Robots. Custom-built chips. Autonomous vehicles. Xpeng's strategy seems similar, with equal challenges to overcome. However, it could s...
Robots. Custom-built chips. Autonomous vehicles. Xpeng's strategy seems similar, with equal challenges to overcome. However, it could surpass Tesla in its own domain.
The robot is impossible to overlook. Its clearly feminine shape draws attention with its almost uncanny measurements and lifelike, model-like walk.
The approximately 5'4" IRON robot's look and gait were so realistic that journalists and influencers, both inside and outside of China, believed that a growing automaker was behind it.Xpenghired a woman to don a cosplay-style shape-shifting suit and robotically perform a supermodel runway walk across the stage of its new auditorium, akin to a more realistic version of Tesla's initial presentation with itsOptimus robot.
But whether or not you think Xpeng's robot is fake—and after seeing the unit up close, it seems to be genuine, at least in terms of its ability to walk—the headline-making robotis significant in numerous ways.
Certainly, Xpeng's internal team hasn't provided a clear explanation yet regarding the design choices made for the robot. Furthermore, Xpeng has not clarified the commercial potential or future direction of the robot, other than a general commitment to mass-produce the androids and offer them for sale in 2026.
But the entire initiative says a lot about Xpeng, which utilized the recentAI Day event in Guangzhouto enter the global awareness for the first time.
Xpeng's robot initiative might be uncertain, but similar to Tesla, it clearly aims to transcend being just a car manufacturer. However, unlike other automotive companies that have evolved into tech powerhouses, Xpeng's future strategy appears significantly more realistic compared to what I've observed from other brands—particularly Chinese ones struggling with profitability, relevance, and visibility in a market that is becoming increasingly competitive.
After spending most of a week at its new Guangzhou global headquarters, I believe Xpeng is one to keep an eye on, both within and beyond China. It's clearly following Tesla's strategy—except Xpeng might be reading it more quickly and understanding it better.
What Is Xpeng?
There are numerous Chinese electric vehicle companies, so I won't blame you if you've never come across Xpeng. The company refers to itself as a "tech startup," even though it is 11 years old andover one million vehicles manufactured worldwide, I believe we have moved far beyond the label of "startup."
The company offers a complete range of technology-driven electric vehicles and has also introduced plug-in hybrids and extended-range EV models this year. These all occupy a somewhat premium-like segment in China. It is similar to Tesla's position in the market, emphasizing refinement and technological features rather than pure luxury.
Notably, a former Tesla engineer was charged with stealing the Autopilot source code and selling it to Xpeng, although Xpeng claims it has upheld the intellectual property rights of everyone involved. (Tesla and the engineer ultimately)resolved the legal dispute in 2021.)
In recent times, Xpeng has begun to explore international markets by offering its range of affordable midsize and small SUVs. Its success in Europe has been limited, with Xpeng's deliveries reaching approximately 25,000 units this year.
Subsequently, Xpeng's approximately $18,000 Mona M03 sedan was introduced. Within just over a year, Xpeng has sold 200,000 units of this vehicle, with intentions for the model to expand into markets beyond China.
Recently, it launched a "Max" version of theMona M03, which includes Xpeng's Level 2 driver assistance features, a technology that was not very common in vehicles at this price point.
Although they have achieved some successes, it is challenging to make a name for oneself in the highly competitive Chinese automotive market. This is why Xpeng is clearly expanding its focus beyond just cars.
Xpeng's Strategy for Self-Driving Technology: Artificial Intelligence, LIDAR, and Beyond
In China, semi-autonomous and self-driving technology is practically essential for any car manufacturer. Really. It's difficult to imagine a car brand in China that isn't developing its own efforts towards Level 3 and higher autonomy—allowing for hands-free and eyes-free driving. The only obstacle is government regulations.
BYD's outstanding "God's Eye" system, a camera-based driver assistance system, is available on all models sold in China, including the very affordable Seagull. I have also used these systems in Nio and Onvo vehicles, which employ either LIDAR (Nio) or camera-only (Onvo) autonomous technologies. Or,Zeekr, which employs LIDAR and certainNvidia Drive Orin ChipsTo enable semi-autonomous driving. Or Li Auto, which incorporates certain NVIDIA Drive Orin chips and some LIDAR tricks to navigate the road.
The main idea is that if you visit almost any brand in the People's Republic, you'll see the same few steps used exactly the same way for every car brand. It's similar across the entire Chinese automotive industry, which is why some brands are having a hard time differentiating themselves in a competitive market filled with otherwise strong entries in China's new-energy vehicle (NEV) sector.
However, now that I've been back home for a short while and had the opportunity to reflect on what I witnessed (and to recover from a severe cold), I realize that Xpeng's emphasis on AI seems to be directly inspired by Tesla's approach. And if they handle it well, it could potentially result in comparable success, not only in China but beyond.
What Is Physical AI?
Xpeng's Global AI Day began with a declaration, a "new creation," as it claimed, named "Physical AI."
On the surface, this is more of a promotional or style-setting phrase for the entire conference, rather than an actual definition or original concept. After navigating through the marketing and exaggerated language, Xpeng's emphasis on "physical AI"—a term that is becoming increasingly common these days—refers to how artificial intelligence can and will engage with and navigate the real world and physical environment. Xpeng made it clear that this is an internally developed model, not something strange, intended for humorous or eerie generative videos using a large language model, or a Tesla Grok-like chatbot that is integrated into the vehicle.
Xpeng views physical AI as the embodiment of its extensive data models, enabling understanding of the environment. The founder and CEO of Xpeng, He Xiaopeng, aims for autonomous driving technology to exhibit human-like reasoning and reactions, rather than appearing mechanical.
For them, this implies that the AI model should attempt to think like a human, or at least mimic that process. According to Xpeng, human logic involves converting "visual information" into actions, while a standard AI model uses visual elements combined with language to understand the environment. In simpler terms, AI models continuously receive data, which can be real text or input from cameras or LIDAR. This data is then structured and analyzed by the model. Remember, this occurs while the vehicle is in motion, so fresh information is consistently being gathered and processed.
Now, consider the aspect that the vehicle (or robot) must operate in an environment that is constantly evolving; this presents a distinct challenge for an AI system, particularly when contrasted with other AI systems that do not need to manage real-time data changes.
Your brain constantly performs these tasks, so it's more capable than you realize. Replicating this in a machine is extremely challenging. According to Xpeng, creating a functional system required significantly reducing both the amount and type of data processed by the AI model.
In order to achieve a more natural autonomous experience, Xpeng states that they must reduce the volume of data being received and processed. While Xpeng collects a significant amount of data from what its vehicles observe, it is not feasible from a technical perspective to process all of it, nor is it beneficial for the final outcome.
Xpeng refers to it as VLA 2.0, standing for Vision, Language, and Action—however, in the 2.0 iteration, the Vision and Language components have been integrated.
In previous VLA models, Xpeng would handle both visual elements (objects) and all other background information—particularly language and text. This large amount of data is the reason Xpeng believes other AI models operate too slowly; according to people I spoke with, the concept is that most humans don't need to read every bit of text to determine the right action. Therefore, AI should function similarly.
Xpeng claims to have gathered approximately 100 million hours of driving data from video recordings. Using this, it has been able to create simulations of "long tail" theoretical scenarios, which aid in training the model to think and respond appropriately in real-world situations.
Does It Actually Work?
Xpeng provided multiple test drives for its affordable competitor to the Tesla Model 3, the Mona M03 Max, as well as itslatest P7 sport sedan. We experienced some point-to-point semi-autonomous Level 2-like urban navigation, indicating that our attention was on the road and our hands remained on the steering wheel.
Certainly, the Mona and P7 were capable of navigating traffic on public roads near Xpeng's Guangzhou campus. They could also maneuver through Xpeng's complex underground parking system, avoiding obstacles and the many Xpeng employees commuting to work as well as contractors working on the final details of Xpeng's new headquarters.
The vehicle was capable of independently moving in and out of a parking spot, and even selecting a different one when a contractor in a Maxus-branded van accidentally obstructed its initial plan.
However, this was not VLA 2.0, nor did it stand out in terms of autonomous driving technology in the Chinese market. Earlier this year, just a few days after the Shanghai Auto Show, I saw a Zeekr 007 perform a similar action at Zeekr's headquarters in Hangzhou, except the car did it completely without a driver—no humans were present inside.
Then, to demonstrate its capabilities to all the Westerners capturing footage with their phones or DJI action cameras, it concluded the impressive demonstration by connecting to a DC fast charger, achieving nearly one megawatt of charging speed.
Hell, Nio, and Onvo's autonomous driving capabilities were just as remarkable as Xpeng's, capable of nearly reaching the door and directly heading to one of its battery swapping stations. Although Xpeng's technology is good, it isn't distinctive—also, the software we evaluated was not part of its AI-focused VLA 2.0 announcement that was mentioned to us the previous day.
However, it could be argued that Xpeng's software performance or demonstration of its current capabilities is not insignificant when viewed within the broader context of its ambitions: global dominance.
Xpeng Aims to Expand Internationally
The VLA 2.0 model serves as a crucial element in Xpeng's future growth strategies. Initially, the company states that the VLA model will be capable of operating beyond China with little additional effort or training, including in Europe, which represents a significant market expansion for the brand.
This represents a significant advancement compared to other Chinese brands, which frequently lack the complete set of assisted driving features due to various factors, often linked to cost and data privacy regulations. BYD's "God’s Eye" software has not been launched outside of China yet, while vehicles such as the Zeekr 7x and Nio ES6 (known as EL6 in Europe) lose their LIDAR-enabled autonomous driving capabilities when sold in Europe.
However, Xpeng states that its upcoming self-driving AI model will operate within the vehicle itself—and adhere to all data-sharing regulations. Xpeng's internal processors, called the Turing Chip, serve as the foundation for Xpeng's future developments. These chips, used in threefold, are designed to manage all of Xpeng's autonomous driving capabilities; forming the basis of its future Robotaxi service (scheduled to debut in 2026), and the upcoming "Robo" version of its current vehicle models.
They are claimed to offer robotaxi-like features, but through personal vehicle ownership. The company states its goal is to achieve genuine Level 4 self-driving capability by next year. That's a significant ambition, especially when considering that Tesla's camera-based Robotaxi is only beginning to launch in very limited locations.
The same chip and AI model logic is believed to power the IRON robot, although it remains uncertain whether any of this technology was operational for the single proof-of-concept walking demonstration Xpeng presented during its conference.
That's A Great Number Of Commitments. Can Xpeng Fulfill Them?
Although these concepts are impressive, I remain unconvinced that Xpeng has demonstrated enough to confidently claim it can succeed. I'm unsure if Xpeng's new AI model functions as its executives and engineers describe. A 2026 goal for full Level 4 autonomy and a robotaxi service relying solely on cameras seems ambitious. The idea of selling a vehicle capable of driving or running errands completely on its own still feels like a distant dream. Tesla made similar promises years ago, and it hasn't delivered yet.
Sure, here's a paraphrased version of the text: "Even Xpeng admits it hasn't figured out the path to making its IRON robot commercially feasible. Although it features "soft skin, bionic muscles, and dexterous hands," the company openly states that it doesn't view the product as viable in manufacturing environments, in contrast to Tesla's ambitions for its Optimus robot."
However, there are certain aspects we are aware of. Xpeng's present set of driver assistance software functions effectively. The vehicles it is installed in are attractive, well-crafted, and reasonably priced. There is a reason why it has become one of the fastest-rising brands in Europe.
Every Xpeng vehicle I have had the opportunity to sit in felt well-built and of high quality, in contrast to Tesla, a brand that has poor initial quality and an outdated product line that has been neglected by a founder who is resting on his past successes. Any technological enhancements added to Xpeng's current range only make an already strong lineup even better. Additionally, VLA 2.0 is good enough for Volkswagen China to become its first client. That must signify something.
Nevertheless, it's simple to visit any Zeekr, BYD, Geely, Li Auto, Leapmotor, Deepal, or other brand's dealership and discover something that seems technologically impressive to use or drive. Similar to Xpeng, when considering a company's financial performance, it's evident that it must go beyond merely manufacturing or selling vehicles if it hopes to achieve significant profits and revenue. This approach was something Tesla began preparing for in recent years, although it has not yet shifted away from being primarily a car company.
And, if Xpeng aims to endure the upcoming significant reduction in Chinese brands, it must realize that commitments alone will not carry it much further.
Contact the author: kevin.williams@insideevs.com
- Why Xpeng Is Adopting Tesla's Disputed Strategy for Self-Driving Technology
- Xpeng, China's Electric Vehicle Manufacturer, Is Now Producing Cars in Europe
- The New Xpeng P7 Is a Beautiful Sedan Designed for Global Markets
- Analysis of Xpeng's Under-$20,000 Vehicle Competing with the Tesla Model 3: Better Value for Money
COMMENTS