Xpeng is no longer treating humanoid robots as a side project. He Xiaopeng taking direct control of the robotics unit shows how seriously the EV maker now sees physical AI as its next battlefield.
Xpeng chairman and CEO He Xiaopeng has moved himself into the center of the company’s robotics push, taking charge of the unit just as its humanoid robot program approaches a more difficult phase: proving that a striking demo can become a repeatable product.
In an internal memo sent on Wednesday and reported by the South China Morning Post, He told staff he would personally lead the robotics department. He also disclosed that over the past year he had spent at least one full day each week inside the robotics operation. That is not a casual oversight role. It is the founder signaling that humanoids now sit beside electric cars, autonomous driving and flying vehicles in Xpeng’s long-term plan.
The timing matters. Xpeng has spent years trying to define itself as more than another Chinese EV brand fighting on price. Its pitch has increasingly been that cars are only one form of intelligent machine. If software, batteries, sensors and manufacturing can be reused across vehicles, robotaxis, flying cars and humanoids, then Xpeng’s real business becomes physical AI, not simply transport.
The company’s humanoid robot, IRON, has already done the thing a new robot must do first: get attention. It appeared at Xpeng’s AI Day in Guangzhou, walked with unusually fluid movement and triggered enough online skepticism that He cut open part of the robot’s exterior on stage to show there was not a person inside. The moment was theatrical, but it also showed the gap Xpeng must now close. A robot that looks convincing for a few minutes is very different from one that can work for hours.
IRON’s current promise is not that it will immediately become a household assistant. Xpeng has pointed first to commercial uses such as customer service, retail, tourism and internal operations. That is a more realistic route. Factories, stores and controlled public environments give robots clearer tasks and fewer surprises than a family home, where floors, objects, children, pets and expectations change every minute.
As the Wall Street Journal recently noted, Xpeng is already under pressure from its core car business after reporting a first-quarter 2026 net loss of 1.78 billion yuan and revenue of 13.03 billion yuan, even as margins improved. That makes the robotics bet more interesting, not less. Xpeng is trying to fund a future platform while still proving it can build a durable EV company in one of the world’s most aggressive auto markets.
The numbers around the car business explain why He’s personal involvement matters. In May 2026, Xpeng delivered 32,158 vehicles, a stronger month than April but still slightly lower than the same period a year earlier. The company has guided for 100,000 to 106,000 deliveries in the second quarter. That is a meaningful scale, but it is not enough by itself to escape the pressure of China’s EV price war.
China’s humanoid race is getting crowded
Xpeng is not moving in a vacuum. Tesla is pushing Optimus. Chinese robotics groups such as Unitree, Fourier Intelligence and EngineAI are racing to lower costs and improve walking, dexterity and autonomy. The broader Chinese manufacturing base gives local companies an advantage in components, batteries, actuators and rapid iteration. The weakness is still intelligence. Walking is hard, but useful work is harder.
That is why the phrase physical AI matters. A humanoid robot is not just a machine with knees and hands. It needs perception, planning, memory, balance, speech, task learning and enough reliability that a customer does not need to babysit it. The companies that win here will probably not be the ones with the most impressive stage walk. They will be the ones that turn messy physical behavior into a manufacturing process.
Xpeng has a credible reason to try. Its vehicles already require many of the same disciplines: sensor fusion, AI chips, power management, motion control and mass production. Its robotaxi plans also fit the same logic. A self-driving car is a robot constrained to roads. A humanoid is a robot asked to move through human spaces. The second problem is much harder, but the engineering culture is related.
The danger is distraction. EV companies have a history of selling investors expansive futures before the present business is fully settled. Xpeng is also developing flying vehicles through its AeroHT unit, including a modular flying car expected to move toward deliveries in 2026. Add humanoids and robotaxis, and the company has a very ambitious map. Ambition is useful only if management can sequence it.
He taking direct control suggests he believes the robotics unit has reached a turning point. The work now becomes less about proving that Xpeng can build an impressive humanoid and more about proving that it can build many of them, at a cost and reliability level customers will accept. That is the real test for every company in this field.
For investors and competitors, the signal is clear. Xpeng wants to be judged not only against BYD, Nio or Tesla’s cars, but against the next generation of machines that use AI in the physical world. The next thing to watch is not another viral robot walk. It is whether IRON can move from demonstration to deployment without losing the confidence that made the demo matter in the first place.
Also read: ZincFive is taking its nickel-zinc data center batteries public through a $600 million SPAC deal that signals a quieter revival in blank-check financing • Plaud is putting Singapore at the center of its Asia expansion • Apollo is now formally screening every software deal for AI displacement risk and the message for founders raising growth rounds is clear
