Megvii Veterans' AI Robot Startup Secures New Funding From Nio Capital, Lenovo
A robotics startup founded by core members of Chinese facial recognition unicorn Megvii has secured a second round of financing within six months, attracting capital from prominent investors and signaling intense market appetite for companies tackling the challenges of embodied intelligence.
Dexmal established in March 2025, recently added investment firms ZhenFund, Hongtai Aplus, Lenovo Capital & Incubator Group, Xichuangtou, and Nio Capital as new shareholders, according to an update in its business registration on August 27. The company has not yet made an official announcement regarding the new funding round.
This follows a RMB 200 million yuan (approximately US$28 million) angel round announced on March 26, backed by investors including Legend Capital, J-Capital, Qiming Venture Partners, Yufu Fund, and Changsheng Investment. The rapid succession of financing highlights a rush of capital into China’s burgeoning embodied AI sector, with investors placing bets on experienced teams aiming to integrate advanced AI models with physical hardware.
The investments underscore a pivotal trend favoring startups with deep expertise in both software and hardware. Investors are increasingly backing teams that pursue a "full-stack" approach—developing both the AI "brain" and the robotic "body"—a strategy considered crucial for achieving breakthroughs in general-purpose robotics and navigating a highly competitive landscape.
A Star-Studded Team From Megvii
Dexmal is positioned as a new venture by the early architects of Megvii Technology Ltd., one of China's most well-known facial recognition firms. The startup’s leadership includes Megvii co-founder Tang Wenbin, who serves as CEO, alongside other early and core technical members from Megvii, including Fan Haoqiang, Zhou Erjin, and Wang Tiancai.
The founding team boasts a distinguished pedigree from both academia and industry. Fan Haoqiang, a graduate of the prestigious "Yao Class" at Tsinghua University, was an early employee at Megvii and rose to become the director of its research institute's algorithm team. The startup’s corporate structure further solidifies its connection to Megvii’s founders, effectively making it a "re-entrepreneurship" by some of the key minds behind the AI unicorn.
‘Full-Stack’ Technology Approach
Dexmal aims to develop general-purpose robots by focusing on a synergistic model of algorithm development, joint hardware design, and closed-loop application scenarios. The company states it is one of the few domestic startups with a founding team experienced in both computer vision and large models.
The company is developing a full, end-to-end technology stack, including a proprietary Multi-Modality Large language-action Agent (MMLA). This model integrates capabilities from large language models (LLMs), visual language models (VLMs), and world models to enable robots to generalize across different tasks and environments.
The team has already demonstrated early technical progress. Before its official incorporation, its researchers developed a model that achieved an average success rate of over 90% across hundreds of tasks on the RLBench benchmark platform. In May 2025, Dexmal won gold medals in two tracks—"Pure Tactile Manipulation" and "Tactile Sensor Design"—at the global ManiSkill-ViTac 2025 robotics challenge.