Xpeng's Humanoid Robot Line Starts, Shifting Race to Mass Production
On September 8, 2026, Xpeng announced that its robot production line had officially started operation. The first high-order general-purpose humanoid robot, IRON, completed automated final assembly and walked off the line on its own.
According to company information, Xpeng designed and built the line, with more than 80% of core manufacturing processes automated.
The line's significance is not whether one robot can walk. It marks a key turning point in humanoid robot competition: from 'can it be built' to 'can it be mass-produced.'
The core challenge in mass production is not technical principle but manufacturing consistency. As output rises from dozens to thousands of units, ensuring consistency across thousands of joint modules and simplifying single-joint assembly from dozens of bolts into an automated process becomes critical to delivery capability.
Xpeng's approach migrates automotive manufacturing systems to its robot line. According to the company, the line deeply integrates mature automotive quality control with precision humanoid manufacturing, using automotive-grade consistency standards to build mass-production quality.
The advantage is that automakers have already accumulated full-scale precision manufacturing capabilities—supply chain management, automated assembly, batch consistency control and cost control. These are precisely the weaknesses of most robotics startups.
IRON differs from most humanoid robots in technical path. It carries three self-developed Turing AI chips with 2,250 TOPS of computing power, and its physical AI foundation model can run directly on the robot rather than relying on remote control.
This positioning is similar to Tesla's Optimus. But Automotive World reported that Optimus mass production has been delayed to the second half of 2027, and Musk admitted in January that the robot 'cannot yet do useful work.'
Figure has chosen another path: a $3.5 billion computing procurement deal with Nscale, planning to deploy up to 100,000 Nvidia GPUs to train the Helix model. This path emphasizes continuous iteration of the 'brain,' but its mass-production timetable is also unclear.
The divergence of these three paths reveals a reality: the humanoid robot industry has not yet converged on a unified technical route or business model.
Xpeng's robotics business has completed a first equity financing round of more than $900 million, led by IDG Capital with participation from Alibaba and Tencent, at a post-investment valuation of about $6.3 billion. It is the largest single round in China's embodied intelligence sector to date.
But funding scale does not equal delivery capability. Xpeng says IRON will initially be deployed in its own stores and campuses. Formal commercial sales have not started, and overseas markets are planned for 2027.
Zhang Yun, global CEO of Ries Strategic Consulting, notes that IRON belongs to a new category of 'humanoid robots,' unlike products that prioritize looking human-like. The ultimate test is whether IRON can operate stably in real scenarios, not merely walk once on a production line.
2026 is viewed as the 'first year of mass production' for humanoid robots. Industry data show China shipped more than 40,000 humanoid robots in the first half, about 97% of the global total. Morgan Stanley has raised its forecast for China's 2026 humanoid robot shipments to 50,000 units.
But shipment growth does not equal commercial success. The launch of Xpeng's line proves Chinese companies have the manufacturing ability to move humanoid robots from lab to line. The next question is whether these robots can continuously create value in stores, factories and homes.
China's Ministry of Industry and Information Technology is advancing a standards system for the humanoid robot industry, planning to complete at least 100 key standards by 2028. As standards improve, the dual test of mass production and scenario deployment will determine who stays at the table.