XPENG unveils next-generation Iron humanoid at AI Day, targeting mass production by end of 2026

XPENG unveiled the next-generation IRON humanoid at its AI Day in Guangzhou, with mass production targeted for the end of 2026.

XPENG unveiled its next-generation Iron humanoid robot at its annual AI Day event in Guangzhou on November 5, 2025, themed Emergence, alongside a second-generation VLA model, a Robotaxi platform, and two flying car systems from its ARIDGE subsidiary. He Xiaopeng, the company's chairman and CEO, formally repositioned XPENG from an electric vehicle manufacturer to what he called a global embodied intelligence company.

XPENG was founded in 2014 in Guangzhou. He Xiaopeng co-founded UCWeb, the mobile browser company, before selling it to Alibaba in 2014; he subsequently served as president of Alibaba's Mobile Internet group before leaving to build XPENG. The company is listed on both the NYSE and the Hong Kong Stock Exchange and has invested approximately RMB 50 billion in its AI and physical AI stack, spanning robots, autonomous vehicles, chips, and flying systems. The Turing AI chip, developed entirely in-house, powers both the Robotaxi platform and Iron.

Five of XPENG's seven robot generations were quadrupeds. He said the company moved to bipedal robots because four-legged designs lacked hands and were poorly suited to tasks in homes and workplaces designed around human bodies.

The next-generation Iron stands 178 centimetres tall, weighs 70 kilograms, and features 82 degrees of freedom across its structure, with 22 degrees of freedom in each hand, enabled by what XPENG describes as the industry's smallest harmonic joint, allowing full 1:1 human-sized hand replication. Iron runs three in-house Turing AI chips delivering 2,250 TOPS of computing power and integrates three large model types simultaneously: VLT, a vision-language-task model described as the core engine for autonomous decision-making; VLA, adapted from XPENG's autonomous driving division; and VLM, also derived from automotive AI work.

At AI Day, He Xiaopeng explicitly dismissed factories and homes as Iron's initial commercialization targets, describing both as too complex and safety-constrained for the first deployment phase. Iron's near-term commercial focus is people-facing service environments including guided tours, shopping guide roles, reception, and traffic management functions. The rationale is data and iteration; service environments expose the robot to high volumes of human interaction in relatively structured settings, building the training data loop before moving into harder operational domains. XPENG plans to release an open Iron SDK, inviting global developers to build customised applications for the platform.

Iron is currently in practical training at XPENG's Guangzhou factory, handling sorting, transport, and other operations in the P7 electric vehicle production line. Baosteel, one of China's largest steel producers, was announced as an ecosystem partner at AI Day; Iron will be deployed at Baosteel to explore inspection and industrial applications and to iterate the platform under live industrial conditions.

XPENG established an embodied intelligence data factory in Guangzhou specifically to address the training data problem for physical AI. Mass production preparation is targeted to begin in Q2 2026; a 110,000 square metre full-chain humanoid robot manufacturing base in Guangzhou's Tianhe district was announced in February 2026, with phase one covering robot factory buildings, power stations, and supporting infrastructure. He Xiaopeng set a target of large-scale mass production of Iron by end of 2026, with long-term investment in humanoid robotics cited at up to approximately $13.8 billion over two decades. Iron is currently deployed internally and at Baosteel for iteration and testing; no commercial customer deployments or production volumes have been disclosed. The 2026 mass production target is the company's own projection.

Have a robotics update Korthos should review? Send news, deployments, product releases, funding rounds, research, or media to [email protected] or reach out on X at @agkorthos.