Yunfeng and Sequoia Lead $2 Billion Investment in Embodied AI Leader: Spirit AI Soars to $10 Billion Unicorn Status

The embodied intelligence sector has started the year with a major financing announcement. On February 24, leading embodied AI company Spirit AI (Spirit AI) announced the recent completion of two funding rounds totaling nearly 2 billion yuan, setting a new record for fundraising in the embodied intelligence field.

The investor lineup for this round represents a gathering of industry heavyweights. Top-tier institutions including Yunfeng Capital, a leading state-owned institution, HunTun Capital (managed by Ge Weidong), and Sequoia China made significant investments. Industrial capital was also heavily involved, with participation from Synstellation Capital, TCL Ventures, and Minghui Investment (the family office of Zhu Xingming, Chairman of Inovance Technology). State capital provided strong support from entities like the Chongqing Industrial Investment Mother Fund and Hangzhou Financial Investment. Strategic investors such as 360 Fund and Houxue Capital also participated, creating a comprehensive empowerment structure covering top capital, industrial giants, state capital, and major strategic investors.

Notably, all existing shareholders, including Shunwei Capital, Prosperity7, Dachen Caizhi, Bairui Capital, Honghui Fund, Huatai-PB, Orient Jiafu, Qiansheng Capital, and GF Xinde, opted for substantial additional subscriptions. This active follow-on investment demonstrates their firm confidence in the company's technological roadmap and development prospects.

Founded in January 2024, Spirit AI is an embodied intelligence company focused on building a "general brain" for robots. The company's founder and CEO, Han Fengtao, is a serial entrepreneur in the robotics industry. Co-founder Gao Yang, one of the "Berkeley Returnee Quartet," is also an assistant professor at the Institute for Interdisciplinary Information Sciences at Tsinghua University. From the perspective of utilizing internet video and pre-trained VLMs, Gao Yang proposed the ViLa and CoPa models in 2024. In May 2025, his team introduced the OneTwoVLA model, achieving a new breakthrough.

The combination of "industry veterans and top scientists" has given Spirit AI a first-mover advantage in both cutting-edge technology exploration and commercialization. Team members primarily hail from top institutions like UC Berkeley, Tsinghua University, and Peking University, with an average age under 30. Despite their youth, they possess deep academic and engineering expertise in core areas of embodied models, including multimodal large models, robotics, reinforcement learning, and imitation learning.

"Spirit AI has recently made progress exceeding expectations at the model level, while also achieving successful validation in mass-production scenarios within CATL's factories. These dual breakthroughs in technology and commercialization made this funding round exceptionally competitive, with investment allocation essentially in a state of high demand. Many investors who missed out are already queuing up hoping to participate in the next round," revealed a source familiar with the matter.

With the successful closing of this round, Spirit AI's valuation has surpassed the 10-billion-yuan mark, establishing it as another unicorn in the embodied intelligence sector. As leading companies enter the "10-billion-yuan club," the embodied intelligence track is moving from an era of "a hundred schools of thought contending" into deeper, more competitive waters. Only companies possessing original technological strength, commercial viability, and sustained capital support can lead the evolving landscape and seize the initiative in this new phase. Representative firms like Spirit AI are sailing into deeper blue oceans.

The backing from a top-tier capital consortium brings financial, industrial, and strategic resources. Following the large model-driven restructuring of the digital world, the wave of "embodiment," where AI penetrates the physical world, has become a global tech industry focus. Data from the China Academy of Information and Communications Technology shows that total financing in China's embodied intelligence sector reached 73.543 billion yuan in 2025, with over 740 investment events, making it one of the hottest investment sectors that year.

The momentum in embodied intelligence has continued into 2026. In January and February, two embodied robotics companies announced new financings of 1 billion yuan each. Meanwhile, new entrants are emerging, with DaXiao Robotics, under SenseTime, announcing the completion of its angel round on February 10. Spirit AI's funding round has pushed sector enthusiasm to a new peak. As the largest announced financing in embodied intelligence so far in 2026, the 2-billion-yuan figure represents a concentrated bet from top-tier capital, industrial capital, state capital, and financial institutions.

Firstly, the round attracted investment from globally renowned venture capital firms Yunfeng Capital and Sequoia China. It is noteworthy that Yunfeng Capital has significantly accelerated its布局 in embodied intelligence this year, having successively invested in companies like embodied intelligence data platform MiFeng Tech and dexterous hand company Tipping Point. Sequoia China is one of the earliest and most extensive investors in the embodied intelligence space, having backed several star companies in robotics, including Unitree, Astribot, Mech-Mind, Hai Robotics, Agile Robots, and Fourier Intelligence.

"Embodied intelligence is a long-term sector with a cycle of at least ten years," stated Sequoia China partner Zhang Han at a recent public forum, indicating an investment strategy favoring early-stage teams with strong entrepreneurial determination and core competencies.

A top-tier investor familiar with Spirit AI commented that the internal approval processes at firms like Yunfeng and Sequoia are extremely rigorous. Their ability to make quick decisions and complete the transaction at this juncture inherently reflects a high recognition of the long-term value of Spirit AI's VLA (Vision-Language-Action) technical roadmap and confidence in its commercialization capabilities.

Concurrently, the round also attracted additional investment from industrial capital players like Synstellation Capital, TCL Ventures, and Minghui Investment. A review of Spirit AI's previous funding rounds reveals that it has become an embodied intelligence startup simultaneously backed by diverse industrial capital, gradually building a unique "full-scenario ecosystem" within the sector. Its industrial shareholders now provide cross-sector coverage of core real-economy fields, ranging from industrial manufacturing leaders like CATL, Inovance Technology, and TCL, to logistics, retail, and financial infrastructure represented by JD.com and China Merchants Capital, and consumer electronics giants Huawei and Xiaomi.

These industrial shareholders bring far more than just financial support. As competition in embodied intelligence moves from the lab into industrial applications, Spirit AI's broad base of industrial shareholders allows it to pre-emptively secure scenario access, data sources, and strategic advantages for collaborative implementation. This fosters a positive cycle where "scenarios feed the model, and the model enhances scenario efficiency," creating a core competency difficult to replicate and crucial for large-scale business deployment.

Furthermore, the round introduced state capital from a leading state-owned institution, the Chongqing Industrial Investment Mother Fund, and Hangzhou Financial Investment. The participation of local state-owned platforms reflects the determination of local governments to treat embodied intelligence as a strategic lever for regional economic transformation and upgrading. At the national level, embodied intelligence has been included in the core scope of "New Quality Productive Forces," seen as a key carrier for the deep integration of AI and the real economy. The backing of state capital not only provides Spirit AI with long-term, stable financial support but also suggests potential strategic resource倾斜 in areas like policy support, industrial implementation, and standard setting.

Currently, a consensus is rapidly forming in the embodied intelligence field, with the initial competitive landscape taking shape. The coordinated entry of top-tier capital, industrial capital, and state capital demonstrates affirmation of Spirit AI's comprehensive capabilities. Simultaneously, the financial resources, strategic assistance, and networks brought by these different types of capital are advantageous for Spirit AI to stand out in the new round of industry competition.

As the embodied intelligence industry enters a deeper development phase, Spirit AI has built a dual technical moat based on its models and data accumulation, a key factor attracting diverse capital. In January 2026, Spirit AI achieved an industry breakthrough in model development. Its open-source model, Spirit v1.5, became China's first open-source embodied model to surpass Pi0.5. This signifies the first time a domestic embodied intelligence model's capabilities have surpassed a leading US player in public benchmarking.

Reportedly, unlike traditional models reliant on specific scenario training, Spirit v1.5 demonstrates powerful zero-shot generalization capabilities. It can complete complex skills like wiping object surfaces, operating hinged objects, and manipulating flexible objects without requiring training on new samples. This ability to "infer other cases from one instance" allows it to excel at handling diverse tasks.

"We adhere to a 'data pyramid' training philosophy. During pre-training, we avoided the traditional 'world model' approach of predicting every frame—a path that consumes massive computing power with low efficiency. Instead, we chose to pre-train based on vast amounts of human internet video, achieving better results with fewer parameters and significantly reducing computing costs," explained Spirit AI co-founder Gao Yang.

Regarding data collection, Gao Yang added, "We reduced data acquisition costs by 90% through self-developed equipment, enabling the large-scale implementation of massive real-world data. It is the叠加 of these technical capabilities that made Spirit v1.5 the world's first open-source base embodied model to outperform Pi0.5 in performance."

To date, Spirit AI has accumulated 200,000 hours of high-quality, multi-type real interaction data, covering categories like internet human videos, teleoperation, wearable device collection, and real-robot rollout. This data scale is projected to exceed 1 million hours by 2026.

The high cost and time-consuming nature of data accumulation have long been common challenges for embodied intelligence startups. Spirit AI established a technical路线 focused on the Scaling Law for data starting in July 2025 and began R&D on wearable data collection devices and validation of corresponding theories. Using this self-developed equipment, the team has now built a complete closed loop of data collection-cleaning-training, reducing embodied intelligence data acquisition costs by tenfold, to just 10% of traditional teleoperation collection methods.

More crucially, while the industry普遍 pursues high-quality data, Spirit AI proposed the counter-intuitive view that "Dirty data is the key to scaling VLA models." By training on diverse "non-perfect data," the team discovered a steeper Scaling curve—data diversity is far more valuable than mere "cleanliness."

From a global perspective, the VLA technical路线 focused on by Spirit AI aligns highly with leading global players. Google DeepMind and Pi are both committed to building general-purpose operational brains, meaning they are持续 investing in the VLA路线. Reportedly, Pi recently began demonstrating technology based on learning from human videos, corroborating that human video data is a highly scalable and valuable data source. Figure is also collaborating with US real estate developers to extensively collect large volumes of human behavior data and plans to scale data volume massively using video formats. Skilled is building a general cross-embodiment embodied intelligence brain, aiming to transcend various physical hardware forms like quadrupedal, bipedal, and wheeled bases to construct a universal brain encompassing Navigation and Manipulation.

Having坚定 invested in human video data for two years, Spirit AI holds a first-mover advantage in this data arms race. It is evident that the dual barriers of models and data form the core foundation of its entry into the embodied intelligence "10-billion-yuan unicorn club." Following this funding round, Spirit AI will持续 increase investment in its foundational embodied models and real-world data systems, deepening co-construction of the industrial ecosystem.

The year 2026 is a critical year for commercialization among embodied intelligence companies. Recently, Morgan Stanley significantly raised its forecast for humanoid robot sales in China for 2026, predicting sales of 28,000 units, double the previous forecast of 14,000 and representing 133% growth year-over-year compared to 2025.

Spirit AI is accelerating its commercialization efforts, with its collaboration with CATL becoming a benchmark case and a "golden key" for unlocking industrial mass-production scenarios. In December 2025, CATL announced the official operation of its Zhongzhou base, the world's first new energy power battery PACK production line featuring the large-scale deployment of humanoid embodied intelligence robots. The Mozi robots working on the production line were developed by Spirit AI.

It is worth noting that Spirit AI and CATL share a connection. In November 2024, Spirit AI received its Angel+ round funding from Bairui Capital, a CVC institution established with capital from Li Ping, Co-founder and Vice Chairman of CATL. After Spirit AI became part of the CATL ecosystem, the two parties further explored the commercial implementation of embodied intelligence around production lines. CATL, involving multiple departments, conducted in-depth production line research and jointly developed a forward-looking yet feasible implementation plan, laying the foundation for the scaled deployment of embodied intelligence robots.

Currently, Mozi robots are operating stably on the real mass-production line for CATL's battery PACKs, having produced nearly a thousand batteries. Through the Spirit embodied model, the Mozi robots demonstrate three key advantages on the production line: 1) Precision Adaptation: They can autonomously handle uncertainties like variations in incoming material position and connection point changes, adjusting their operation posture in real-time. 2) Flexible Operation: When plugging/unplugging flexible wiring harnesses, they can dynamically adjust force to ensure reliable connections without damaging components. 3) High Efficiency and Reliability: In actual operation, their connection success rate remains stable above 99%, and their work cycle time has reached the level of skilled workers.

The integration of Mozi robots as indispensable members of the production line marks a critical leap for embodied intelligence from "technically feasible" to "industrially reliable." In the future, Mozi robots will be scaled and replicated across different battery types, production lines, and operational processes.

Simultaneously, Spirit AI is exploring commercial deployment in diversified scenarios. For instance, in industrial settings characterized by multi-variety, small-batch, high-flexibility production requiring frequent changeovers, the company aims to enhance robot interaction capabilities during production operations using the generalization power of embodied models, achieving generalization across workpiece types, incoming material positions, and operational environments. In commercial scenarios, applications are being developed for supermarket retail, hotel services, and inspection. For home settings, the focus is on high-frequency household chores, enabling tasks like categorizing and storing everyday items, waste sorting, and sorting, folding, and storing laundry.

From task-level intelligence to general embodied models, and from simulation to real-world operation, embodied intelligence is entering a critical phase of technological industrialization. As robots truly acquire the ability to understand the physical world, moving from "being seen" in showcase demonstrations to "being needed" in daily production and life, human-robot collaboration methods, production processes, and even societal operational efficiency are poised for profound transformation.

Having built dual technical barriers in models and data and demonstrating outstanding commercial viability, Spirit AI is quietly positioned at the starting point of the convergence of this wave of technological revolution and industrial transformation.