Inside India's Humanoid Actuator Supply Chain : Strong at the Edges but Empty at the Core

Quick summary. Every humanoid robot carries forty or more actuators, and the actuator is where the money sits: reducers, screws and magnets alone are roughly a third of the bill of materials. India's factories are excellent at the actuator's periphery, housings, gears, windings, industrial bearings, controllers, and absent from its precision core, where no Indian company makes harmonic reducers, encoders, torque sensors or sintered NdFeB magnets. Even a modest 10,000 robots a year would need more harmonic reducers than the entire world bought in 2025. The realistic near-term prize is not building humanoids. It is owning as much of the precision-manufacturing stack underneath them as India can convert.

What an actuator is, and why it matters. An actuator is any device that converts an electrical command into controlled physical motion, and in a humanoid it does the job a muscle does in a body. Every joint the robot has, the shoulder, elbow, wrist, hip, knee, ankle and gripper, is driven by one, which is why a machine with forty or more joints carries forty or more actuators. The task sounds simple: move a limb to a commanded position, hold a load, absorb a shock, and repeat it millions of times without drifting out of tolerance. The difficulty is that each actuator must combine the torque density of an electric motor, the positioning precision of a gearbox measured in arc-seconds, and the feedback of sensors that know where the joint is at every instant, all packed into a volume smaller than a human thigh. That combination is why actuators are among the largest slices of a humanoid's bill of materials, and why whoever masters them effectively owns both the robot's cost curve and its performance ceiling.

The humanoid robot industry is leaving the laboratory. China is projected to ship around 50,000 units in 2026, and the component that decides whether any of them walk well is the actuator, the motor-plus-gearbox-plus-sensor package that serves as each joint. A humanoid carries more than forty of them. Each one is a dense bundle of engineering: a frameless torque motor, a harmonic or planetary reducer, two absolute encoders accurate to a tenth of an arc-second, precision bearings, in the linear joints a planetary roller screw, a torque sensor, a servo drive. Get any one of these wrong and the joint runs hot, noisy or short-lived.

That bundle is where the industry's value and its bottlenecks live. Reducers and roller screws together account for roughly a third of a humanoid's bill of materials; torque sensors and encoders are among the largest remaining slices. Whoever masters the actuator owns the robot's cost curve. The question for India, where the auto component industry has spotted its next decade of growth, is how much of that bundle the country can actually make.

The global supply chain is narrow, Chinese at the base, and qualification-bound

The map of who makes what is remarkably concentrated. Harmonic and strain-wave reducers trace back to Japan's Harmonic Drive Systems, challenged at the commodity end by China's Leaderdrive, Shuanghuan and Laifual. Roller screws come from a handful of specialists: Rollvis, GSA, Ewellix, Rexroth, Nanjing. Rare-earth magnets are a Chinese near-monopoly, roughly 94% of sintered output, resting on 87-91% of global rare-earth processing. Encoders sit with Heidenhain and Renishaw; the dominant encoder chip is made in Austria. The precision grinding machines that cut gear teeth and screw threads are built by Klingelnberg, Nidec, Gleason and Höfler, and the metrology stack that inspects their work is ZEISS and Mitutoyo.

Two facts define the global bottleneck. First, apparent capacity is looser than it looks. Laifual, the Chinese harmonic reducer maker, doubled shipments to 239,500 units in the first half of 2026, grew revenue 80% to RMB 142.2 million, and still posted an adjusted net loss, the result of a 30-40% price war in commodity-grade units even as it plans to add 800,000 units a year of capacity. Price is set at the unqualified margin; availability is set at the qualified one. Second, qualified supply is genuinely scarce. Harmonic drives quote lead times around 26 weeks, roller screws six to nine months, and the qualification currency is 50,000-hour reliability data that takes years to accumulate. One modelling exercise puts 2030 demand for planetary roller screws at 770% of assessed capacity, rotary actuators at 385% and six-axis force-torque sensors at 440%. Even at the extreme of the chain, OEMs are redesigning joints around ball screws largely because roller screws are too scarce and too costly.

This is the market India wants to enter. The entry ticket is not a factory. It is a qualification.

What India actually has: a broad process base with a hollow core

Judge India's position honestly and a clean split emerges between process and product. The country has deep, filing-evidenced capability in the processes an actuator needs: five-axis machining (Sansera machines everything from 5-axis parts to 4-metre components in titanium and Inconel for ISRO and HAL; Craftsman runs 14 plants), gear cutting and heat treatment at automotive scale, industrial bearings, copper winding wire, motor laminations and electronics assembly. What it does not have, across every disclosed record, is the product core: not one Indian maker of a harmonic or RV reducer, not one encoder manufacturer, not one robot-joint torque sensor, no cross-roller or thin-section bearing, and no operating sintered NdFeB magnet plant.

The distinction matters because an industrial product is not a humanoid product by default. Elecon Engineering machines gearboxes to micron-level tolerances, but its units are megawatt-class industrial drives, not 20-arcsecond miniaturised joint reducers. Schaeffler's Savli plant makes angular-contact ball bearings in India, the only verified Indian production of that type, but the cross-roller bearings a rotary joint needs are a different discipline. Amber's subsidiary PICL makes BLDC motors by the million, for air conditioners and washing machines, an appliance product nowhere near robot-grade torque density and feedback.

Where India stands across the actuator stack

The hidden chain, layer by layer

Three layers deserve the detail, because they carry the story.

Magnets are the deepest gap and the most-discussed one. A humanoid carries 2 to 4 kg of NdFeB, one and a half to two times an electric vehicle, to a higher quality bar that requires grain-boundary diffusion capability, and global humanoid magnet demand is projected to cross 5,000 tonnes by 2030, roughly a hundred times the 2025 level. India imports about 95% of its rare-earth magnets, and the 2025 Chinese licence curbs showed what that means in practice: Maruti's April-September e-Vitara production target fell from 26,500 units to 8,200. The response is real but early. The government's ₹7,280 crore scheme targets 6,000 tonnes a year of integrated magnet capacity. NAN GreenMet's ₹1,250 crore first phase targets 1,200 tonnes, scaling to 10,000, with commercial production from Q1 2028 at the earliest. Permanent Magnets holds approvals and a phased roadmap to 5,000 tonnes by FY31, but its magnet assembly line ran on fully imported components, and grain-boundary diffusion is disclosed by no Indian project. Midwest reports 500 tonnes of capacity in Hyderabad with 5,000 planned, while its own FY26 annual report says industrial-scale alloy and magnet capacity remains absent. Announced capacity is not installed capacity, and installed capacity is not robot-qualified capacity.

Motors are the strongest near-core story. CG Power builds LV motors to 710 kW and MV motors to 20 MW, with low-voltage drives that are, by its own description, almost entirely indigenous. Bharat Bijlee makes synchronous AC servo motors in-house, the only verified Indian servo maker. Pitti Engineering makes the stator and rotor cores and die-cast rotors inside other people's motors for ABB, CG and Siemens, and is candid about it: the company makes the components, not the motor. KSH International derives 75.3% of revenue from magnet winding wires and is adding capacity; Ram Ratna runs at about 76% utilisation developing EV hub-motor wire. The missing product is the frameless torque motor itself, which exists only as a development programme at Sona Comstar.

Reducers are the gap that defines the chain. India's gear industry is genuinely impressive: Carraro hobbed, shaved and ground 1.68 million gears in FY25; Hi-Tech Gears runs a full carburising, nitriding and induction suite; Shanthi Gears was the first Indian gear maker with aerospace-grade AS-9100D certification; Bharat Gears pioneered bevel-gear grinding for automotive. Not one of them discloses a harmonic, cycloidal or RV reducer programme. The strain-wave reducer's flexspline, a thin-wall forged cup ground to backlash tolerances of twenty arc-seconds, is a metallurgy and grinding problem that took Japan three decades and China a decade of technology transfer to solve. India has not started.

Made in India, assembled in India, imported

The honest reading of that table is that India would today assemble a humanoid actuator rather than manufacture one. The layers where India captures real economic value are the housings, the windings, the laminations, the industrial bearings and the electronics assembly, meaningful work, but the commodity half of the BOM. The value-dense half, reducers, screws, sensors, magnets, is imported, and so is the equipment that would make it.

The machine-tool layer underneath, and the queues that define it

The most under-appreciated finding is that India's component gaps repeat one layer down in the equipment. RACL Geartech's management has said plainly that gear-grinding machines are not manufactured by Indian makers, with order queues of one to two years. Macpower imports a 7-axis spindle grinder from Switzerland, describing it as something India is not producing. Wendt (India) makes surface, cylindrical and tool grinders, but its range stops short of gear and thread grinding. Jyoti CNC, the strongest Indian machine-tool builder, with installed capacity of 6,000 machines a year heading to 10,000 and an order book of ₹4,848 crore, still buys roughly 30% of its own content, controllers, drives, motors, sensors, from Germany and Japan. Batliboi imports its ball screws and guideways from Taiwan, as, it says, everybody does.

The numbers frame the dependency. India imported machine tools worth over ₹20,000 crore in FY2025-26, 60-65% of domestic consumption of ₹31,780 crore, with local production at ₹16,478 crore. China, Japan and Germany supplied 60% of imports in FY24. A humanoid actuator is a ground product; its flexsplines, screws and gears exist because of precision grinding. India localised the machining half of its equipment base and never localised the grinding half. If actuator production scaled in India tomorrow, its components would be cut on imported machines, ground on imported machines, and inspected on imported machines.

The companies that would have to build it

Tier A: already making actuator-relevant components.

Tier B: adjacent capability, no disclosed robotics product. Amber Enterprises (appliance BLDC motors through PICL), Avalon Technologies (motor drives and a delivery robot system in its FY26 report), Syrma SGS and Cyient DLM (industrial automation electronics), Siemens India (automation at scale, with management noting local PLC production would need substantial domestic volume), Igarashi Motors (small traction motors, a 3.3 kW unit developed and a 7.5 kW in progress), Sintercom (SMC stator parts, first EV-designed motor order won), RIR Power Electronics (₹618 crore SiC device project), and Timken India (35% of FY26 sales imported, which is both a dependence and a headroom).

Tier C: no Indian manufacturer disclosed at all. Harmonic and RV reducers, flexsplines, encoders, robot-joint torque sensors, cross-roller and thin-section bearings, sintered NdFeB with grain-boundary diffusion, gear- and thread-grinding machines, and robot-grade FOC servo drives. These categories also happen to hold the most concentrated value in the BOM.

What scaling would actually demand

Take a reference robot with 40 actuators, split 28 rotary and 12 linear, one motor, two encoders, a torque sensor and two precision bearings per joint, 3 kg of magnets and about 1,300 ICs per robot. The arithmetic then runs as follows.

Magnet demand by scenario, against India's announced capacity anchors (tonnes of NdFeB per year): 10,000 robots/yr, 30 t; 100,000 robots/yr, 300 t; NAN GreenMet's initial line, 1,200 t; 1 million robots/yr, 3,000 t; the national scheme's 6,000 MTPA target, 6,000 t. At one million robots a year, humanoid magnets alone would consume half the national scheme's 6,000-tonne target, and every tonne needs sintering capability no Indian plant has yet commissioned.

At 10,000 robots a year, less than a quarter of China's projected 2026 shipments, the housings, windings, laminations, bearings and controllers could scale on existing capability. Every harmonic reducer, encoder, torque sensor, magnet and grinding machine is imported or unbuilt, and the 140,000 harmonic units alone exceed the entire global humanoid volume of 2025. At 100,000, reducer demand of 2.8 million units across types exceeds the announced capacity of any single Chinese supplier, meaning India would need a domestic reducer industry of near-Leaderdrive scale, created from zero and qualified, while magnet demand of 300 tonnes equals a quarter of NAN GreenMet's first line. At one million, 14 million harmonic units is 163 times the 2025 global volume, larger than China's entire announced expansion, and 3,000 tonnes of magnets consume half the national scheme target while competing for feedstock against every other national programme on earth.

At every scale, the binding constraint is the same and it is not tonnage. It is the qualification ladder: an OEM to qualify against, since every cycle in this chain runs 30 to 48 months per component by the Indian benchmark Azad Engineering has disclosed, the grinding machines that must be ordered one to two years in advance, and the know-how recipes that no Indian filing yet shows anyone holding.

What India could realistically localise by 2030

Sort the landscape by how firm each layer is. Operating now: the machining base, gear unit-processes at automotive grade, industrial bearings with high localisation, motor components, winding wire, BLDC controllers. Under construction: OSAT packaging lines at Kaynes Sanand and CG Semi, where the G1 line went live in July 2026 and a second line of roughly 14.5 million chips a day is expected by end-2026; KSH's phase-two wire capacity; MTAR's push to complete screw certification. Announced, not built: the magnet pipeline, NAN GreenMet from Q1 2028 at the earliest, Permanent Magnets to 5,000 tonnes by FY31, Midwest's expansion, GMDC's planned 12,000 tonnes of rare-earth oxide separation, the ₹7,280 crore scheme itself. Early-stage or absent: grain-boundary diffusion, sintering at scale, harmonic reducers, encoders, torque sensors, frameless motor production, and any Indian gear-grinding machine.

By 2030, a realistic India localises the actuator's commodity periphery almost entirely, assembles the electronics layer from an OSAT base it is building now, supplies screws, gears, bearings and windings into global actuator supply chains, and, if the magnet plants commission and climb the certification curve, covers a meaningful share of its own magnet demand. It does not, on today's evidence, make its own harmonic reducers, encoders or robot-qualified magnets. Timing is unforgiving: an Indian reducer programme that won a design-in today could not ship qualified volume before roughly 2028-2030, the same window in which the demand surge is projected to strain global qualified capacity by a factor of four to eight.

The real question for India

The documented record points one way. India's verified strengths sit in the process layers; its verified absences sit in the product core and the qualification layer. India has the factories. It has none of the qualifications, and qualification, not capacity, is the binding constraint from 10,000 robots to one million.

That said, the two paths are not separable, and the honest framing is conditional. Every qualification cycle needs a customer to qualify against, and India's only humanoid OEM link today is an MoU between Sona Comstar and Germany's NEURA Robotics covering joint development and the industrialisation of robots for India and agreed markets. If a humanoid OEM ultimately builds in India, the ecosystem opportunity becomes self-reinforcing, and the component layers would convert far faster than they ever would serving exports from a standing start. If it does not, India's opportunity is as a precision component supplier to the global actuator industry, which is a real business, but a different and smaller one than the robot narrative implies.

The highest-leverage signal to watch is unglamorous: an Indian precision-grinding machine programme. Machines with one-to-two-year queues must be procured before qualification can even begin, which makes the equipment layer the slowest input in the stack and the one that decides everything above it. India localised the machining half. The grinding half is the country's move to make.

What to watch

• Sona Comstar's development programmes crossing into named qualifications or a first robotics revenue line, and whether the NEURA MoU acquires volumes and timelines.
• MTAR's screw certification completing, and any disclosure of screw-specific capacity or a dedicated grinding line.
• Magnet projects commissioning: NAN GreenMet's Q1 2028 target, Permanent Magnets' powder-to-block phase, and whether any Indian project discloses grain-boundary diffusion.
• Any Indian machine-tool builder entering gear or thread grinding.
• Kaynes and CG Semi OSAT ramps against their FY27 targets, the substrate for a future robot-grade drive.

Data note. Company figures are from filings, investor presentations and earnings calls through the Q1 FY2026-27 season; market and capacity data as of late September 2026. Global component volumes, lead times and concentration figures reflect industry disclosures and third-party estimates as reported. Scenario requirements are derived arithmetic from the stated component assumptions and should be read as planning estimates, not forecasts. Announced capacities are distinguished from operating capacity throughout.