The case for a humanoid robot is that the world is already shaped for humans: a machine in human form could work in an unmodified factory rather than requiring the factory to be rebuilt around it. The case against is arithmetic. As of 2026 Reuters reports humanoids completing tasks at about 20% of human output, and practitioners note that conventional robotics already does the repetitive work better and cheaper.
The most instructive datapoint in this collection is the most detailed one. Figure reported in November 2025 that a single Figure 02 robot completed an eleven-month pilot at BMW Spartanburg: over 1,250 hours of runtime across ten-hour weekday shifts, more than 90,000 sheet-metal parts loaded into welding fixtures. Read the task, not the hours. The robot performed one pick-and-place operation — loading parts onto a fixture so that a conventional six-axis robot could weld them — a single repetitive station inside a line otherwise run by ordinary automation. That is a real result and it is also almost the narrowest possible version of the promise.
Nobody agrees how many exist. For 2025, TechCrunch reports 13,317 units shipped globally; Reuters reports roughly 20,000, with Chinese manufacturers at 95% of them. Neither states its counting basis, and this collection publishes both rather than picking. Against that base of roughly thirteen to twenty thousand, Bank of America projects 1.2 million units shipped annually by 2030, and a 2.6 million projection for 2035 has been reported. The gap between a five-figure present and a seven-figure forecast is the single most important thing to hold in mind when reading anything else about this industry.
The binding constraint is not hardware. Reuters reports humanoids being trained by human demonstration at roughly one usable movement per 300 attempts for a novice trainer, and one per 50 for an experienced one. At those yields, training throughput rather than build cost is what limits deployment — which is why falling prices, real as they are, do not straightforwardly translate into more working robots.
Prices are falling on supply chain rather than breakthrough: the Kuavo robot fell 26% year-on-year to about $46,000, and demonstration-robot rentals fell from roughly $1,500 to $440–600 per day. Chinese government purchases rose from $6 million in the first half of 2024 to $62 million across 2025 to $230 million in the first half of 2026 — a demand curve with a specific and non-commercial shape. One maker, Deep Robotics, sold four humanoids across 2024 and 2025 while subsidies ran at 42% of its $4.2 million net profit.
Two commercial agreements in this collection — Agility Robotics with Toyota Motor Manufacturing Canada in February 2026, and with Mercado Libre in December 2025 — disclose no unit count, site, task, timeline or contract value in the company’s own announcements. They are recorded as announced agreements and nothing more, because nothing more is stated.