Author | Qiu Xiaofen
Editor | Yuan Silai
Exclusive information from Yingke reveals that embodied AI company "Yuanli Lingji" has recently completed a new round of financing. The investors are primarily several major large model companies, including Zhipu, Jieyue Xingchen, and SenseTime. Additionally, industrial investment parties such as Huaqin and SAIC Hengxu have continued to invest.
"Yuanli Lingji" is a general embodied large model company. It was founded in March 2025 by Tang Wenbin, co-founder and CTO of Megvii Technology. The core founding team consists of former members of Megvii Technology.
Interestingly, this financing also marks a rare "meeting" between former rivals SenseTime and Megvii amidst the embodied AI boom.
Furthermore, including Alibaba as the exclusive lead investor in the A+ round, this is a rare gathering of four domestic large model manufacturers in the embodied AI track. Previously, Zhipu had only made small-scale investments in the embodied AI field through its Z Fund, while Jieyue Xingchen had almost never invested in embodied AI.
This collective action also signals a shift: as the main battlefield of large model competition moves from Token to Action, embodied models with the ability to interact with the physical world have become the next high ground targeted by model companies.
Alongside this financing, Tang Wenbin is consolidating forces and beginning to integrate robotics assets.
Yingke exclusively learned that "Yuanli Lingji" has recently completed a merger with the logistics robotics company "Atomix" (Yuanli Juhe) through equity acquisition, aiming for large-scale deployment and global expansion of embodied AI.
The business origins of "Atomix" can be traced back to 2016—at that time, Tang Wenbin led the intelligent logistics and robotics scheduling business (Hetu System) within Megvii Technology, primarily promoting multi-form logistics robotics solutions.
In July 2024, following changes in Megvii's business, Tang Wenbin spun off the logistics robotics business from the Megvii system, establishing "Atomix" as an independent entity.
After several years of exploration, "Atomix" has achieved the second-highest global sales volume of pallet shuttle robots, cumulatively serving over 500 projects. Clients include Uniqlo, Mixue Ice Cream & Tea, CATL, etc., with annual company revenue nearing ten billion yuan.
As the embodied AI hardware supply chain matures, the industry is approaching a wall that must be scaled: the embodied brain. Compared to the clear evolution path of language models, the embodied AI model currently lacks even low-cost, massive, and high-quality data, let alone a convergent training paradigm. It can be said that the entire industry is groping in the dark.
In this situation, the integration of body, brain, and data may become the new norm in the embodied AI track.
Traditionally, the ideal state for the embodied AI industry has been to create a genuine data flywheel. However, the reality is that the industry is in a state of "data deadlock"—models need error data from real-world scenarios to evolve, but without being equipped with a good model, robots cannot enter scenes and thus cannot collect real data.
Therefore, insiders say the merger of the two companies essentially aims to close the loop between the model and the scenario, breaking the data deadlock.
As Tang Wenbin mentioned in a previous interview, Picking is the "atomic task" of the embodied AI era—Picking is to embodied AI what Coding is to large models. "Atomix" is like a continuously operating Picking data engine.
"Yuanli Lingji" Robot Making Breakfast (Source / Company)
It is understood that in the future, the real-world data generated from "Atomix's" operations across over 20 countries and 500+ projects will directly become the fuel for "Yuanli Lingji's" model training; meanwhile, the embodied AI model trained by "Yuanli Lingji" will quickly achieve collaborative operations with "Atomix's" existing robots.
This vision may not be a castle in the air but is built upon a certain technological foundation. Prior to this, "Yuanli Lingji" has already launched the general embodied large model "DM0".
Tang Wenbin mentioned in a previous interview that at the data level, "Yuanli Lingji" has completed the industry's first "integration of three types of data"—conducting mixed training on internet semantics, autonomous driving physical rules, and robotics operation data to enhance data scale and quality.
This cross-domain mixed training approach allows "DM0" to break free from dependence on specific hardware parameters. Like an experienced "veteran driver," it abstracts universal physical laws from massive heterogeneous data and can transfer across various robot body configurations regardless of hardware differences, achieving a universal operational logic.
"Yuanli Lingji" Robot Mixing Drinks (Source / Company)
More crucially, "Yuanli Lingji" also attempts to extend the "chain-of-thought reasoning" of large models into physical space—this enables "DM0" to achieve sub-millimeter precision operations with a small parameter scale of just 2.4B, significantly improving success rates in long-horizon continuous tasks.
Through a series of combinations, "DM0" is attempting to break the pain points of traditional embodied models: single-source data, paralysis upon robot change, and bloated parameters.
Following this merger and financing, China's embodied AI industry is welcoming a strong player. More importantly, it also signals that the industry is entering the next phase—finding the scaling law for embodied models.
This is a formidable challenge that cannot be overcome merely by amassing robot bodies.
Just this week, media reports disclosed that ByteDance is heavily recruiting a Head of Embodied AI Technology, targeting core technical talents from leading startups. Meanwhile, overseas embodied AI star company Skild AI just completed the acquisition of Zebra Technologies' robotic automation business.
The moves of giants at home and abroad are strikingly similar—as body manufacturers, data asset holders, model developers, and scenario operators begin to accelerate their convergence, the industry has officially entered deep waters.
Homepage Image Source | Network
Layout | Fan Xinya
Welcome to Connect
