A Western reference humanoid robot cost roughly $200,000 to build in January 2024. By June 2026 the same machine costs about $94,600 — a fall of ~53% in 18 months. China's curve runs lower and parallel: ~$46,000 to ~$29,900, a 35% cut. On the index that is IHCI-West 47.3 and IHCI-China 65.0 (2024-01 = 100). Sensors — not actuators — lead a ~$55,000 bill at 37%, and they are where the next leg is won or lost — according to InfraMosaic's Humanoid Cost Index (IHCI).
For two years the loudest argument against humanoid robots was a price tag. Morgan Stanley's February 2025 build estimates put a Western machine near $200,000; at that level the labour-substitution math only worked in a handful of niches. The InfraMosaic Humanoid Cost Index (IHCI) was built to do one thing the point estimates could not: track the number monthly, on a consistent reference design, so the cost-down can be watched rather than asserted.
The track is unambiguous. Indexed to 100 at January 2024, the Western reference robot has dropped to 47.3 as of June 2026 — a 52.7% decline — while the Chinese reference, already cheap, has eased from 100 to 65.0, or 35%. In dollar terms the Western machine moved $200,000 → ~$94,600 and the Chinese machine $46,000 → ~$29,900. China still builds for roughly a 68% discount to the West, but the gap is the story's second act; the first is that both curves are bending down fast enough to change the adoption calendar.
Why the curve matters more than the level
Adoption of an expensive durable good is governed by where its cost crosses the value of the labour it displaces — and, crucially, by how fast that crossing approaches. Morgan Stanley's own framework sketched negligible humanoid adoption to about 2035, then an S-curve toward a ~$3 trillion US labour-substitution TAM by 2050. The shape of that S-curve is set by the cost-down rate. A robot that is merely cheap invites a pilot; a robot that is visibly halving every 18 months invites a capital plan.
The "too expensive" objection doesn't need the robot to be cheap today. It needs the buyer to believe it will be cheap on the depreciation schedule they're signing — and the IHCI is what makes that believable.
That is why InfraMosaic moves "cost" out of the bear column. The cost gap closed faster than the con-side assumed: a Western reference machine at ~$94,600 already clears the substitution threshold for structured, high-duty tasks, and the slope says the addressable task set widens every quarter. What stays in the con column is not price — it is reliability and autonomy in unstructured environments. The economics objection is being retired on schedule.
The cost-down curve: Western and China reference robots, monthly
Node by node: where the $55,000 sits
The IHCI is not a black box. Underneath it sits a reference bill of materials re-based directly on Morgan Stanley's Exhibit 54 (Optimus Gen-2, by component), and it carries a correction worth stating plainly: the headline cost driver is sensors, not joint modules. The original analysis repeatedly anchored on "joint drive modules = 60–70% of cost," yet its own exhibit shows sensors as the #1 line. InfraMosaic foregrounds the data — and locks it with a regression test so the correction cannot silently revert.
Sensors lead the bill of materials — at 37%, ahead of motors and screws
| Subsystem | Assessed cost | Share of BOM |
|---|---|---|
| Sensors (6-axis force-torque + IMU + encoders + vision + tactile) | $20,350 | 37% |
| Motors (frameless + BLDC) | $11,165 | 20% |
| Screws (planetary roller + ball) | $11,110 | 20% |
| Reducers (harmonic + cycloidal) | $6,930 | 13% |
| Encoders | $2,145 | 4% |
| Compute (SoC + chips + ISP) | $2,090 | 4% |
| Bearings, battery & structure | $1,210 | ~2% |
This matters for the curve's slope. Motors at $11,165 (20%) and screws at $11,110 (20%) are the classic China-scale story — and they are already deflating fast. The proven template is the planetary roller screw, which fell from roughly $3,000 to ~$800 as Tesla-scale linear-actuator volume arrived. Multiply that mechanism across 28 actuators, 14 screws and 14 reducers per robot, layer in Chinese supply for reducers ($6,930, 13%) and encoders ($2,145, 4%), and most of the Western cost-down to date is explained. The mechanical stack is doing what mechanical stacks do at volume.
The bottleneck — and the reason the curve could stall rather than glide — is the line that leads the BOM. The six-axis force-torque sensor embedded in the hands and feet is both the top cost and the tightest supply node InfraMosaic tracks: the order-book model flags force-torque sensors at 440% of projected 2030 capacity, behind only planetary roller screws at 770%. A component that is simultaneously the most expensive and the most supply-constrained is precisely the one that resists the cost-down. If sensors don't follow the screw template, IHCI-West flattens above its projected path; if they do, the curve continues toward the modelled ~$40,000 Western / ~$18,000 China endpoints by 2030.
Verify every number
Rystad asks you to trust the trace; InfraMosaic lets you check it — every figure here resolves to hash-chained evidence in the Publication Ledger. The cost-curve endpoints carry their own ledger figure-IDs; click through to the inputs, methodology, QA and content hash behind each one.
What to watch
- Force-torque sensors. The hero line and the hardest node: 37% of BOM, 440% of 2030 capacity. Whether it follows the roller-screw cost-down decides if IHCI-West stays on its projected glide path or flattens above it.
- The China discount. Holding near 68% today. The 2030 model has it compressing toward ~55% as Western scale catches up — watch whether each monthly print narrows or holds the gap.
- Rare-earth export controls. The magnet input behind the motors line is a named near-term hurdle. A supply shock there reverses part of the actuator cost-down regardless of factory scale.
- The 2028 inflection. Nothing in 2025–26 deployment data (122,302 cumulative units across 26 programs) argues for pulling the at-scale S-curve earlier — pilots are hundreds-to-thousands of units, not millions. The cost curve leads; the volume curve follows.
The bear case used to start with a price. As of June 2026 the price is roughly half what it was, the track is public and monthly, and the bottleneck has a name. That is the difference between an objection and a forecast — and it is the difference InfraMosaic was built to measure.
Methodology. The IHCI indexes a Western and a Chinese reference humanoid build to 100 at January 2024, tracked monthly against a bill of materials re-based on Morgan Stanley Humanoid-100 Exhibit 54 and triangulated with Goldman Sachs, Bain and observed factory-gate prices (Unitree, Beijing pilot lines). Pre-2026 points are HARD; the forward cost-curve to 2030 is ASSESSED. Material BOM totals ~$55,000; whole-robot figures include integration, assembly and margin. Every headline figure resolves to a hash-chained record in the Publication Ledger (build of 13 Jun 2026). Figures cited from the live dataset: IHCI-West 47.3 (~$94,583), IHCI-China 65.0 (~$29,917).