Last updated: 2026-08-26
XPeng began as an electric-vehicle manufacturer and, like several Chinese carmakers, concluded that the components a humanoid robot needs — actuators, battery systems, sensors, autonomy software and the silicon to run it — overlap heavily with what it already builds at volume for cars. Robotics was run inside the carmaker before being spun out into a standalone subsidiary.
That inheritance is the company’s central argument. Most humanoid developers must build a supply chain from nothing; XPeng argues it can borrow one. Whether that advantage survives contact with the very different tolerances and part counts of a humanoid is the open question.
IRON was shown publicly at XPeng’s AI Day in November 2025. The walking demonstration was smooth enough that a portion of the audience online assumed it was staged with a person in a suit — an accusation the company felt obliged to answer directly. Beyond the gait, XPeng emphasised the robot’s synthetic skin over a flexible lattice frame, and 21 degrees of freedom in each hand.
The August 2026 financing round was the point at which outside investors put a number on the effort: more than $900m raised at a $6.3bn valuation, with IDG Capital leading and Tencent and Alibaba among the participants. For a robot that has never been sold, that valuation rests on the production thesis rather than on revenue.
XPeng’s stated plan is mass production by the end of 2026 inside its own facilities, deliveries to customers in 2027, monthly capacity above 1,000 units, and a million units by 2030. Those are targets. The humanoid sector has a consistent record of missing them — Tesla’s Optimus programme being the most visible example — and IRON has no published price, no order book and no independently verified specification for height, weight or runtime.
Against Tesla Optimus, IRON is further along in disclosed specification but behind in nothing measurable, because neither robot is on sale. Against Figure, which by mid-2026 was building at roughly one robot per hour at its BotQ facility, IRON has yet to demonstrate a working production line. Against the broader Chinese field — Unitree, AgiBot, UBTECH, Galbot — IRON is unusual in that its parent already ships complex hardware at automotive volume, and unusual again in designing its own inference silicon.
Some content on this page was created with the assistance of AI tools.