Xpeng’s Physical AI Pivot: How China’s EV Challenger Is Building a Multi-Carrier Embodied Intelligence Platform (2025–2030 Outlook)

In China, conversations about AI used to center on large language models and internet services.

Over the last few years, however, the keywords in China has been shifting toward “物理AI (Physical AI)” and “具身智能 (embodied intelligence)”.

At the center of this shift stands Xpeng (小鹏汽车) – a company that started life as an EV startup and now sits at the core of a rapidly forming ecosystem that spans humanoid robots, industrial robots, telecoms, and cloud providers.

Xpeng: From “AI in the Head” to “AI with a Body”

Xpeng no longer describes itself simply as an electric-vehicle company.

It now calls itself “a mobility explorer in the world of Physical AI, and a global embodied intelligence company”.

In other words, Xpeng is not just selling smart EVs. It is positioning itself as a company that builds AI systems that move in the physical world.

Xpeng’s Physical AI concept

When Xpeng talks about Physical AI, the idea can be summarized as:

AI should no longer live only on screens and in the cloud processing text.  It should become an intelligence that interacts with the real world through hardware such as robots, vehicles, and flying machines.

Because of that, Xpeng always frames its Physical AI strategy as a combined stack:

• Physical world 
• Hardware – vehicles, robots, flying vehicles
• AI foundation model – a large-scale model designed for perception and action

These three are treated as a single, tightly coupled system – not as separate layers.

Naver’s Leap into Physical AI – From Cloud Intelligence to Industrial Reality

VLA Gen-2: A foundation model for the physical world

Xpeng’s in-house second-generation VLA (Vision–Language–Action) model is explicitly described as a “model for the physical world”.

If traditional LLMs are optimized for “text → text” as language engines,  VLA takes a different approach:

• It connects visual signals from cameras directly to action commands,
• It tries to remove as much of the intermediate text step as possible.

In practice, that means:

Even if a human does not describe the scene in words, the robot or vehicle should be able to see, understand, and act on its own.

From the design stage, this model targets multiple physical carriers (multi-carrier):

• Autonomous passenger cars
• Robotaxis
• IRON humanoid robots
• Flying vehicles for low-altitude mobility

Chinese coverage repeatedly stresses that these four types of hardware are meant to share the same model architecture and evolve on a common Physical AI platform.

IRON, flying cars, robotaxis: Testbeds for Physical AI

Source: AI Strategica

Crucially, these are not four unrelated product lines.  They are “four different bodies operating on a single Physical AI platform.”

Why Xpeng’s Physical AI strategy matters

Xpeng’s Physical AI strategy can be captured in a single line:

From AI in the head, to AI with a body.

The company wants to move beyond AI that only manipulates text, images, and code in digital space, and drive AI into the core of real-world industries such as manufacturing, transportation, mobility, and aviation.

To do that, several capabilities inevitably become critical:

• Semiconductors – high-performance AI chips and NPUs
• Sensors and perception – cameras, LiDAR, and other sensing hardware
• Massive datasets – roads, factories, logistics, flight and low-altitude operations
• Training infrastructure – large-scale compute for model training
• Hardware–software integration – tightly coupling models with vehicles, robots, and flying platforms

Taken together, this shows Xpeng trying to move beyond the role of a traditional OEM and become a system player that integrates the full Physical AI stack – from chips and models to hardware and real-world deployment.

China’s Semiconductor Pivot — From Self-Reliance to Competitiveness

Implications & Strategic Questions

The strategic implications and answers to the following questions have been addressed in the AI Strategica CoreBrief.

Implications

• Embodied AI as a strategic pivot: In China, the narrative is visibly shifting from cloud-based LLMs to Physical/Embodied AI that controls real machines in real environments.
• Xpeng as a reference architecture: Xpeng’s combination of EVs, robotaxis, humanoid robots, and flying vehicles on one model stack is emerging as a template for multi-carrier Physical AI ecosystems.
• Rising hardware & integration premium: AI-driven differentiation is moving toward those who can control both the model and the machine – including chips, sensors, data, and integration.

Strategic Questions for Industry Stakeholders

1. If Xpeng succeeds, will “one model → many bodies” become the dominant paradigm for Physical AI – and who else is realistically positioned to copy or counter this play?
2. How should automotive, robotics, and aviation players respond if the competitive frontier shifts from “better vehicles/robots” to “better AI-hardware platforms”?
3. For semiconductor and sensor vendors, what does a unified Physical AI stack mean for future design requirements and ecosystem lock-in?
4. How can non-Chinese players in Korea, Japan, Europe, and the U.S. build their own embodied-intelligence roadmaps without being pulled into someone else’s platform orbit?

🔒Want deeper insights?

This SpotPulse® provides only a snapshot of the issue.   Access the full CoreBrief® and InDepth report® for in-depth analysis, data charts, and strategic implications tailored for decision-makers. Contact@AIStrategica.com