In the first half of this year, major internet giants have been overtly announcing their forays into the embodied intelligence arena. In contrast, ByteDance has maintained a notably low profile. However, this quiet demeanor does not signify absence. For this domain, widely regarded as a future gateway, ByteDance is, in fact, executing a stealthy strategy, advancing on both the software and hardware fronts to secure its position in the next wave of intelligent systems.
On September 2, Stardust Intelligence (Shenzhen) Co., Ltd. announced a strategic partnership with Xian Gong Intelligence, planning to deploy over a thousand AI robots in phases over the next two years across industrial manufacturing and warehouse logistics scenarios. As of September this year, only a handful of players in the embodied intelligence field, such as Yushu Technology, Zhiyuan Robotics, and Songyan Power, have secured “thousand-unit” level orders. Notably, the investment backing Stardust Intelligence includes not only prominent institutions like Ant Group, Yunqi Capital, and Daotong Capital but also the Jinqiu Fund, which has deep ties to ByteDance.
According to information, ByteDance is one of the many limited partners of the Jinqiu Fund. Established by Yang Jie, former head of ByteDance’s financial investment department, the Jinqiu Fund takes its name from ByteDance’s first office location, “Jinqiu Home.” Its core team largely hails from ByteDance’s investment department and other top-tier investment institutions and industrial firms. This connection underscores ByteDance’s indirect yet strategic involvement in advancing embodied intelligence technologies through targeted investments.
- ByteDance’s Quiet Foray into Embodied Intelligence
This year, ByteDance has been quietly making moves in the embodied intelligence sector. From investment layouts to self-developed robots, and from large models to hardware carriers, its integrated software-hardware strategy is gradually becoming clear. Data from the IT Juzi platform reveals that from April to August this year, the Jinqiu Fund has invested in five embodied intelligence enterprises: Stardust Intelligence, Incus, Enjoy Technology, Yushu Technology, and Pompote.
Furthermore, according to reports from LatePost, ByteDance’s self-developed robots have entered the mass production stage. Since 2023, the robotics team under ByteDance’s AI Lab has been pushing forward with robot mass production. Currently, robots developed by ByteDance have cumulatively produced over a thousand units. The form of these mass-produced robots is wheeled logistics robots, lacking robotic arms for sorting, primarily used for transporting packages and parts in warehouses and production lines, mainly serving ByteDance’s own businesses such as Douyin e-commerce warehouses.
In response to inquiries, no reply was received from ByteDance at the time of reporting. Indeed, ByteDance’s preparations for this hardware “expedition” extend beyond production capacity to include models. In July this year, ByteDance’s Seed team launched a new Vision-Language-Action model—Seed GR-3. According to the Seed team, this model can understand abstract instructions, manipulate flexible objects, and possesses strong generalization capabilities. Unlike previous VLA models that relied heavily on large amounts of robot trajectory training, GR-3 requires only a small amount of human demonstration data for fine-tuning to adapt to new tasks, significantly reducing deployment costs.
Moreover, thanks to improved model architecture, GR-3 can effectively handle long-horizon tasks and perform high-dexterity operations, including bimanual coordination, flexible object manipulation, and full-body operations integrated with base movement. To fully leverage GR-3’s capabilities, ByteDance has simultaneously developed the universal dual-arm mobile robot ByteMini. This robot has 22 degrees of freedom, adopts a unique wrist ball joint design, and can perform bimanual coordination, precise grasping, and movement operations in confined spaces. ByteMini, as the “body,” equipped with the GR-3 “brain,” forms a complete embodied intelligence solution.
ByteDance has been vigorously promoting AI development this year. For large models to be widely applied across various industries, they not only need a powerful “brain” but also must find a “hardware body.” While advancing self-developed technologies, ByteDance is also expanding its hardware layout through ecosystem resources. It is understood that at the Volcano Engine Primordial Power Conference in June, intelligent robot devices from third-party partners like Gizwits, Quectel, and Songyan Power were all exhibited in the display area.
Currently, industry consensus is increasingly leaning towards “integrated software and hardware.” ByteDance has quietly positioned itself across multiple layers—technology, investment, manufacturing, and ecosystem—laying ample groundwork for its long-term development in the field of embodied intelligence. Pan Helin, a renowned economist and member of the Expert Committee on Information and Communication Economy of the Ministry of Industry and Information Technology, stated in an interview that the “self-research + investment” strategy is correct.
In Pan Helin’s view, self-research enables ByteDance to possess AI+ capabilities, allowing it to use AI to empower its content platforms in the future. Investments can compensate for shortcomings, absorbing external teams’ technologies to acquire algorithms, computing power, reducers, motors, and more. The core objective is both for current business and to seize entry points. Pan Helin also pointed out that for ByteDance, its greatest opportunity in embodied intelligence lies in having direct traffic entrances, thus holding better prospects for commercial application. Its challenge, however, lies in the lack of technical accumulation in execution control. “Internet companies are better at developing robot brains, but they have shortcomings in joints, limbs, and cerebellum functions, requiring the introduction of external technologies. Another challenge is how ByteDance solves the current problem of limited application scenarios for embodied intelligence.”
- Embodied Intelligence Emerges as a New Competitive Arena
Data announced at the opening ceremony of the 2025 World Robot Conference shows that in the first half of this year, China’s robot industry’s operating revenue increased by 27.8% year-on-year, with the output of industrial robots and service robots growing by 35.6% and 25.5% year-on-year, respectively. China has been the world’s largest application market for industrial robots for 12 consecutive years, indicating that China’s embodied intelligence industry is in a stage of rapid development.
According to incomplete statistics, within the year (as of June 23), there have been 91 investment events in the domestic embodied intelligence industry, with an increase of 47 events in just over two months. IT Juzi data shows that in 2023, there were only 28 investment events in embodied intelligence, which more than doubled to 77 in 2024. As of September 5, 2025, there have been 148 investment events in the domestic embodied intelligence industry within the year, with a total investment amount of 18.418 billion yuan.
This year, embodied intelligence has become a new track fiercely contested by internet giants, with companies like JD.com, Alibaba, and Meituan noticeably accelerating their pace, particularly JD.com, which has shown the most vigorous momentum. From May to July, JD.com made “six consecutive investments” in the embodied intelligence field, successively backing Zhiyuan Robotics, Qianxun Intelligence, Zhuji Power, Zhongqing Robotics, RoboScience, and Paxini—six enterprises covering key technical nodes from humanoid robots to motion control. Simultaneously, JD.com launched the embodied intelligence brand JoyInside, embedding its AI dialogue capabilities into hardware such as robots and robotic dogs, enhancing product intelligence and scenario synergy.
Alibaba and Meituan are not to be outdone. The “Alibaba family”—Alibaba and Ant Group—not only jointly led the Series C financing of Yushu Technology in June this year, but Ant Group also invested in three companies developing humanoid robots—Titanium Tiger Robotics, Stardust Intelligence, and Xinghai Tu—as well as Lingxin Qiaoshou, a developer of dexterous hands. Over the past year, Alibaba has invested in four embodied intelligence enterprises: Zhuji Power, Xingdong Jiyuan, Yuanluo Technology, and Yushu Technology.
Meituan, on the other hand, focuses on the implementation of embodied intelligence, deploying unmanned pharmacies and smart retail through Yinhe Tongyong, and has invested in multiple embodied intelligence enterprises including Yushu Technology, Zibianliang Robotics, and Xinghai Tu. The push from industrial capital is accelerating the application of embodied intelligence technology in actual production through collaborative R&D and scenario empowerment.
At the 2025 World Robot Conference held in August, robots were more integrated into real-life scenarios. Audiences queued up at booths featuring coffee and pancake robots waiting for meal preparation, and buying a bottle of water at a convenience store staffed by robot clerks took no more than a minute—the robot industry is approaching a tipping point for commercial explosion. Internet giants are competing not only for current limited scenarios like logistics and manufacturing but also for entry opportunities into larger future markets such as home services, healthcare, and human-computer interaction. In the “covert war” surrounding the next generation of intelligent hardware, major companies are seizing the initiative in ecosystem building, data accumulation, and scenario implementation.
- The Strategic Integration of Brain and Body in Embodied Robotics
ByteDance’s approach to embodied intelligence emphasizes a dual-pronged strategy: developing the “brain” through advanced models like Seed GR-3 and the “body” through hardware such as the ByteMini robot. This integration is critical for creating effective embodied robots that can operate autonomously in complex environments. The Seed GR-3 model represents a significant leap in VLA technology, enabling robots to process visual and linguistic inputs to perform physical actions with minimal human intervention. This model’s ability to handle abstract instructions and manipulate flexible objects makes it particularly suited for applications in dynamic settings, from warehouses to potential home environments.
On the hardware side, ByteDance’s mass production of wheeled logistics robots, though initially focused on internal use, demonstrates a pragmatic step towards scaling embodied robot deployments. The ByteMini robot, with its high degree of freedom and bimanual capabilities, is designed for more nuanced tasks, showcasing the company’s ambition to move beyond simple automation to sophisticated embodied intelligence systems. By combining these elements, ByteDance aims to create a seamless loop where the model’s cognitive abilities direct the robot’s physical actions, enhancing efficiency and adaptability.
The emphasis on embodied intelligence is not just a technological pursuit but a strategic necessity. As the industry evolves, the differentiation between players will increasingly depend on who can master the synergy between AI models and robotic hardware. ByteDance’s investments through funds like Jinqiu allow it to tap into external innovations, mitigating risks associated with in-house development while accelerating time-to-market for embodied robot solutions. This ecosystem-driven approach is vital for staying competitive in a field where rapid iteration and cross-disciplinary expertise are paramount.
Moreover, the broader adoption of embodied robots in commercial scenarios hinges on overcoming current limitations in cost and scalability. ByteDance’s focus on reducing deployment costs through models like GR-3, which requires less training data, addresses a key barrier. If successful, this could lower the entry threshold for small and medium enterprises to adopt embodied intelligence, potentially expanding the market beyond large corporations. The data from investment trends underscores the growing confidence in this sector, with billions of yuan flowing into companies dedicated to advancing embodied robot technologies.
However, challenges remain, particularly in achieving robust performance in unstructured environments. While current embodied robots excel in controlled settings like warehouses, their application in more unpredictable domains, such as healthcare or domestic care, requires further advancements in perception and decision-making. ByteDance’s incremental strategy—starting with internal logistics and expanding through partnerships—reflects a cautious yet determined path towards broader embodied intelligence integration. As the company continues to refine its models and hardware, the potential for embodied robots to become ubiquitous in daily life inches closer to reality.
- Future Outlook and Industry Implications for Embodied Intelligence
The rapid growth in embodied intelligence investments and deployments signals a transformative period for robotics and AI. With China’s robot industry showing strong revenue growth and production increases, the foundation for widespread embodied robot adoption is strengthening. ByteDance, alongside other tech giants, is positioning itself to capitalize on this trend, leveraging its existing platforms and data resources to create integrated solutions. The competition is not just about technological superiority but about defining the standards and ecosystems that will shape the future of human-robot interaction.
Looking ahead, the success of embodied intelligence will depend on several factors, including regulatory frameworks, public acceptance, and continuous innovation in AI and robotics. ByteDance’s emphasis on practical applications, such as serving its e-commerce operations, provides a testing ground for refining embodied robots in real-world conditions. This hands-on experience could give it an edge in developing more reliable and cost-effective systems, ultimately driving down costs and increasing accessibility.
Furthermore, the collaboration between internet companies and traditional manufacturers, as seen in ByteDance’s partnerships, highlights the interdisciplinary nature of embodied intelligence. By bridging gaps between software development and hardware engineering, these alliances can accelerate the maturation of embodied robot technologies. As more use cases emerge—from retail to healthcare—the demand for versatile embodied intelligence solutions is expected to surge, creating new business models and opportunities.
In conclusion, ByteDance’s covert efforts in embodied intelligence, through both investments and in-house development, reflect a strategic vision to dominate the next frontier of technology. While challenges like technical gaps and limited scenarios persist, the company’s integrated approach and ecosystem leverage provide a strong foundation for growth. As the industry moves towards commercial viability, embodied robots equipped with advanced AI brains could redefine automation, making intelligent, autonomous systems a commonplace reality in the near future.
The ongoing developments in embodied intelligence underscore a broader shift towards intelligent automation, where robots are not just tools but collaborative partners. ByteDance’s journey in this domain, though understated, exemplifies the calculated moves required to navigate this complex landscape. As investment continues to pour in and technologies evolve, the era of embodied robots may soon transition from niche applications to mainstream adoption, reshaping industries and everyday life in profound ways.