WeRide spun off its data business into Jingshuo, an embodied AI and data infrastructure unit targeting 300 million yuan in 2026 revenue — up from 159.6 million yuan in 2025 — as autonomous driving companies seek to monetize their data engineering capabilities for robotics and physical AI customers beyond the vehicle.
Autonomous driving companies are repurposing the data pipelines they built for robotaxis into products for a new set of customers: robot makers, logistics operators and industrial AI developers. WeRide's move to spin off its data business into a standalone subsidiary shows how quickly that shift is taking shape.
Jingshuo, incorporated in May 2024 as a wholly owned unit of WeRide, targets 300 million yuan ($41 million) in revenue for 2026, up from an estimated 159.6 million yuan in 2025 and 55.8 million yuan in 2024, according to people familiar with the matter. The subsidiary has completed a Series A funding round and is positioning itself as an embodied AI and data infrastructure software provider — covering data generation, collection, simulation and proprietary data systems that serve not just vehicles but any physical AI system.
"Robotaxi companies spent years building the engineering systems to turn real-world driving data into training material for autonomous models," said a person familiar with Jingshuo's strategy, who spoke on condition of anonymity because the details are private. "The question is whether those same systems can be repackaged for robots that need to grasp, move and operate in factories and warehouses."
The spin-off reflects a broader reckoning in the autonomous driving industry. For years, companies like WeRide, Pony.ai and Baidu's Apollo unit competed on fleet size, operational miles and the number of cities where their robotaxis could run. Those metrics still matter, but investors are increasingly asking whether the underlying data infrastructure — the tools for collection, labeling, simulation, model evaluation and iteration — can generate revenue outside the vehicle.
Jingshuo's platform covers the full data lifecycle: raw data management, training data pipelines, annotation systems, model selection and evaluation, and deployment. It also includes an embodied AI business that handles data synthesis, teleoperation-based data collection and processing. The company says its accumulated real-world data already extends beyond automotive scenarios, though it has not disclosed specific non-automotive customers.
Why embodied AI needs a different data pipeline
Road data does not translate directly into robot data. Autonomous driving and embodied AI use different sensors, action spaces and task objectives. A robot needs data about grasping, moving, manipulating objects and interacting with humans — not just navigating streets. What can transfer across domains is the engineering methodology: data governance, scenario mining, simulation generation, model evaluation and the delivery workflow that turns raw sensor feeds into production-ready training sets.
This is the gap Jingshuo aims to fill. Rather than selling labeled images or video clips, it offers a system that generates synthetic data from real-world inputs and runs models through a continuous evaluation loop. For robotics companies, the bottleneck is not a lack of samples from any single scenario — it is the absence of a system that can produce high-quality training data at scale and at lower cost.
The market is still early. Frost & Sullivan, in a March 2026 report on physical AI simulation and data platforms, said the sector remains in a growth phase and that platform value depends on downstream application maturity and commercialization progress. The report noted that both smart vehicles and embodied AI are increasing demand for long-tail scenario reproduction, sensor simulation and closed-loop algorithm optimization.
A technical executive at an embodied robotics company, who asked not to be named discussing internal challenges, said limited training data and fragmented simulation stacks are the two hardest problems in physical AI research. World models and simulation engines can compress training and evaluation cycles, the person said, but they ultimately depend on real-world deployment to validate performance.
The revenue question
Jingshuo's 300 million yuan target represents a roughly 88% jump from the estimated 2025 figure, though the base is small. WeRide's overall financials remain tied to its robotaxi and robovan operations; the company previously raised independent funding for its Robovan unit at a valuation exceeding $400 million.
Nomura, in a July 2026 report, said data-as-a-service can generate revenue quickly on an hourly or project basis, but cautioned that providers lacking model evaluation and application capabilities risk being cut out as robotics companies build those functions internally. The distinction matters: data collection is visible and easy to charge for, but the harder-to-outsource work is continuously improving model performance.
The competitive landscape is already forming. In June 2026, Ruqi Mobility, a ride-hailing company backed by Guangzhou Automobile Group, launched its own embodied AI data platform, attempting to extend robotaxi data capabilities into robotics scenarios. The difference between players may come down to whether they keep data, world model and simulation capabilities inside the company to serve vehicle scale-up, or spin them out as independent businesses selling to external customers.
For WeRide, the bet is that Jingshuo can become a cross-scenario data infrastructure provider — not just a cost center supporting the parent company's autonomous driving R&D. If the math works, the competition among autonomous driving companies will shift from who has the largest fleet to who can sell their data engineering pipeline to the most physical AI customers.
WeRide shares trade on the Nasdaq under the ticker WRD. The company declined to comment on Jingshuo's operations or financial targets.
This article is for informational purposes only and does not constitute investment advice.
