overstate the pull. The only part of vision that is even arguably specialized is the high-end global-shutter or depth (time-of-flight, structured-light) sensor used for precise motion and grasping; that is a thinner premium tier, but it rides on the same vast wafer base as the commodity RGB sensors.

Because a camera fitted to a robot stays in it for life, the sensors the fleet needs are a cumulative stock, the total number of robots ever built worldwide multiplied by about four and a half cameras each¹, not a per-year flow. Integrating the consensus shipment ramp¹⁰, 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 two and a half million, forty-five million, and two hundred and twelve million image sensors respectively. Hold those figures against an industry that ships around seven billion sensors every single year²: even the entire cumulative 2040 fleet need is well under a single month 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.

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 CIS output runs about seven billion sensors a year today, growing from roughly six and four-fifths billion units in 2023 toward eight and a half billion by 2029 at about a four-and-seven-tenths-percent unit growth rate², and the model flattens it toward about ten and four-fifths billion by 2040 as smartphone volume saturates, smartphone sensor shipments grew only two percent in 2024³, while automotive and security add slower unit growth. The market is led by Sony at roughly forty-five to fifty percent⁸, then Samsung, GalaxyCore, OmniVision at about eleven percent, SmartSens and onsemi, with the top five companies making up about eighty percent of the market⁵; revenue is heading past thirty billion dollars by 2030⁹. China is strong here by unit volume, not just the roughly nineteen percent revenue share⁴: GalaxyCore is the world’s number-two image-sensor maker by shipments and grew thirty-four percent in 2024, SmartSens grew about forty percent in the first half of 2025, and OmniVision is China-owned³, so we put China-affiliated output around two billion sensors a year today and rising toward four and a half billion by 2040, capped only at the sub-micron high end by lithography restrictions, not in the high-volume tier. The US is a thin niche around onsemi’s East Fishkill fab in New York, the only secure, federally-qualified CMOS image-sensor fab in the country, auto-led, which began in-house sensor production in 2024⁷, sized at perhaps two hundred million sensors a year rising to about four hundred and fifty million, anchored on the automotive image-sensor base that is itself growing from about three hundred and eighty million units in 2023 toward four hundred and eighty million by 2028⁵. The post-2030 figures are a modeled extrapolation of the mature CIS base, not a reported unit forecast.

For every view, the answer is never. Cumulative global production stays hundreds of times above the fleet’s cumulative need at every year through the 2040 horizon, so there is no global crossover¹⁰; by 2040 the world has produced well over a hundred billion image sensors against a cumulative fleet need of about two hundred and twelve million², coverage of several hundred times. China-only never falls behind either: on its own, at around two billion-plus sensors a year, it banks roughly two hundred times the cumulative 2040 need, because China is a genuine volume leader in this category rather than a marginal entrant³. And, unusually for this report, even the thin US-only niche never crosses: onsemi’s few-hundred-million-a-year auto-led output banks something like twenty times the cumulative 2040 fleet need entirely on its own⁷. So all three crossover years are null. This is one of the links where the “build it from a single bloc” question has no sting at all, not because of any policy subtlety, but because the part is a high-volume commodity made in enormous quantity on three continents. The only genuine qualification is qualitative: high-end global-shutter and depth sensors are a thinner, more concentrated premium tier than the commodity RGB base, so a robot design leaning heavily on specialized depth imaging would draw on a smaller pool, but that tier still rides the same wafer fabs, and the commodity base outruns the total robot need many times over. The other embodiments only reinforce this: a quadruped carries about two cameras, so even a quadruped fleet of about 5.15 million by 2040 adds only some ten million sensors to an industry shipping billions a year, a commodity draw that never binds. Drones draw a little more per unit, about three cameras each for first-person-view, gimbal and obstacle sensing, but even the largest fleet in the report by units, a cumulative roughly four hundred and seventy-eight million drones by 2040 (units built, since about half are single-use military craft), adds only about one and four-tenths billion sensors, still tiny against an industry shipping more than seven billion CMOS image sensors a year.

Most links in this report bind because some deep node, precision grinding, suspended-coil winding, heavy rare earths, cannot scale fast enough. Vision has no such node. A CMOS image sensor is fabricated on mature and specialty wafer nodes that are already enormously scaled for phones and cars, and the supply base carries visible slack: global CIS wafer capacity is on the order of six hundred and thirty-eight thousand wafers a month running at only about seventy-two percent loading, forecast to meet demand comfortably through 2030⁶. Vision is therefore deliberately not placed in the shared precision-grinding pool that throttles roller screws, harmonic flexsplines and cross-roller raceways, its bottleneck is CIS wafer fab plus camera-module assembly, neither grinding-gated, so its effective output equals its standalone output, with no pool throttle. Stack a multi-billion-unit-a-year base² with idle wafer capacity against a humanoid pull of about five and a half cameras per robot¹ and the conclusion is unavoidable: there is no binding input to map. If anything ever tightened the picture it would be the narrow high-end depth-sensor tier rather than commodity imaging, or a demand-side surprise, some robot generation standardizing on far more cameras than the current bill of materials suggests, but even a large multiple of today’s five-and-a-half count would still vanish against seven billion sensors a year. The honest record is that production is broadly distributed with China a true volume leader³; on the question this report exists to answer, can the world build enough, vision is among the clearest yes answers in the set.