Humanoid robot bill of materials: where the money actually goes
A humanoid's cost is not evenly distributed. Actuators alone consume 40-60% of the hardware BOM - three to five times the next largest block. Which means: of all the cost-reduction moves available, only the ones that land on actuators move the total. Everything else can be squeezed to zero and the price barely shifts.
A humanoid robot is a collection of joints, not a machine. Around 25-30 joints per unit, and every joint is a self-contained cost centre - motor, reducer, encoder and driver, bought and priced as one unit. So total cost is joint count multiplied by cost per joint, not the price of any single hero component. There are only two levers that matter: fewer joints, or cheaper joints. B
On this page
Last verified: 2026-10-09. Share ranges are from McKinsey's 17 April 2026 supply-chain analysis. They are directional estimates and move considerably with architecture and maturity.
1. Five domains: actuation is the only large one
Split a humanoid's hardware BOM into five domains and the shares fall out like this. The five together account for 85-90% of unit cost; the remainder is cooling, harnesses and similar. B
| Domain | Share of hardware BOM | What it buys |
|---|---|---|
| Actuation | 40-60% | Every joint module in legs, torso, arms, neck and hands |
| Sensing and perception | 10-20% | Cameras, IMU, force/torque, tactile, encoders |
| Compute and control | 10-15% | Edge AI SoCs, real-time controllers, connectivity |
| Structural components | 5-10% | Frame, housings, linkages |
| Battery modules | 5-10% | Cells, BMS, thermal management |
Basis: hardware BOM only. Excludes R&D amortisation, software, integration and after-sales. Shares are directional ranges. B
Hands are often broken out as their own line, but their actuators and sensors are usually already counted inside actuation and sensing above - so listing them separately double-counts. If "hands" is a separate line in your BOM, check first whether the joint modules inside it are being counted twice. This is the single most common inconsistency between supplier quotations.
2. The second-level split inside an actuator
An actuator is itself a small machine. Open it up and the gearbox is the largest piece at 30-50% of a single actuator's cost - which is the most important sentence on this page: inside the largest cost block of the whole robot, half is the reducer. B
| Inside one actuator | Share of that actuator's cost |
|---|---|
| Gearbox | 30-50% |
| Driver electronics | 15-20% |
| Motor | 10-20% |
| Mechanical components | 10-20% |
| Sensors | 5-10% |
Basis: a representative integrated actuator, not a specific model. B
Run the multiplication: actuation is 40-60% of the machine, the gearbox is 30-50% of an actuator, so gearboxes alone land somewhere around 12-30% of total hardware BOM. That derivation is ours, not a figure published in the source - treat it as an order of magnitude, not a number. C
3. Absolute BOM range per unit
Shares are only half the picture. Current hardware BOM runs roughly $30,000 to $150,000 per unit, and the long-term target widely cited across the industry is under $20,000 - which implies substantial compression is still required. B
The range is that wide because it covers two different kinds of machine at once: validation units assembled from off-the-shelf industrial servos, and machines redesigned around integrated joints for volume production. When you get a quote, ask which end of the range it sits at, and why. Treating a prototype BOM as a production cost is the most common misread in this cycle. C
4. Why this structure resists change
The rigidity is physical, not a supply-chain maturity problem. Bipedal locomotion needs continuous dynamic balancing and torque control; every degree of freedom needs a joint that can produce force on its own. Joint count is set by the configuration and cannot be squeezed. Even with a fully mature supply chain, actuation stays the dominant block - maturity lowers unit prices, it does not change shares. C
The second structural constraint is a mismatch: where the value concentrates is where the supplier ecosystem is least ready. Actuation is both the largest cost block and the least mature supply base, while compute and batteries ride on electric-vehicle and consumer-electronics infrastructure that already runs at scale. Put plainly: the parts that are easiest to make cheaper were never the expensive ones. B
5. The three places cost can actually move
- Gearbox selection - the largest piece of the largest cost block, and the segment where domestic substitution is advancing fastest. See should you substitute a domestic harmonic drive.
- Joint count and configuration - removing one degree of freedom removes the motor, reducer, encoder and driver along with it. This is the only lever with a multiplier effect, but it lives in the design phase and cannot be pulled at the purchasing stage.
- Buying integrated joint modules versus components - the choice decides whether you are paying for hardware or for integration labour. Quotations in the two forms are not directly comparable.
There is no "standard BOM table for a humanoid robot". Such a table does not exist - joint count, payload, dexterity, duty cycle and maturity all move the mix substantially. Any calculator handing you fixed proportions has already assumed an architecture on your behalf. What this page gives you is magnitude for judgement, not a list for quotation.
6. Sources and evidence grades
Grades: A official primary · B authoritative secondary, traced to the origin · C our own derivation, not citable as fact.
- McKinsey & Company, Turning humanoid supply chain constraints into billion-dollar wins, 17 April 2026 - five-domain shares (actuation 40-60%, sensing 10-20%, compute and control 10-15%, structural 5-10%, battery 5-10%); the five domains at 85-90% of unit cost; BOM of $30,000-$150,000 per unit; a long-term target below $20,000; and the intra-actuator split (gearbox 30-50%, driver 15-20%, motor 10-20%, mechanical 10-20%, sensors 5-10%). B The source presents these as directional estimates and states explicitly that robot size, payload, dexterity, duty cycle and maturity can move the mix considerably.
- Joint count of 25-30. C Synthesised by us from configuration estimates in several secondary analyses (7-8 per arm, 6-7 per leg, 2-3 for torso and neck). Variation across designs is large; use it as an order of magnitude only.
- "Gearboxes at roughly 12-30% of total BOM" is our multiplication of the two ranges above. C Multiplying two ranges compounds the error - do not quote it as a precise figure.
- The claims that cost rigidity is physical, and that value concentration mismatches supply maturity. B The latter restates the McKinsey conclusion; the former is our own inference. C
Related
Last updated 2026-10-09. Revised when the underlying McKinsey analysis changes or a new edition is published.