The state of humanoid work, mid-2026

The honest number of humanoid robots doing real, paid industrial work today is small — industry trackers put it in the low hundreds worldwide. They move totes, load sheet metal, and carry parts a few metres to a conveyor. They run for an hour or two on a charge, often sit inside safety cages, and still need a human nearby. This is not the world the launch videos imply.

But the word that matters is paid. Figure has run its robots inside BMW's Spartanburg plant on real shifts; reporting describes a fleet on the order of forty machines billed by the robot-hour, logging thousands of runtime hours and handling tens of thousands of sheet-metal parts. Agility's Digit has moved totes in a live GXO logistics operation. Apptronik has work underway with Mercedes-Benz and Jabil, and Sanctuary is partnered with Magna. These are early and narrow, but they are not demos.

On price, mid-2026 is a study in spread. Unitree's G1 is listed around USD 16,000–18,000 and the company shipped several thousand units in 2025 — more than the rest of the field combined — while targeting far higher volumes this year. 1X has put its NEO at roughly USD 20,000 outright, or a monthly subscription, with first home deliveries expected late in the year. Tesla's Optimus is projected in the USD 20,000–30,000 range but is not yet sold to outside buyers.

The capable industrial machines cost far more — six figures per unit is a fair estimate once the work is real. So the USD 20,000 figure is best read as a floor set by consumer-grade hardware, not the price of a robot that can hold down a shift. The gap between the cheapest walking humanoid and the cheapest useful one is the whole game.

The near-term trajectory is legible. Several hundred industrial humanoids in 2025 become low thousands across 2026 and 2027. In Germany, Schaeffler has said robots will begin operating at two of its sites before the end of 2026, with deployment running into mid-2027. Nobody serious is claiming broad factory-scale adoption this year; the consistent read across analysts is that the mainstream shift belongs to the 2030s.

That timing is the point, not a disappointment. It is exactly the window in which a company can build the layer that decides where these machines are worth deploying — before the machines themselves are commodities.

Moduloa keeps 25 dated claims on the record. Two are worth checking against this note. The first — that humanoids are doing real, paid industrial work, not pilots or demos, by 2031 — looks directionally on track: the paid work exists, but most of it still reads as pilot-scale, so it stays open. The second — that humanoid unit cost falls well below the 2026 level of roughly USD 20,000 by 2031 — cannot be scored yet, because that USD 20,000 is the present baseline, not a fall beneath it.

Neither claim moves this week. That is the correct outcome: the register is scored only when the evidence is decisive. What this note records is the direction of travel — and the direction is toward the world the thesis was written for. See the register →

This is a synthesis of current public reporting, not a primary source itself. Key references:

Interesting Engineering — humanoid robots to work in German factories (Schaeffler) · AI Business — what a humanoid robot actually costs in 2026 · Robotics Center — Tesla Optimus price and availability

Figures on deployments and pricing are drawn from company statements and trade reporting as of July 2026 and will age quickly. Corrections are welcome: shk@moduloa.com.