Humanoid robotics has crossed from demonstration to deployment. InfraMosaic now tracks 122,302 cumulative units on order or already working across 26 distinct customer programs — Figure at BMW, Apptronik at Mercedes-Benz, Agility's Digit at GXO, with China's UBTech and AgiBot shipping by the thousand. Five of those programs are in live commercial deployment, six are firm orders, and four of 58 catalogued robot programs have reached mass production — a status that did not exist when the canonical 2025 forecasts were drawn — according to InfraMosaic's Deployment Tracker.
For a year and a half, the bull case for humanoids rested on a single chart that nobody could yet populate: the curve where pilots become fleets. Morgan Stanley's February 2025 Humanoid 100 framed it honestly as a forecast — humanoids would "move from demo to deployment over the decade," reaching tens of millions of US units and, on Musk's telling, more than a billion globally by the 2040s. What the paper could not do was count. In early 2025 there was, in effect, nothing on the floor to tally. That is what has changed.
The mechanism: orders, not announcements
InfraMosaic's tracker deliberately separates three things the hype cycle tends to blur — robots being paid to work, robots under firm order, and robots in supervised pilots. The 122,302-unit figure is the sum of disclosed unit estimates across all 26 programs, and its composition is the story. It is not 122,302 robots clocking in tomorrow; it is a backlog whose largest lines are forward commitments that have now been signed, priced and dated.
Two orders dominate the total. Figure AI's general-purpose backlog targets 100,000 units over four years — vendor-claimed, and flagged as such — while 1X Technologies has a disclosed 10,000-unit deal for its NEO home robot. Strip those two forward commitments out and the remainder — roughly 12,300 units — is what is genuinely shipping or imminently shipping right now: Unitree's 5,500+ units shipped in 2025, AgiBot's 5,100+, Tesla's ~1,000 internal Optimus target at Fremont, UBTech's 500-plus Walker S pre-orders into Foxconn, BYD and Geely, and a 200-robot Chinese eldercare cohort. The split matters: the demo-to-floor inflection is real, but it is still measured in thousands of working units against a backlog measured in six figures.
| Maker → Customer | Model | Region | Status | Units (est.) |
|---|---|---|---|---|
| Figure AI → backlog | Figure 02/03 | N. America | Order | 100,000 |
| 1X → undisclosed partner | NEO | N. America | Order | 10,000 |
| Unitree → mixed | G1 / H1 / R1 | Asia-China | Commercial | 5,500 |
| AgiBot → mixed | A2 | Asia-China | Commercial | 5,100 |
| Tesla → internal (Fremont) | Optimus | N. America | Pilot | 1,000 |
| UBTech → Foxconn/BYD/Geely | Walker S | Asia-China | Pre-order | 500 |
| Apptronik → Mercedes-Benz | Apollo | Europe | Pilot | — |
| Agility → GXO Logistics | Digit | N. America | Commercial | — |
| Figure AI → BMW (Spartanburg) | Figure 02 | N. America | Pilot | 2 |
Node by node: where the robots are landing
By status, the 26 programs break down as 12 supervised pilots, 6 firm orders, 5 commercial deployments and 3 pre-order books. By geography they cluster in two poles: 12 programs in North America and 10 in Asia-China, with three in Europe (the German automakers) and one in Korea (Boston Dynamics' Atlas into Hyundai). That bimodal map is the single most important pattern in the data. The West is deploying through marquee enterprise pilots — automotive assembly and warehouse logistics, two robots here, a small fleet there — while China is deploying through volume: Unitree and AgiBot are out-shipping the Western brands outright, and UBTech's Walker S2 has booked more than 800 million yuan of industrial orders.
The application mix is narrower than the marketing. Of the 26 programs, the overwhelming majority sit in just two verticals — automotive assembly and warehouse/manufacturing logistics — the structured, repetitive, high-wage environments where a humanoid's economics close first. Consumer (1X NEO), retail (Galbot, XPeng IRON) and care (eldercare and clinical trials) appear, but as pilots and pre-orders, not fleets. This is exactly the order an economist would predict: deployment follows the marginal value of the task, and the factory floor pays best.
The bottleneck: the body, not the brain
Here is the implication the unit count forces. The robots being deployed are not constrained by intelligence — the Brain layer has already been paid for by the AI trade. They are constrained by the body, and specifically by what they cost to build. InfraMosaic's bill-of-materials work, re-based on Morgan Stanley's own Exhibit 54, shows that sensors are the single largest cost line at 37% of an Optimus-class BOM — ahead of motors and screws — with six-axis force-torque sensors flagged at 440% of projected 2030 capacity. A 100,000-unit Figure order or a 10,000-unit NEO deal is a demand signal pointed straight at the tightest, most China-and-near-monopoly-exposed nodes in the chain.
That is why the deployment number is the leading indicator now, and why mentions and unveils are the lagging ones. The question for 2026 is no longer "who showed a robot" but "who can build the one they already sold." Total material cost still sits near $55,000 per unit and is modelled down to roughly $40,000 (Western) and $18,000 (China) by 2030 — and every signed order tightens the race down that curve.
What to watch
Three things will tell us whether the inflection accelerates or stalls. First, conversion: how much of the 110,000-unit forward backlog (Figure plus NEO) actually ships on schedule — vendor-claimed lines are tracked as exactly that until they post. Second, the China volume curve: Unitree and AgiBot already out-ship the West on count; if UBTech's S2 order book and XPeng's end-2026 mass-production target convert, the centre of gravity in deployed units moves decisively east. Third, the sensor bottleneck: watch force-torque capacity, because that is the node where a six-figure order book meets a physical supply ceiling. The base-case market path — 550,000 units by 2030 — assumes that ceiling lifts in time.
Verify every number
Rystad asks you to trust the trace. InfraMosaic lets you check it — every figure in this article resolves to a hash-chained record in the Publication Ledger, with its inputs, methodology and QA suite attached. The headline figure is on record:
kpi.deployments_total = 26 programs · content_hash 6b1b05f6eaa88f0a…
ledger chain_head 15a725e0…41c29d · as-of 2026-06-13 · 363 records
Composition of the 122,302-unit total: 100,000 (Figure backlog, vendor-claimed) + 10,000 (1X NEO order) + 5,500 (Unitree 2025 shipments) + 5,100 (AgiBot 2025 shipments) + 1,000 (Tesla Optimus internal target) + 500 (UBTech Walker S pre-orders) + 200 (China eldercare cohort) + 2 (Figure–BMW Spartanburg pilot). Programs without a disclosed unit estimate (small fleets, undisclosed networks) are counted in the 26-program total but contribute zero to the unit tally — the figure is therefore a conservative floor.