- Startup builds real-world data and training systems for robotics, autonomous driving and AI models
- Funds to support expansion of training sites, data generation and validation capabilities
Chinese AI infrastructure startup BodenAI (博登智能) has raised several hundred million yuan across two tranches of financing, as investors bet on growing demand for real-world data and testing systems underpinning the next generation of AI applications.
The Ningbo-based company said it completed Series A+ and A++ funding rounds backed by investors including CDH Investments, Fresh Capital, Lucion Venture Capital and Shenzhen Capital Group.
BodenAI plans to use the proceeds to expand what it calls a three-layer ecosystem covering real-world scenario networks, automated data-generation engines and physical-world validation systems.
‘Train at scale, validate in reality’
The effort supports the company’s strategy of training AI models at scale while validating them in real-world environments.
The fundraising comes as AI developers increasingly focus on obtaining high-quality real-world data to train and validate systems that operate beyond text and images, including robots, vehicles and other physical machines.
Founded in Ningbo, BodenAI develops infrastructure for what the industry increasingly refers to as Physical AI — systems that allow AI to perceive, interact with and operate in physical environments.
Its platform covers data collection, annotation, anonymization and synthetic data generation.
“AI’s first phase belonged to models and algorithms. The next phase will belong to the real world,” founder and chief executive Zhao Jie said.
“The upper limit of AI capability will depend not only on model architecture, but also on a complete system of real-world training, physical validation and continuous learning.”
Data collection farms
The company’s BASE platform provides data training services for sectors including autonomous driving, embodied AI, healthcare and large language models.
BodenAI has established three physical training sites for embodied AI in Ningbo, Ma’anshan and Huzhou, spanning more than 30,000 square meters.
The facilities house nearly 500 robots and thousands of proprietary data-collection devices, giving the company capacity to generate about 500,000 hours of robot training data annually, it said.
The company said it works with domestic and international companies and research institutions on large-scale data production.
In autonomous driving, it has processed hundreds of millions of frames of road-scene data and counts automakers such as Geely, Leapmotor and Zeekr among its customers, alongside autonomous-driving technology developer Horizon Robotics (地平线).
