Summary of Key Points
Xpeng Robotics’ first round of financing exceeded $900 million, with a valuation of 43 billion yuan (approximately $6.3 billion), which is significantly higher than most leading embodied intelligence startups (typically around 20 billion yuan). What capital is really interested in is not the robot itself, but rather Xpeng’s existing automotive industry infrastructure, including factories, supply chains, quality systems, user channels, and data collection capabilities. Since 2026, domestic automakers have collectively entered the embodied intelligence field, each with their own strategies, but all share a core advantage: they possess “scenes” and “data.” Automotive factories and user channels can continuously generate high-quality training data, which is currently the most scarce resource in the embodied intelligence sector. The competition in embodied intelligence will ultimately depend on who can control the scenarios, generate data, and create a closed-loop iteration.
1. Xpeng Robotics’ Valuation of 43 Billion Yuan: Capital Is Buying into the “Automotive Foundation,” Not Just the Robot
Why did Xpeng Robotics receive a valuation of 43 billion yuan in its first round of financing, three times higher than that of many leading startups that have gone through multiple rounds of funding?
- Unique Existing Resources: Xpeng is a robotics company that emerged from the automotive industry, so it already has the necessary factories, supply chains (such as parts procurement and assembly lines), and quality control systems (meeting automotive-grade standards). Startups have to build factories from scratch and negotiate supply chains, while Xpeng can take a shortcut.
- Data Collection is Crucial: What’s lacking in embodied intelligence today is not technology, but data from robot operations. Xpeng’s factories have various tasks like stamping, welding, and final assembly, providing real-world data for training robots—no need to borrow facilities or worry about production interruptions due to trial and error.
- Capital Bet on Rapid Deployment: Xpeng plans to mass-produce robots by the end of 2026, with thousands of units per month, initially targeting stores and industrial parks. With existing sales channels and user bases, the path from prototype to market launch is much shorter for Xpeng.
2. Automakers Collectively Entering the Embodied Intelligence Field
Since 2026, almost all major automakers have started developing robots, each with different approaches, but they all have one thing in common: they have real-world scenarios. Examples include:
- Independent Business Models: Xpeng, Chery (MoJia Robotics), GAC (HuiLun Technology), and Changan (TianShu Robotics) have established separate robotics companies for research, production, and sales. Chery’s MoJia has already delivered 3,000 units and plans an IPO.
- Internal Collaboration: FAW, Dongfeng, and Seres are starting with in-house applications, first using robots in their factories (e.g., using robots for bolt tightening) before expanding to other scenarios.
- Investment and Cooperation: NIO (invested in Ren Shaoqing’s robotics company), SAIC (invested in ZhiYuan Robotics), Geely (invested in YuShu Technology), and Great Wall (partnered with YuShu) are binding together through investments or collaborations to leverage their technology and provide real-world scenarios for robot training.
Common Factor: All automakers have the real-world scenarios (factories) and user touchpoints (stores, car owners) that robotics startups lack; startups need to rent facilities and find customers, while automakers can use what they already have.
3. What Really Matters in Embodied Intelligence: Data, Which is More Difficult to Obtain than Large Models
Embodied intelligence robots need data to learn effectively, but collecting data for them is more challenging than for large models like ChatGPT or autonomous driving:
- Limited Data Sources: Large models have internet-based text and images, and autonomous driving has camera data from vehicles on the road. There are few robots that can work independently for long periods, making data collection difficult.
- Complex Data Dimensions: Robots need to record multiple aspects of their operations (vision, touch, joint movement, force applied, task success), and the data varies across different robots and sensors.
- High Collection Costs: Options include using humans to control robots (expensive), creating simulations (which differ significantly from real-world conditions), or replicating scenarios (but still not perfect).
Conclusion: Those who can collect high-quality data from real-world scenarios at low costs will have a competitive advantage in embodied intelligence.
4. Automakers’ Factories as a Gold Mine of Data
Automakers’ factories are ideal for data collection:
- Understanding Real Needs: They know which tasks in their factories require robots (e.g., lifting heavy objects, dangerous tasks) and which tasks are better suited for robots rather than traditional machinery.
- Comprehensive Data Collection: They have detailed information on task requirements (part tolerances, bolt tightening forces, fault handling, quality standards), which can be used to train robots effectively.
- Closed-Loop Data Circulation: Robots can learn from successful operations and failures, and the data can be fed back into models for continuous improvement.
- Low Risk of Trial and Error: Automakers can afford mistakes and even see them as cost-saving opportunities (e.g., ZeroRun aims to recoup investment in three years).
5. From Factory to User: Automakers in a Prime Position for Robot Commercialization
Automakers’ advantages extend beyond their factories to the user end:
- Stores as Transition Scenarios: Xpeng plans to deploy robots in stores, and Chery’s MoJia is already used in automotive and public services. These scenarios are relatively simple, allowing robots to gain user feedback and data.
- Understanding User Needs: Automakers understand how to turn technology into market-ready products, considering factors like price, noise, range, and after-sales support.
- Future Market: While factory scenarios are important for initial commercialization, the broader market is home users. Automakers have direct access to consumers through their channels (4S stores, car owner apps), making the path to market clearer.
Conclusion
The strategy behind automakers’ involvement in embodied intelligence is simple: they focus on creating scenarios that generate continuous high-quality data. This is where they have a significant advantage. The competition will not be about who makes the most realistic robots, but who can sustainably provide the necessary data. Automakers are in a perfect position, with factories for data generation and user channels for interaction data. Xpeng’s high valuation reflects capital’s recognition of this advantage. Their collective entry into the field indicates that they have grasped the core of embodied intelligence—data.
(The entire analysis is written in plain language, making it easy for non-financial professionals to understand.)