Embodied AI Firms Jointly Invest in Data Infrastructure Startup Kaiwang Data
Kaiwang Data, a full-chain AI data infrastructure company, announced the completion of a new round of strategic financing exceeding 100 million yuan. The round was led by Beijing Yizhuang Industrial Upgrading Fund, Huafang Capital, and Skybound Capital, with participation from Xinding Capital, Myriad Asset Management, Xinghua Dingli, Huazhang Investment, and Lundachuanrun. Existing investors including Yasheng Capital, Qingzhi Capital, and Puyue Capital also added their investments.
Notably, several leading embodied AI companies, including Luming Robot, Songyan Power, Self-Model Robot, and Zhiyuan's affiliated company Mifeng Technology, joined as strategic investors for the first time. This collective move underscores the critical importance of high-quality physical-world training data, which is becoming the most scarce resource constraining industry development.
Why Kaiwang Data? The founding team of Kaiwang Data brings a cross-disciplinary background spanning internet platforms, autonomous driving algorithms, and automotive companies. Founder and CEO Yu Xu was the fourth employee of Uber China and the first builder of the data production and operations system at Momenta and ByteDance, with extensive experience in building big data platforms from scratch. Co-founder and COO Ren Dandan has experience in operations processes at Mercedes-Benz and in setting up overseas innovation projects. Core members come from leading companies such as ByteDance, Alibaba, Uber, and Momenta, forming a deep competitive moat in AI data services.
Founded in 2022 and headquartered in Beijing, Kaiwang Data positions itself as a 'data service middleware,' focusing on the entire value chain of data definition, scenario collection, annotation processing, and data trading, covering autonomous driving, world models, and embodied AI. The company has established a complete business loop from data definition to standardized trading, offering both standardized products and customized delivery services. In the embodied AI direction, it covers multimodal data types such as first-person bare-hand, tactile sensing gloves, UMI grippers, and third-person panoramic video, with data collection in real environments including supermarkets, restaurants, homes, and automotive production lines. Monthly effective output of physical interaction data for embodied AI and world models is approaching 100,000 hours, breaking the previous industry threshold of 10,000 hours. Additionally, it has obtained the Grade B qualification for geographic information engineering surveying, the first among domestic data annotation companies, building a differentiation in compliance and data accuracy.
The financing structure is a unique outcome of the data bottleneck, turning the supply chain into shareholders. According to estimates from Guojin Securities and other institutions, the global high-quality real physical interaction data totals only about 500,000 hours, while training general embodied models requires at least tens of millions of hours. QYResearch predicts the embodied AI data market will grow from over $1 billion in 2025 to nearly $9 billion by 2032, with a CAGR exceeding 36%. Controlling the production and circulation of high-quality physical data becomes key to next-generation intelligent agents. Previously, embodied AI companies relied on self-built collection teams, which were costly and limited in coverage. Data service providers standardize collection, annotation, and trading, turning one-off collection into a standardized industrial division. This is why companies like Luming, Songyan, Self-Model, and Zhiyuan chose to invest rather than simply purchase, participating deeply in capacity building while sharing the dividends of expansion. For Kaiwang Data, the investor list is also a customer list; for the embodied AI firms, it signals a collective judgment on the next competitive focus.
The funds will be directed to three major areas: developing an integrated data trading platform, recruiting specialized talent for world models, and expanding large-scale data capacity in real scenarios across commercial, industrial, and home sectors. CEO Yu Xu stated, 'The AI industry is not a solo effort; it requires government guidance, capital support, and deep collaboration among data service providers, algorithm companies, and physical scenario operators to realize the full flywheel from data to intelligent robot deployment.' The company aims to build a healthy AI ecosystem with all investors and partners, supporting domestic world models and embodied AI in global competition.