Recently, Spirit AI (千寻智能), a leading embodied intelligence enterprise, completed two consecutive funding rounds totaling nearly 2 billion CNY.
The investor lineup in this round represents a major concentration of industry capital: ultra-first-tier institutions such as Yunfeng Capital (云锋基金), a leading state-owned investment institution, Chaos Investment (混沌投资), and HongShan (红杉中国) have made heavy entries. They are joined by industrial capital including Synstellation Capital, TCL Ventures (TCL创投), and Minghui Zhiyuan (明荟致远); state-owned capital such as the Chongqing Industrial Investment Funds (重庆产业投资母基金) and Hangzhou Finance and Investment Group (杭州金投); and strategic investors like the 360 Fund (360基金) and Houxue Capital (厚雪资本), forming a comprehensive empowerment structure across top-tier, industrial, state-owned, and strategic capital.
Notably, existing shareholders including Shunwei Capital (顺为资本), Prosperity7, Fortune Capital (达晨财智), Borui Capital (柏睿资本), HighLight Capital (弘晖基金), Huatai Zijin Investment (华泰紫金), Oriental Jiafu (东方嘉富), Fibonacci Capital (千乘资本), and GF Xinde Investment (广发信德) all opted for significant follow-on subscriptions. Their continued investment demonstrates firm confidence in the company's technical roadmap and development prospects.
Industrial Synergy Behind the CNY 2-Billion Windvane
As the large model reconstructs the digital world, the "embodied" wave of artificial intelligence penetrating the physical world is becoming the core narrative of the global tech industry. Data from the China Academy of Information and Communications Technology shows that in 2025, total financing in the domestic embodied intelligence sector reached 73.543 billion CNY, with over 740 investment events, and the sector's popularity continues to climb. However, alongside the influx of capital, the market is also rapidly differentiating. From concept hype to value realization, enterprises truly possessing technological moats and industrial scenario penetration are becoming scarce targets contested by capital.
Spirit AI's current financing round is a concentrated reflection of this trend. This is not just financial support, but a "full-ecology assembly" covering top-tier, industrial, state-owned, and financial capital—top-tier capital bets on technological foresight, industrial capital injects scenarios and industry chain resources, and state-owned capital provides long-term stable support. The collective alignment of diverse capital is essentially a high recognition of Spirit AI's comprehensive strength.
At the level of industrial capital layout, Spirit AI has secured joint bets from diverse industrial capital and built an "all-scenario ecology" unique to the embodied intelligence track. From industrial manufacturing leaders like CATL (宁德时代), Inovance Technology (汇川技术), and TCL, to logistics, retail, and financial infrastructure represented by JD.com (京东) and China Merchants Venture (招商局创投), and consumer electronics giants Huawei (华为) and Xiaomi (小米)—the industrial shareholders have achieved cross-boundary coverage of the real economy's core tracks. This means that Spirit AI is not only a leader in technological routes but also a scarce target favored by top industrial capital. As competition in embodied intelligence moves from the laboratory into industrial depths, Spirit AI has already locked in the strategic advantages of scenario entry, data sources, and collaborative implementation by relying on its broadest industrial shareholder layout, building an irreplicable core competitiveness for large-scale implementation.
Dual Engines: Technology and Data Barriers
In January 2026, Spirit AI's open-source Spirit v1.5 model became the first Chinese open-source model to surpass Pi0.5 in performance. The model possesses powerful zero-shot generalization capabilities, able to complete complex tasks such as wiping objects and manipulating hinges and flexible objects without new sample training, demonstrating efficient execution across scenarios. Behind this breakthrough is Spirit AI's long-term deep cultivation of its technological path. Spirit AI co-founder Gao Yang stated: "We adhere to the 'data pyramid' training philosophy, avoiding the old path of predicting every frame with traditional 'world models' during the pre-training phase—that approach consumes massive computing power and is inefficient. We chose to pre-train based on massive human internet videos, achieving better results with fewer parameters and significantly reducing computing costs."
Regarding data collection: "Through our self-developed equipment, we reduced data collection costs by 90%, which enabled the large-scale implementation of massive real-world data. It is the superposition of these technical capabilities that made Spirit v1.5 the world's first open-source base embodied model to surpass Pi0.5 in performance." Spirit AI firmly advances its Scaling route with "diversity" as the core. To date, it has accumulated over 200,000 hours of multi-type real interaction data, covering dimensions such as internet videos, teleoperation, and wearable collection, with the total expected to exceed 1 million hours in 2026. Its self-developed wearable data collection equipment has iterated to the 5th generation, reducing data costs to 1/10 of traditional methods and clearing obstacles for large-scale collection.
More crucially, Spirit AI proposed a counter-intuitive viewpoint: Dirty data is the key to scaling VLA models. By training on diverse "imperfect data," the team discovered a Scaling curve with a steeper slope—data diversity is far more valuable than "cleanliness" itself. This data philosophy highly aligns Spirit AI's focused VLA technology route with global top-tier companies like Google DeepMind and Pi, validating the foresight of its technological layout.
Commercialization: The "Capability Output" Model
Spirit AI has achieved strong breakthroughs in commercial implementation: the world's first humanoid embodied intelligence production line has been put into operation at CATL's Zhongzhou base. The Xiaomo (小墨) robot serves as the core equipment on the production line, achieving zero-fault mass production for nearly a thousand batteries. Its operational tempo is comparable to or even faster than skilled workers, demonstrating millisecond-level rapid adaptation to production uncertainties and flexible operational capabilities that surpass humans. Relying on a 100% self-developed full-stack technology moat and massive process data accumulation, Spirit AI has not only established industry barriers but also pushed "capability output" to the extreme: shifting from simply selling hardware to exporting full-stack capabilities in perception, decision-making, and execution. This allows for rapid customization to adapt to different production lines and achieve global large-scale replication. In the future, it will cover various battery types and process segments, thoroughly reshaping the commercial boundaries of new energy manufacturing.
In the commercial sector, the partnership with JD.com further expands these boundaries. The Mozi (墨子) robot is replacing humans in retail scenarios for interactive explanations and product operation demonstrations. Both parties are jointly exploring the implementation potential of JD Cloud and the Joyinside large model within large-scale retail networks, achieving two-way empowerment of technological R&D and application scenarios.
Behind these commercialization achievements lies Spirit AI's profound accumulation of core capabilities. The industry consensus is that the competition in embodied intelligence is essentially a competition among top-tier model teams. Spirit AI possesses what is known as the "irreplicable AI dream team"—members mostly come from top universities like UC Berkeley, Tsinghua, and Peking University. With an average age of under 30, they already possess deep academic and engineering genes in core areas of embodied models such as multimodal large models, robotics, reinforcement learning, and imitation learning. Not only do they continuously output core achievements in the international open-source community, but they also have the full-link capability to bridge "models, algorithms, and hardware." It is exactly this young and scarce team that provides the technological confidence for Spirit v1.5 to surpass Pi0.5. Investors have commented: "Spirit AI's team configuration is benchmarked against that tier of Pi and Google DeepMind."
Dong Huaijin, Executive Director of Yunfeng Capital, stated: "An AI brain exclusively for the physical world is an important prerequisite for the leap in embodied intelligence, and the core barrier across cycles comes from the actual creation of productivity by technology. As a new-generation embodied intelligence scientist, Gao Yang leads the creation of AI foundation models and self-developed core equipment, efficiently and high-quality driving the flywheel upgrade of 'diverse data - foundation models - real scenarios,' and iterates the closed loop in actual industrial scenarios based on full-stack software and hardware integration capabilities. We look forward to Spirit AI continuing to deeply cultivate Physical AI foundation models, driving the large-scale implementation of industrial scenarios and the release of collaborative value with its technological base."
Geng Yixuan, Partner at Shunwei Capital, stated: "Shunwei has long focused on the cross-innovation of AI and hard tech. Spirit AI is the core project we discovered earliest and continuously backed across multiple rounds in this direction. The team combines top academic heritage with practical industry genes. What's even more commendable is that they chose a 'difficult but correct' path from the very beginning—building barriers with the 'data pyramid' philosophy and breaking industry data bottlenecks with low-cost collection. It is exactly this pragmatic innovation path that allowed Spirit AI to be the first to complete the positive cycle of 'scenarios feeding back into the model.' We firmly believe that this young dream team is expected to become the backbone defining the next generation of embodied intelligence."
Following this financing round, Spirit AI will continue to increase its investment in embodied foundation models and real data systems, deepening the co-construction of the industrial ecology. Spirit AI firmly believes that embodied intelligence will become the core infrastructure of the next decade. When machines truly possess the ability to understand the physical world, human-machine collaboration, production processes, and even social operational efficiency will usher in profound transformations. Taking this financing as an opportunity, the company will continuously deepen technological innovation and industrial implementation, joining hands with industrial partners to build a new ecology for embodied intelligence, and injecting new momentum into industry development and social progress.