Embodied Intelligence: From Frenzy To Breakthrough Through Real‑World Deployment

In the second phase of artificial intelligence development, embodied intelligence has emerged as a strategically contested domain. At the Consumer Electronics Show (CES), robots were ubiquitous—vacuum cleaners, lawn mowers, agricultural machines and numerous humanoid platforms were on display.

The sector currently presents a stark duality: major corporations and investors are concentrating capital on leading players under intense public scrutiny, while many smaller firms face funding shortfalls and some formerly high‑profile companies confront existential crises. After an intense period of competition, the focus is shifting from securing financing and showcasing technical prowess toward achieving practical deployment and sustainable profitability.

Capital Influx And Divergent Outcomes - The embodied intelligence track in 2025 displayed both extraordinary investor enthusiasm and pronounced concentration of capital. According to IT Juzi, by December 21, 2025, more than 600 investors had committed capital to the sector, with over 304 financing events totaling RMB 37.9 billion—approximately 2.95 times and 4.05 times the financing frequency and amount of 2024, and 7.24 times and 4.37 times those of 2023. Some companies, such as Songyan Power, completed six financing rounds within a single year, advancing rapidly from Series A to Pre‑B+.

Large technology groups have been particularly active. Baidu Ventures, Lenovo Ventures, Ant Group and other strategic investors together executed 62 investments during the year, with Baidu Ventures leading at 13 deals and Lenovo Ventures/Lenovo Star following with 11. Guoxiang Capital (SenseTime) and Ant Group each recorded eight investments, forming a primary investment tier. Tencent led an initial investment in Zhiyuan’s humanoid program and quickly increased its exposure, later joining Unitree’s shareholder base—evidence of its focused interest in embodied intelligence. Meituan’s founder deployed capital through strategic and venture platforms, making eight investments since 2023 and concentrating activity in the first half of 2025, while JD.com  executed six investments within three months to build capabilities across bodies, components and algorithms, with an estimated aggregate commitment near RMB 4 billion.

This intense backing reflects a simple strategic thesis: embodied intelligence is viewed as a potential transformative terminal for AI, and major players are racing to secure positions. Yet capital concentration has produced clear stratification. Data compiled by the Touzhong Jiachuan team show that among 168 funded embodied intelligence firms in 2025, the top ten recipients—including Zibianliang, Tashi Zhihang, Leju Robotics, Xingdong Jiyuan, Xinghaitu and Zhongqing Robotics—received RMB 13.472 billion, representing 40.95% of total sector financing. As funding concentrates at the top, mid‑tier companies face narrowing financing windows, accelerating market differentiation. Several smaller firms have gone two to three years without fresh capital, struggling under simultaneous R&D and production pressures.

Notable examples include Yunji Technology, which once held a 6.3% share of China’s robot service market and whose “Run” series became a hotel standard, but which has lacked new financing for over three years since its Series D. CloudMinds, a humanoid unicorn that raised more than RMB 5.4 billion historically, has not secured new investment since 2023 and now confronts significant legal and enforcement actions. Capital has shown a readiness to withdraw where business models and path to commercialization are unclear; GSR Ventures partner Zhu Xiaohu’s exits from multiple humanoid investments in early 2025—citing unclear commercial models—illustrate a broader shift toward investment discipline. Meanwhile, leading firms that demonstrate orders and revenue growth continue to consolidate advantages: Unitree secured nearly RMB 1.2 billion in orders, Zhiyuan exceeded RMB 1 billion in sales, and Galaxy General recorded over RMB 700 million in orders, providing tangible evidence of commercial traction.

From Demonstration To Commercial Viability - The central question for embodied intelligence is practical deployment—“where are the customers?” Demonstrations of advanced capabilities serve as proof of concept and a foundation for iterative improvement, but commercial value accrues when systems reliably perform productive tasks in real environments. Public demonstrations—such as humanoid robots performing complex movements—generate attention, yet industrial deployments that automate repetitive, mission‑critical tasks deliver measurable cost savings and operational benefits. For example, a humanoid deployed in a battery manufacturer’s production line that autonomously inspects harness connections and adjusts insertion force can reduce labor costs and material damage by substantial amounts annually.

This transition from technical showmanship to revenue generation is the defining leap. As Wang Xiang, former president of Xiaomi and partner at Gaoshan Xinyu, observed, technological sophistication does not guarantee commercial success; identifying genuine demand and suitable application scenarios is essential. Humanoid robots aim for general‑purpose adaptability but face high costs and technical complexity, so early adoption is concentrated in structured industrial settings such as automotive manufacturing. Zhiyuan’s A1 humanoid has undergone assembly and material‑handling trials at BYD, and the company reports cumulative orders exceeding RMB 100 million from customers including Joyson Electronics, Longcheer Technology and Chery’s supply chain. Wheeled and quadruped platforms, by contrast, have achieved large‑scale commercial use in cleaning and logistics, with companies like Gaussian Robotics deploying fleets across dozens of countries and accumulating extensive operational data.

Real‑world deployment is now a primary competitive moat. Successful projects generate revenue and create a virtuous cycle of data, improved algorithms and product refinement. However, scaling deployments across diverse, unstructured environments remains technically demanding. Field tests reveal persistent gaps: humanoid systems still exhibit nontrivial error rates in crowded or complex industrial settings, and the so‑called “uncanny valley” constrains household adoption due to trust and comfort issues. Sensor noise, actuation latency and imperfect force control continue to produce failures in seemingly simple tasks, underscoring the gap between laboratory performance and open‑environment robustness.

Addressing The Last Mile Requires Ecosystem Coordination - High unit costs—ranging from tens of thousands to hundreds of thousands of dollars—limit both industrial roll‑out and household adoption. Tesla’s Optimus and Boston Dynamics’ Atlas exemplify the cost intensity of humanoid platforms, driven by expensive sensors, actuators and custom components. Domestic manufacturers have reduced costs through localization, but achieving a step‑change in price‑performance requires industry‑wide standardization and supply‑chain coordination. Fragmented interfaces, inconsistent data formats and divergent performance metrics prevent scale effects; common components account for a minority share of parts, constraining procurement efficiencies.

Standardization is therefore a strategic priority. The Ministry of Industry and Information Technology established a Humanoid Robot and Embodied Intelligence Standardization Technical Committee in December 2025, initiating work on 12 core standards covering environmental perception, operational safety and performance evaluation, with the first set of norms planned for 2026. Industry collaborations are already producing results: Unitree’s coordination with key component suppliers to standardize servo motor interfaces reduced procurement costs by 35% and shortened lead times by 40%, enabling lower‑cost educational models. The National Humanoid Robot Innovation Center’s “White Tiger” dataset unified data formats and shared 150,000 hours of real‑world data across multiple training sites, cutting model‑training costs by 70%. International players have also engaged in standardization efforts, contributing to global supply‑chain alignment and cost reduction.

With capital, industry and policy increasingly aligned, the embodied intelligence sector remains a focal point of investment and innovation, but it is maturing toward a more disciplined, deployment‑oriented phase. Achieving leadership will require sustained technical progress, patient capital, coordinated standards and collaborative ecosystems that lower costs and accelerate real‑world adoption. Ambitious production targets announced by global players underscore the intensifying competition, while domestic firms that demonstrate scalable deployments and robust supply chains are building defensible positions. The ultimate survivors will be those that combine visionary ambition with pragmatic execution—advancing core technologies while delivering tangible value in real operational contexts. The future of China’s embodied intelligence industry will be written in each technical breakthrough, every successful deployment and every instance of disciplined strategic thinking.