Embodied intelligence company MouShen Intelligence has announced the completion of a nearly RMB 500 million Pre-A+ financing round. The investment was led by a consortium including Shenbao Yiben Fund, Orient Securities, Shaanxi Provincial High-Tech Industry Investment Co., Ltd., along with Anyu Fund, Tianmeng Investment, and Jianyuan Tianhua, while existing shareholders Chuanghehui Capital, Xuhui Capital, and Gengxin Capital continued to increase their stakes.
MouShen Intelligence stated that the proceeds from this round will primarily fund research and development in areas such as robot-native large models, on-device intelligence, and domestic computing power adaptation, as well as support the industrial deployment of embodied intelligence products. According to company disclosures, its valuation in the first half of 2026 had grown more than tenfold compared to the beginning of the year.
Public records indicate that since its founding less than two years ago, MouShen Intelligence has sequentially completed an angel round, an angel+ round, a RMB 300 million Pre-A round, a follow-on Pre-A round of nearly RMB 100 million, and the current Pre-A+ round. The composition of investors in this round includes local state-owned capital, brokerage-affiliated funds, and industrial investment institutions.
Several investment institutions believe that as robot hardware matures, the competitive focus in the embodied intelligence sector is shifting from hardware manufacturing to the closed-loop capabilities of models, data, and application scenarios. A general-purpose robot "brain" capable of cross-platform deployment is expected to become a critical infrastructure for large-scale robotic applications.
In terms of technical approach, MouShen Intelligence has not fully adopted the industry's mainstream vision-language-action (VLA) model pathway. Instead, it places action modeling at its core, exploring a "World Motion Model" that aims to enhance robots' execution and generalization capabilities in complex environments through the understanding, generation, and control of actions.
Previously, MouShen Intelligence launched the latent-space action diffusion model MLD and MotionGPT. MotionGPT decomposes robot actions into approximately 3,000 "action primitives" and generates complex behaviors by predicting tokens in a manner similar to large language models. In 2026, the company further introduced the spatiotemporal-integrated world action model STI-WM, which seeks to unify the modeling of spatial structure, temporal evolution, physical consistency, and execution robustness.
Training data is composed of internet videos, motion capture data, and real robot data, aiming to reduce the model's reliance on costly real-world robotic datasets. Shaanxi High-Tech Investment noted that the core competitiveness of embodied intelligence companies is now shifting toward the efficiency of the "model-data-scenario" loop. MouShen Intelligence has built up certain expertise in action representation, generation control, model compression, and on-device inference, though it still needs to further strengthen its data flywheel and product standardization capabilities.
On the commercialization front, MouShen Intelligence disclosed that it has completed deliveries of robot products to listed property management companies and leading A-share sanitation companies. It is also advancing humanoid robot application development with top retail groups and white goods manufacturers.
Financial data shows that MouShen Intelligence's audited billing revenue for 2025 reached the tens of millions of yuan level. In the first half of 2026, billing revenue had already reached RMB 30 million, and the company projects full-year revenue to surpass RMB 50 million.
CEO Mu Zelin of MouShen Intelligence stated that the company will promote native development based on domestic computing power chips throughout the entire model training and inference pipeline, further improving software-hardware synergy efficiency and the autonomy of its technology stack. According to the company's positioning, MouShen Intelligence aims to build a general-purpose embodied intelligence system adaptable to different robot platforms, reducing the costs enterprises incur from repeatedly developing algorithms for different machine types and industry scenarios.
Following the completion of this financing round, MouShen Intelligence stated it will continue research and development around robot-native large models, on-device deployment, cross-platform adaptation, and the domestic computing power ecosystem, while advancing the transition of these technologies from single-project validation to large-scale industrial applications.