Browse
Log in / Register
Share
On March 14, 2025, Chinese electric vehicle (EV) maker XPeng Motors announced a strategic expansion of its in-house Turing AI chip: what was originally designed as the brain for intelligent driving is now powering the company's humanoid robot, IRON. This move marks a pivotal step in XPeng's Physical AI vision, where the same silicon architecture handles perception, planning, and control across different physical platforms—from cars to bipedal machines.
The Turing chip first debuted in XPeng's flagship EV models, delivering high-performance computing for autonomous driving functions like real-time object detection and path planning. Now, with the IRON humanoid robot rolling off XPeng's newly launched production line on March 14 (the company quietly began production at its Guangzhou facility earlier this month), the chip's role has transcended its automotive origins. According to a company spokesperson, the first batch of IRON units has already completed automated final assembly, a milestone that underscores XPeng's ability to vertically integrate hardware and software across product lines.
XPeng's Physical AI strategy hinges on the idea that the same underlying compute platform can serve multiple physical forms. The Turing chip, built on a 7nm process, features a custom neural processing unit (NPU) capable of 508 trillion operations per second (TOPS) under typical workloads. This level of performance is not only sufficient for Level 4 autonomous driving on the road but also provides the real-time reasoning required for a walking, balancing, and object-manipulating robot.
"Our philosophy is to separate the AI brain from the body," said He Xiaopeng, CEO of XPeng, during a recent internal briefing. "The Turing chip is the brain; whether it's in a car or a robot, the core intelligence remains the same, and we can iterate on it much faster." This approach contrasts with companies that design dedicated chips for each form factor, giving XPeng potential cost and development speed advantages.
The IRON humanoid robot, standing 5.6 feet tall and weighing 132 pounds, shares more than just a chip with XPeng's EVs. Its vision system leverages the same multi-camera and lidar fusion algorithms originally trained on millions of miles of Chinese road data. The robot's navigation stack borrows heavily from XPeng's XNGP (XPeng Navigation Guided Pilot) system, allowing it to map unfamiliar indoor environments and avoid obstacles with minimal latency.
One unique aspect of IRON is its ability to perform tasks that require fine motor control, such as picking up tools or assembling small components. XPeng claims the robot achieved a 95% success rate in factory trials for handling battery modules—a task that demands both strength and precision. The company plans to initially deploy IRON units in its own EV factories to assist with repetitive assembly work, with potential customer sales targeted for late 2026.
The expansion of the Turing chip into humanoid robots places XPeng in direct competition with players like Tesla (Optimus), Boston Dynamics (Atlas), and Chinese rivals such as Xiaomi (CyberOne) and Fourier Intelligence (GR-1). However, XPeng's vertical integration—designing the chip, the robot, and the training pipeline in-house—differentiates it from many competitors who rely on off-the-shelf components.
Industry analyst Li Ming from Counterpoint Research commented: "XPeng is doing something unique: they're not just adding a robot as a sideshow. By reusing the chip that already powers their self-driving cars, they dramatically reduce the cost of intelligence. If they can scale the production to hundreds of units, they could undercut robot makers who rely on expensive GPU-based compute modules."
According to Counterpoint's estimation, the Chinese humanoid robot market could exceed $6 billion by 2030, driven by industrial and service applications. XPeng's early move to embed its Physical AI chip into two high-growth product lines—
