Boden AI博登智能科技
LatestA++ (several hundred million RMB across A+ and A++ completed in Jun 2026)
Undisclosed (RMB 700M at Series A; latest to be confirmed)
Ningbo Boden Intelligence Technology Co., Ltd.
A leader in Physical AI data infrastructure, 'Train at Scale, Validate in Reality'; the BASE automated data annotation platform; three embodied robot innovation centers producing 500,000 hours of real-machine data per year; order backlog near RMB 1B
数据截至2026-08-19
About
Boden AI was founded in Ningbo in 2019. Founder Zhao Jie holds a PhD from the University of Kaiserslautern in Germany and was a senior core-algorithm engineer at a German automaker. Starting from autonomous-driving data annotation, the company positions itself as a Physical AI data infrastructure company (benchmarking the US Scale AI). Its core product, the BASE automated data annotation platform, has iterated to its sixth generation with 200+ built-in pre-labeling models, delivering up to 7x efficiency gains, over 99% overall accuracy and about 40% cost reduction. It has built three embodied robot innovation centers in Ningbo, Huzhou and Ma'anshan (over 30,000 square meters in total, with 500+ robots and thousands of self-developed Ego collection devices), producing 500,000 hours of real-machine training data per year. The business mix is about 50% embodied intelligence, 30% autonomous driving and 20% large models, with nearly 500 enterprise partners cumulatively, a roughly 104% three-year average growth rate and PB-scale multimodal data. It closed a RMB 100M Series A in May 2025 (Guohe Investment, RMB 700M valuation) and several hundred million RMB across A+ and A++ rounds in June 2026; it is a national-level specialized 'Little Giant' and participates in over 20 national and industry standards.
Technical Approach
'AI for AI' technology philosophy: AI algorithms deeply participate in data processing (weak supervision + self-supervision + active learning), cutting reliance on manual labeling by over 70%, with the automated labeling ratio above 60% in intelligent scenarios (a 90%+ mid/long-term target). It builds a three-layer capability ecosystem of a 'real-world scenario network + fully automated data engine + real-world validation system', forming a full-chain loop of collection (BRIC) → annotation (BASE) → management (Blink) → training → validation; drawing on autonomous-driving experience, it combines real-scenario data with synthetic data to handle long-tail effects.
Commercialization（7）
- Order backlog near RMB 1B, planning to deliver about RMB 500M within the year (orders require a 20-30% prepayment)
- Business mix: about 50% embodied intelligence, 30% autonomous driving, 20% large models; roughly 104% three-year average growth
- Nearly 500 well-known domestic and international enterprise partners (including research institutions); PB-scale multimodal data, one of the largest data-training companies in Zhejiang
- Core customers span automakers Geely, Leapmotor, Zeekr and Seres; tech giants ByteDance, Tencent, Alibaba, NVIDIA and SenseTime; China Unicom; and Midea (humanoid robot data collection and annotation)
- Open-sourced the RW-RL-Dataset jointly with PIA Automation (均普智能) and SJTU's MINT lab (1,000+ hours of real robot reinforcement learning data)
- The Ma'anshan embodied intelligence training ground phase 1 entered operation in March 2026, planned to become one of China's leading humanoid robot data factories
- Named a national-level specialized 'Little Giant', a 2025 national high-tech enterprise and a 2026 national AI high-quality dataset pilot consortium; participates in over 20 national and industry standards
Position & Strengths（6）
- Positioning upgraded from 'data service provider' to core infrastructure for the Physical AI era, benchmarking the US Scale AI (valued at tens of billions of dollars)
- A scarce heavy-asset moat: three embodied robot innovation centers + 500+ robots + thousands of self-developed collection devices, producing 500,000 hours of real-machine data per year — hard to replicate short-term
- AI-for-AI automation engine: 60%+ automated labeling (90%+ target), 7x efficiency, 99% accuracy, 40% cost reduction
- Full-chain loop: collection (BRIC) → annotation (BASE) → management (Blink) → training → validation (Boden Cloud), one-stop delivery
- Standard-setter status: participates in over 20 national/industry standards, moving from 'technology provider' to 'standard setter'
- Judges that autonomous driving and embodied intelligence share highly similar data technology logic (wheeled robots being a subset of embodied intelligence), giving clear transfer advantages