XPeng to Begin Selling Humanoid Robot Iron in Second Half of 2027

XPeng plans to begin selling its humanoid robot Iron in the second half of 2027, according to reports. humanoid.guide reported that Brian Gu, vice chairman and president, gave the timeline at the Forbes Global CEO Conference in Singapore. The company expects to begin mass production in 2026 and use the robot to assist sales staff and handle reception duties at XPeng’s Guangzhou headquarters early next year.

By laying out a sequence of mass production, internal operations and external sales, XPeng has made clear its intention to move humanoid robots from the prototype stage into its business plans. With no reported price or production volume, the focus is less on product performance itself than on whether the company can transition to mass production and validate operations in stores and offices.

What is new in this report is that XPeng is discussing Iron not as a research-and-development display piece, but through a phased plan: mass production in 2026, internal operations early next year and sales in the second half of 2027. However, the company has not disclosed production volume, price, operating conditions or battery life. The schedule is therefore laid out ahead of details on supply capacity and profitability. A related story from this newspaper, Amazon Unveils Physical AI Toolchain on AWS, Adopting NVIDIA’s Physical AI Stack, showed a trend toward opening development platforms and computing stacks to external users. By contrast, XPeng appears to be prioritizing real-world operational testing, moving from trials in its own stores and offices to external sales, rather than opening its development platform. Iron’s use of the company’s own Turing chip also sets it somewhat apart from the common robotics development path that relies on NVIDIA, linking in-house computing infrastructure with the commercialization of the robot itself. One implication for the industry is that competition in humanoid robots is shifting beyond demo videos and prototype quality to how companies sequence sustained internal operations and the transition to mass production. If XPeng first tests human-facing applications in its own stores and offices, uses such as conversation, reception and sales support — closer to light-duty tasks than factory work — will likely come first. Key areas will include real-world data collection, software updates and the stability of inference on the company’s own chip, although the report does not specify how these will be evaluated. Unresolved questions include the scale of mass production in 2026, which markets will be targeted for sales in the second half of 2027, and the level of autonomy planned for internal trials. In addition, claims about performance compared with the NVIDIA Thor chip remain unverified, with no testing conditions provided. Independent evaluations and results from real-world operations will therefore be worth watching.