XPENG has switched on dedicated humanoid production lines in Guangzhou, and its next-generation IRON robot finished assembly and walked off the line under its own power. The company calls it a leap from research prototypes to factory manufacturing for advanced general-purpose humanoids.
Look closer at that more than 80% automated claim, though, and the story gets stranger — and more exciting. Roughly one-fifth of the core work still needs human hands. The obvious next move, if IRON and its kin keep getting better, is to put humanoids into those remaining gaps: robots helping build robots, a factory that starts to look like a soft von Neumann machine.
From lab prototype to the factory floor
According to XPENG, more than 80% of core processes on the new lines are automated. The setup borrows automotive-grade quality systems from its smart EV business and applies them to precision robot manufacturing. Chairman and CEO He Xiaopeng said the lines were built from scratch with no existing playbook, and that XPENG will keep pushing for faster cycles and greater scale.
Commissioning the lines is not the same as full mass production. XPENG still targets mass production by the end of 2026, with early robots going into its own stores and campuses first. An official launch and deliveries in China and overseas markets are planned for 2027. Capacity and pricing were not disclosed.
What IRON brings to the floor
Unveiled in November 2025, the latest IRON is built as a general-purpose humanoid platform:
- 76 degrees of freedom across the body, with 21 per hand - A proprietary fully enclosed flexible lattice structure meant to balance a human-like look with safety - Three in-house Turing AI chips providing up to about 2,250 TOPS, so XPENG’s physical-AI model can run onboard
That onboard compute is meant to let IRON handle complex tasks without remote operation, with lower inference latency and tighter control over data. In plain terms: the robot is designed to see, decide, and act on its own — the same toolkit you would need if a humanoid were ever asked to wrench, inspect, or finish another humanoid two stations down the line.
Why it matters: the 20% that still needs a body
Humanoid demos have been plentiful. What has been scarce is repeatable, high-automation manufacturing that can turn those demos into units at commercial volume. XPENG’s line is a real test of whether general-purpose humanoids can leave the lab on a schedule the industry can plan around.
The unfinished fifth of the process is the plot twist. Every station that still needs a person is a station a future IRON might one day staff. Mathematician John von Neumann imagined machines that could construct copies of themselves. Nobody is claiming XPENG has that — yet. But stack an 80% automated humanoid line next to a general-purpose robot with human-scale hands, onboard AI, and factory-floor ambitions, and you get the first industrial sketch of how that loop might close: humanoids filling the automation gaps on the lines that build the next humanoids, then pushing automation toward 90%, then 95%, while capacity compounds.
Capital markets are already treating the bet seriously. In late August 2026, XPENG’s robotics unit Dogotix signed roughly US$900 million in financing commitments, with an implied post-money valuation around US$6.3 billion if the equity incentive pool is fully used. IDG Capital led; Alibaba and Tencent joined as strategic investors. XPENG expects to keep control and consolidate the unit’s results. Money like that is not only paying for smarter hands — it is paying for the factories those hands may eventually run.
What’s next
Watch the year-end mass-production claim, the first store-and-campus deployments, and whether 2027 deliveries land on time. Watch harder for any sign that IRON units are assigned back onto the production floor itself. For Rich Brew Co readers tracking physical AI, IRON’s walk off the line is not just a robot leaving a factory. It is a glimpse of a factory that might one day staff itself.
Sources: XPENG / PR Newswire, CnEVPost.