WeRide splits its data business, betting on "outside-the-car" opportunities.

```

Author | Zhou Zhiyu

Autonomous driving companies are leveraging the capabilities accumulated by their fleets and are re-pricing these abilities in the physical world beyond automobiles.

In the past, Robotaxi companies mainly used their data for vehicle development: collection, annotation, simulation, and model training—all ultimately aimed at ensuring safe operation of cars on more roads. With the rise of embodied intelligence and physical AI, this data production capacity is now serving new clients: robots, logistics equipment, and industrial intelligent agents that require perception, decision-making, and actions in real environments.

This has changed the competitive boundaries of autonomous driving companies. They are competing not only for whether vehicles can be deployed, but also for whether they can turn their long-accumulated engineering systems into infrastructure usable by more physical AI scenarios.

Wallstreet News has learned from insiders that WeRide has promoted the independent operation of its data business, which is now undertaken by its wholly-owned subsidiary, Jingshuo. One insider said Jingshuo's business positioning is embodied intelligence and data infrastructure software services, including data generation and collection as well as its own data systems. The accumulated real-world data currently come from more than just automotive scenarios.

In response, on July 22, WeRide told Wallstreet News that there was no information to disclose at the time.

WeRide’s move means autonomous driving companies are transferring their capabilities across scenarios, building engineering systems for data collection, processing, simulation, evaluation, and model iteration.

For the Robotaxi industry as a whole, competition is shifting from accumulating more road mileage to creating repeatable products from the capabilities gained on roads. Whoever can keep generating orders for vehicle-independent systems will have the chance to shift from competing on autonomous vehicle solutions to competing in physical AI infrastructure.

Independent Operation

The business WeRide has spun off is a set of software services centered around data production and model iteration. It’s also the first time WeRide has spun off a business segment for independent operation.

Founded in May 2024, Jingshuo is a wholly-owned subsidiary of WeRide, leveraging experience from autonomous driving to build closed-loop AI data and solution systems.

Previously, Jingshuo already had some foundation in data services and delivery, albeit mainly as a supporting business in WeRide's autonomous driving R&D framework.

Now, Jingshuo’s externally displayed business has further expanded. Its AI platforms cover management of raw data, training data, and annotated data, as well as model selection, training, evaluation, and deployment. Its embodied intelligence business covers data synthesis, remote operation data collection, and data processing.

Wallstreet News has learned that Jingshuo is positioned as an embodied intelligence and data infrastructure software service provider, covering data generation, collection, and self-owned data systems, and its real-world data sources extend beyond automobiles.

What Jingshuo truly sells is not the number of samples, but the ability to turn real-world data into training material. Also, data collection is just one path—once real data enters the system, it can generate simulation data.

For robotics companies, the value of this ability is not in obtaining another batch of road videos, but in reducing data production cost and integrating data from different sources into a unified training and validation process. What the robotics industry lacks is not samples from a particular scenario, but a system that can continually generate high-quality data.

The earlier insider said this business has already accumulated over a long period; the decision to start independent operations and fundraising was made after recent evaluations.

That is, Jingshuo isn't starting from scratch in embodied intelligence, but is reorganizing its existing capabilities into a business for external clients.

Jingshuo's revenue target for 2026 is 300 million RMB. According to WeRide’s financial reports, revenue from smart data services in 2024 was about 55.8 million RMB, with an estimated 2025 revenue of 159.6 million RMB based on the reports.

Wallstreet News has learned Jingshuo recently completed its Series A fundraising round. WeRide’s internal judgement is that demand for this business is substantial and investors are confident in its prospects. Previously, WeRide also promoted the independent financing of Robovan, with a valuation exceeding $400 million USD.

Robovan’s and Jingshuo’s revenue logic are not the same. Robovan targets vehicles, logistics operations, and technical services—clients care about vehicle delivery, operational efficiency, and whether unit vehicle costs can be decreased. Jingshuo, on the other hand, targets data generation, simulation, and model training services—clients care whether their own models and robotic bodies can be connected.

But both rely on the WeRide One foundation. WeRide One is WeRide's general-purpose autonomous driving technology platform, supporting R&D, deployment, and operations across different vehicles and scenarios.

This is precisely the basis for Jingshuo’s independence: WeRide already possesses an internal platform for cross-model and cross-scenario technical reuse. Jingshuo’s task is to further package the platform’s data production and engineering capabilities into services for external clients.

Pricing Capabilities

Spinning off the data business may bring new imagination to the capital markets in the short term, but it will not automatically become a new source of growth.

For Robotaxi companies, real-world road data, autonomous driving algorithms, and operational experience accumulated in the past primarily served their own vehicles and fleets. Spinning off the data business truly changes the organization of these capabilities: it turns what was internal R&D cost into a business that can have independent pricing, financing, and accounting.

The issue Jingshuo faces is not whether data exist.

Road data cannot directly become robot data. Sensors, action spaces, and task objectives differ between autonomous driving and embodied intelligence. Robots need lots of data about grasping, moving, operating, and human-machine interaction. What can be transferred across scenarios is data governance, scenario mining, simulation generation, model evaluation, and engineering delivery processes.

Thus, the key for Robotaxi companies to enter embodied intelligence data business is not simply moving vehicle data to robots, but reorganizing engineering methods that served autonomous driving into data products suitable for various robot bodies and tasks.

Such moves have already appeared in the market. In June 2026, the data business of Ruqi Mobility released an embodied intelligence data platform, attempting to extend Robotaxi-related data capabilities to robotic scenarios.

The difference between companies isn’t who first proposed physical AI or whose marketing rhetoric is closer to large models. The key distinction is that some companies keep data, world modeling, and simulation abilities internal for vehicle scaling and tech iteration, while others spin these abilities off into independent businesses facing external clients.

In March 2026, Sullivan released a report on physical AI simulation and data platforms, stating that the sector is still in an early growth stage, and the platform value depends on downstream application technical maturity and commercialization progress. The report also noted that smart cars and embodied intelligence are increasing demand for recreating long-tail scenarios, sensor simulation, and closed-loop algorithm optimization.

Engineers at embodied robotics companies told Wallstreet News that limited training data and fragmented simulation stacks are the core challenges in physical AI R&D. Although world models and simulation engines can shorten training and evaluation cycles, real scenarios are ultimately indispensable.

This sets a higher bar for data service providers like Jingshuo. Simply offering annotators and collectors risks becoming a manpower service billed by project. True product-like offerings combine real data, simulation data, scenario libraries, model evaluation, and delivery processes to reduce clients’ time and cost in robot development.

Nomura Securities pointed out in a July report that data as a service can quickly generate cash flow by the hour or project, but suppliers lacking model evaluation and application capabilities may later see their offerings integrated by robot companies themselves.

This points to a reality in the data business: data collection is easy to spot, but continually improving model effectiveness is harder to outsource.

This is the new problem facing the Robotaxi industry. Previously, the market mainly measured such companies by fleet size, operating mileage, orders, and unit vehicle economics. In the future, what matters is whether companies can turn long-term R&D spending into orders beyond automobiles—and whether internal toolchains can become products external clients will continually purchase.

WeRide’s spin-off of Jingshuo offers an observation sample. What needs to be verified isn’t whether “embodied intelligence” as a concept can attract short-term attention, but whether the data and engineering systems accumulated by Robotaxi companies over years can be transformed from serving major R&D to cross-scenario commercial capabilities.

If the numbers add up, the competition among autonomous driving companies will shift from whose fleet is bigger to who can sell their accumulated data and engineering processes to more physical AI clients.

Only then will orders, repeat purchases, and gross margins truly enable them to receive new pricing.

Risk warnings and disclaimersThe market involves risks and investments need to be made cautiously. This article does not constitute individual investment advice, nor does it take into account any user’s special investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. Investments based on this are at your own risk. ```