At the 2025 China Mobile Global Partners Conference, a significant gathering of industry leaders highlighted the rapid evolution of domestic computing infrastructure and its pivotal role in accelerating artificial intelligence applications. In his keynote address, China Mobile Chairman Yang Jie outlined comprehensive upgrades across digital-intelligent facilities, technologies, applications, security, and collaboration. He emphasized that AI is fundamentally reshaping production factors and value creation paradigms, driving the digital economy toward a higher-level intelligent economy and injecting sustained momentum into the development of new quality productive forces. On-site observations revealed that embodied intelligence emerged as a standout highlight of the event, with multiple manufacturers showcasing key products, underscoring the trend of embodied intelligence accelerating toward mass production and deployment this year. Simultaneously, Chinese companies have made substantial progress in enhancing the computational efficiency of computing power infrastructure, addressing critical bottlenecks in data transmission and cluster performance.
-
The Acceleration of Computing Power Foundations
In the wave of AI large models, computing power infrastructure has become a crucial competitive advantage. Chairman Yang Jie detailed efforts to build efficient intelligent computing facilities, advancing a cloud-intelligence integrated and training-inference integrated computing resource system. This initiative aims to upgrade cloud computing to “cloud-intelligent computing,” accelerate the transformation of computing-network brains into “computing-network intelligent brains,” and strengthen the interconnection and supply-demand matching of intelligent computing. Explorations include the construction of 100,000-card-level intelligent computing centers and quantum technology infrastructure, promoting computing-power coordination, and continuously optimizing the integrated supply of “general-intelligent-super-quantum” computing power.
From an industry chain perspective, achieving high-efficiency computing capabilities hinges critically on high-speed interconnection. As large models continue to evolve, Scaling Law, though slowing in pre-training phases, has not become obsolete, meaning that running models imposes increasingly higher demands on underlying AI computing clusters. The challenge of scaling “single-point efficiency” to “cluster efficiency” is a common hurdle for domestic computing power foundations. One chip company executive pointed out that future data transmission will be a bottleneck for AI infrastructure, highlighting the background for the industry’s active promotion of super-node capabilities.
Shen Yichen, founder and CEO of Xi Zhi Technology, addressed this in the Computing Power Network Joint Innovation Sub-forum, noting that with the development of trillion-parameter large models, computing power demand is exploding, while electrical chip performance approaches physical limits, widening the computing power gap. Two paths to破解 this bottleneck are improving multi-card collaboration efficiency or enhancing single-card computing power. Xi Zhi Technology introduced super-node solutions supporting the OISA protocol, including optical-interconnected electrical switching super-nodes and the Yue Guang 1024 optical-interconnected optical switching super-node. The latter, reportedly the first optical-interconnected optical switching GPU super-node solution in China, will be deployed in Shanghai Yidian’s domestic super-node computing power cluster.
The core technology involves a globally首创 distributed optical circuit switch chip based on silicon photonics technology, analogous to eliminating a central logistics hub in a factory and setting up small material transfer stations near each workshop or workstation, making material transmission paths shorter and more direct. A chip industry executive analyzed that the key point is that data movement in physical space requires time, so issues that can be resolved between chips should not be addressed within machines or between cards. Currently, super-nodes are an inevitable development direction. The Yue Guang 1024 optical-interconnected optical switching super-node solution achieves Scale-up network connections between GPUs from multiple domestic and foreign manufacturers without dedicated switches; moreover, as optical switching itself does not rely on specific data transmission protocols, it can seamlessly兼容 different interconnection protocols used by various vendors.
-
Exploration and Deployment of Embodied Intelligence
The conference emphasized the future trend of carbon-silicon life symbiosis, with on-site embodied intelligence practitioners demonstrating the progress of robots in “working” schedules. In terms of trends, Chairman Yang Jie stated in his speech that AI continues to release “two scale effects” in technical capabilities and economic benefits, integrating into various sectors of the economy and society at an unprecedented speed, promoting carbon-silicon integration and symbiosis as an important trend in the development of human civilization. In terms of silicon-based life, future silicon-based life forms, with sensors, processors, memory, controllers, and other physical hardware as the “body,” and computational intelligence, perceptual intelligence, cognitive intelligence, and motor intelligence as the “neural center,” will emerge collectively. AI intelligent terminals, intelligent connected vehicles, embodied robots, and others will become widely popular, with the scale of silicon-based individuals far exceeding humans, forming new “talent dividends” and appearing in new “360 industries.”
In terms of carbon-based life, breakthroughs in brain-computer interfaces, biological intelligence, and other technologies will promote the gradual integration and co-evolution of carbon-based and silicon-based life at physiological and cognitive levels, giving rise to new life forms such as exoskeletons, cyborgs, and digital life, helping humans enhance physical abilities, break through cognitive bottlenecks, and achieve dual leaps in physical and mental power. Yang Jie pointed out that in the future, more and more “enhanced humans” deeply empowered by AI will be born, redefining the boundaries of life. From the perspective of carbon-silicon life co-creation, silicon-based individuals will upgrade from equipment tools to collaborative partners, organizational forms will accelerate toward “human-machine collaboration,” “super platforms” and “one-person companies” will rapidly rise, and “carbon-silicon symbiosis” culture will accelerate formation, triggering systematic changes in organizational systems, ethical norms, legal institutions, and more.
At the Zhiyuan Robot booth during the conference, multiple models of humanoid robots and robotic dogs of different sizes were displayed. According to introductions, based on different forms of embodied intelligence products, the company has currently built solutions covering eight areas: security inspection, industrial intelligent manufacturing, logistics sorting, explanation and reception, cultural entertainment and commercial performances, education and research, data collection and training, and commercial cleaning. Among them, the Zhiyuan精灵 G1 embodied robot demonstrated sorting capabilities on-site; this category can operate in industrial manufacturing and logistics sorting scenarios, while also serving as a data collection and training gripper, providing data “fuel” for the evolution of embodied intelligence. The four-legged robotic dog D1Ultra can be applied in scenarios such as security patrol, scientific research education, and interactive entertainment.
Zhiyuan Robot is actively deploying in industrial scenes. Recently, Zhiyuan Robot and ODM leader Longqi Technology announced a deep strategic cooperation on the application of embodied intelligence robots in industrial scenarios, with Longqi Technology placing a framework order for Zhiyuan精灵 G2 robots worth billions of yuan. It is reported that the Zhiyuan精灵 G2 will initially focus on tablet production lines, playing roles in flexible grasping, multi-station collaboration, and production line data linkage. In August, Zhiyuan Robot also reached a project cooperation with Fulin Precision involving tens of millions of yuan, with nearly a hundred Yuan Zheng A2-W robots landing in Fulin Precision’s factory, marking the first case of large-scale commercial signing of embodied robots in the domestic industrial field.
From an application trend perspective, Wang Xingxing, founder, CEO, and CTO of Yushu Technology, stated during a forum that robots are currently landing faster in consumption and entertainment aspects, but face relatively greater challenges in home and industrial scenarios. In his planning, he hopes that in the future, embodied intelligence in home scenarios can serve tea and water, do housework, and buy groceries, but this imposes high comprehensive requirements on robots, involving both technical and ethical aspects, thus affecting the speed of deployment. In contrast, robot categories requiring only single functions land faster, typically like扫地 robots. Wang Xingxing pointed out that robots still face certain challenges in industrial scenarios; currently, Yushu Technology is promoting robot entry into factories with Nio and Geely Group, mainly facing the comprehensive performance of efficiency and cost needing to be higher than current traditional automation equipment and mechanical arms. Once this critical point is crossed, embodied robots can truly scale in industrial scenes. “Although facing challenges now, overall I am optimistic,” he summarized, noting that in the first half of this year, China’s robot industry is developing vigorously, many overseas large companies are also receiving attention, and the overall technological progress of the industry exceeds expectations.
-
The Distance and Proximity of AI Applications
Of course, embodied intelligence must also face external challenges. Wang Xingxing has mentioned on different occasions that humanoid robots have larger body sizes, so ensuring safety is particularly important. However, as a relatively new product type, robots are currently facing the annoyance of being hacked. He found that recently, some hackers and white hats overseas are attempting to crack robots. But he also thanked these discovered vulnerabilities; it is revealed that Yushu Technology is also advancing EU safety certification related work. “We also hope to receive friendly help from the whole society, with more supervision to help us solve vulnerability problems.” This is similar to the early days of the PC industry when computer viruses were easily attacked, and software development required joint driving from the whole society. Wang Xingxing pointed out that the development of robot products must also focus on the improvement of capabilities from underlying software to AI applications, security, and even laws and regulations, which is an important direction for building a sound embodied intelligence system.
He also explained why some robots suddenly fall during work: in the process of designing safety protection mechanisms for embodied intelligence, obvious emergency stop buttons or specific voice commands are usually added. The sudden fall of robots may include reasons such as the robot’s emergency stop button being accidentally triggered. As the “year of intelligent agents,” industry chain manufacturers are also actively promoting related work in building industrial ecosystems. According to China Mobile Vice President Zhang Dong, China Mobile has achieved a full-domain refresh of AI+ product families, with the cumulative number of AI+ product users exceeding 210 million, and the “Lingxi” intelligent agent monthly active users exceeding 70 million. In the next stage, the “2026 Action Plan” will be fully promoted: first, building a “1+1+N2” intelligent agent product system, covering 1 set of intelligent agent tools, 1 set of intelligent agent protocols, and an N2 intelligent agent matrix, constructing an Agent+robot product matrix and service ecosystem, and promoting the deep empowerment of the “Lingxi” intelligent agent; second, building a digital-intelligent life mall, based on communication services, creating a “one foundation + dual cores + two wings” system, defining new business forms with intelligent agents; third, establishing an intelligent agent ecosystem alliance, deeply coordinating in commercial value, high-quality resources, and development policies.
In the process of intelligent agent application deployment, interconnection protocols are also a construction focus. China Mobile took the lead in the industry in releasing the A2A-T intelligent agent interaction protocol stack. China Mobile, jointly with partners, increased the overall efficiency of multi-agent interaction by 20%. At the same time, a batch of network operation Lingxi intelligent agents were launched to enhance network elasticity and business quality. The continuous upgrade of computing power infrastructure and the deep integration of industrial collaboration are jointly promoting AI from technological concepts to commercial reality. Under the vision of “carbon-silicon symbiosis,” the virtuous cycle of computing power and applications is accelerating construction, with embodied intelligence and embodied robots playing a central role in this transformation.
The advancements in embodied intelligence are not limited to industrial applications; they extend to various sectors, including healthcare, education, and services. For instance, in healthcare, embodied robots can assist in patient care, rehabilitation, and surgical procedures, enhancing precision and reducing human error. In education, these robots can provide personalized learning experiences, adapting to individual student needs and promoting engagement. The service industry benefits from embodied intelligence through automated customer service, logistics, and hospitality, improving efficiency and customer satisfaction. However, the widespread adoption of embodied intelligence requires addressing several challenges, such as cost-effectiveness, interoperability, and public acceptance. Companies are investing in research and development to overcome these hurdles, focusing on making embodied robots more affordable, reliable, and user-friendly.
Moreover, the integration of AI with other emerging technologies, such as 5G, IoT, and blockchain, is amplifying the capabilities of embodied intelligence. For example, 5G networks enable low-latency communication for real-time robot control, while IoT devices provide vast data streams for training and operation. Blockchain technology can enhance security and transparency in robot transactions and data sharing. These synergies are crucial for scaling embodied intelligence applications and achieving the vision of carbon-silicon symbiosis. As Yang Jie highlighted, the future will see an increasing number of AI-empowered “enhanced humans,” blurring the lines between biological and artificial intelligence. This convergence demands robust ethical frameworks and regulatory standards to ensure responsible development and deployment.
In conclusion, the 2025 China Mobile Global Partners Conference underscored the transformative potential of computing power and embodied intelligence in shaping the future economy and society. With continued innovation and collaboration, the era of carbon-silicon symbiosis is becoming a tangible reality, driven by advancements in embodied intelligence and embodied robots. The progress in computing infrastructure, exemplified by super-node solutions and high-speed interconnections, provides the necessary foundation for this evolution. As embodied intelligence moves from concept to commercialization, it promises to redefine industries, enhance human capabilities, and create new opportunities for growth. The journey ahead involves addressing technical, ethical, and societal challenges, but the collective efforts of industry leaders, policymakers, and researchers are paving the way for a harmonious integration of carbon and silicon life forms.