How China’s XPeng Is Using AI to Make Customers’ Lives and Their Cars Better

As one of the hottest topics currently in the zeitgeist, it feels like every industry is looking to snag their own piece of the artificial intelligence pie. Some companies have been using it for some not-so-great reasons. And then there are the ways where it can legitimately help people. As with any tool, its primary role should be to assist and enhance our lives. Hopefully, Chinese automaker XPeng’s big bet on AI leads to its customers benefiting in tangible ways through its cars and robotics.

At a brand and product launch event in Munich, Germany, earlier this month, XPeng unveiled the 2026 XPeng L03 for the European, Australian, and Chinese markets. It’s approximately a Ford Mustang Mach-E-sized SUV that’ll be available in either all-electric or EREV-powered forms. But beyond just cars, XPeng wants AI to be the lifeblood of the company, so much so that it’s calling its current era Physical AI for All.

At the core of everything lies XPeng’s own in-house-developed AI Turing chips that boast 40-core processors and 750 Trillions of Operations Per Second (TOPS) of computing power, and foundation models. Needless to say, that’s a lot. Around these, XPeng says it is building exclusive power systems, hardware, and operating systems that will run on its AI and bring it into our physical realm—a vision far greater than something as simple as a large-language model.

The XPeng L03

Advanced Semi-Autonomous Driving

Powered by three AI Turing chips, Ultra versions of the L03 SUV will offer the latest version of the automaker’s Navigation Guided Pilot (NGP) system that’s powered by its emerging VLA 2.0 semi-autonomous tech, a setup that should give the supervised version of Tesla’s Full Self-Driving a run for its money. In certain locations, the company is working on training the system with the goal of rolling out mapless driving capabilities in short order. XPeng also revealed that it started testing its first robotaxi in Guangzhou, China, earlier this year, running on the most advanced version of VLA, with the ultimate end game being Level 4, fully autonomous driving for all of its vehicles.

XPeng has identified unstructured data, continuous signals, and real-world interactions as some of the challenges of physical AI. Basically, there’s so much data coming in from the car’s sensors, XPeng claims that the only way to label it all is to use a model so there’s no information loss. And because the company wants to roll out NGP in Europe, it has to train it to recognize the cyclists, pedestrian scooters, and signs that are more prevalent on European roads.

This part at least sounds familiar. Driving habits absolutely do differ from market to market, so it’s crucial that any semi-autonomous system is well-versed in the native environment before it’s unleashed onto public roads.

An image of XPeng’s Iron humanoid robot at a museum.

Iron Robot

As for its humanoid robot, Iron, XPeng says the Turing chip means it won’t need a remote control at all. On its own, Iron will be able to “perceive, decide, and interact all by itself.” Also, it’s human-shaped because XPeng decided this was the best way for a robot to assist us in factories, homes, and businesses. Iron will see a global launch next year, according to the company.

This is potentially a huge leap from what’s commercially available elsewhere. Among what’s out there now, there are the Boston Dynamics robot dogs that perform quality checks at automotive plants, and there are automated pallets for transporting parts across a factory floor. Boston Dynamics also has a humanoid robot, Atlas, that will be AI-trained and eventually see additional customers beyond Hyundai (its owner) and Google starting in 2027. Tesla is also pushing hard into the robotics sector with its Optimus project, and we recently reported on the Unitree G1 humanoid robot that’s presently available for sale and got a deep dive into its operation.

In the past, Iron was apparently so lifelike in its walk that online users accused it of being a person in a costume. To disprove it, XPeng CEO He Xiaopeng had the robot’s leg cut open onstage before a live audience to prove there was no person underneath.

NGP for All

We’ll withhold our final say on Iron until we see it in action, but in the time we spent with NGP in an XPeng G6 last year, we came away impressed. It was generally smooth and erred on the side of caution. We’re just not yet able to say whether Tesla FSD or XPeng NGP is the better bet, having not yet had the opportunity to test both across the same location.

However, customers can get NGP on XPeng’s Mona line of models, which are the automaker’s most affordable offerings. That means the technology will be placed at a more accessible entry point as it develops, much like Tesla has done with its supervised version of FSD.

Given the current climate around Chinese vehicles in America, there’s no saying when or if XPeng and its products will enter our market. But we hope that someday they will, because the technology sounds and feels state-of-the-art, and more choice is always good for the consumer.