Summary of Key Highlights
The 2026 World Artificial Intelligence Conference (WAIC) showcased a comprehensive breakthrough in China's AI industry chain, covering everything from "underlying computing power" to "mid-level models" and finally to "upper-layer terminal applications": Domestic computing power is no longer just about competing on individual chips; instead, there is a focus on optimizing system-level capabilities and efficiency. Large models are extending from the digital world to the physical world (embodied intelligence), with companies beginning to design dedicated "brains" for robots. Terminal devices such as intelligent phones, glasses, and robots have entered the stages of practical use and mass production. At the same time, domestic companies are accelerating ecosystem construction through partnerships and are also open to global cooperation (with foreign firms like Seagate participating). Faced with international competition from companies like Tesla's robotics division, Chinese enterprises believe that this will foster mutual growth in the industry.
1. Computing Power Foundation: From "Competing on Chip Quantity" to "Competing on System Efficiency"
In the past, people compared who had the most GPUs; now, the focus is on generating greater value from computing power.
- System-Level Breakthroughs: Huawei,摩尔 Threads, and Alibaba Cloud have all launched "super-node" products that integrate hundreds or even thousands of chips into a single cabinet, functioning as a supercomputing unit. For example, Huawei's Atlas 950 can incorporate 1,024 Ascend cards, while Moore Threads' super-nodes address the issue of high network latency. Alibaba Cloud has also made this type of computing power available directly to enterprises through its public cloud services.
- Collaborative Ecosystem Building: SenseTime, in collaboration with Cambricon and Huawei, has initiated the "Galaxy Plan" to build five AI computing clusters with tens of thousands of cards each. This is not a solitary effort; rather, it's about jointly building a platform for all participants.
- Domestication Is More Than Just Replacing GPUs: Experts suggest that the replacement in the next five years will involve a complete reconstruction of the entire upstream hardware ecosystem, including chips, software, and storage. Improving computing power efficiency could unlock trillions of dollars in value, shifting the focus from simply accumulating hardware to optimizing its use.
- Global Cooperation to Fill Gaps: Foreign companies like Seagate are providing large-capacity storage solutions, enabling more efficient data flow for AI both in the cloud and at the edge. After all, AI generates a vast amount of images and videos that require substantial storage, and such cooperation can reduce China's AI training costs.
2. Model Evolution: From "Digital Players" to "Physical Learners"
Large models are no longer limited to processing text and images; they are beginning to understand the rules of the physical world and are being developed to serve as the "brains" for robots.
- Advancing in Ultra-Large Parameter Models: Yuezhi Dianmian has released the K3 model with 2.8 trillion parameters, while MiniMax and iFlytek are working on models with around 2 trillion parameters. These models can handle more complex tasks and work continuously for extended periods.
- Embodied Intelligence as a New Direction: Companies have realized that models trained for smartphones cannot be directly applied to robots. The physical world is not a copy of the digital one; robots need to perceive object weights, judge road conditions, etc., and therefore require specialized models. For instance, Ant Lingbo's "Full Stack Brain 2.0" is designed specifically for physical tasks, and Zhiyuan also emphasizes "embodied native design."
- When Will There Be a Breakthrough?: Experts predict that by the end of 2027 to early 2028, embodied intelligence will see a significant advancement, with robots being able to handle real-world tasks as flexibly as humans.
3. Terminal Device Transformation: AI Moving From "Answering Questions" to "Proactively Performing Actions"
Terminal devices are no longer just equipped with AI; they are becoming tools that truly assist users.
- Explosion of Intelligent Phones: Honor, Jieyue Xingchen, and Nubia have all launched intelligent phones that restructure the operating system to understand user intentions. For example, you can say, "Book a high-speed train for me to Shanghai tomorrow and set an alarm," and the phone will automatically handle the ticketing, ordering, and setting the alarm without requiring you to perform each step manually.
- Intelligent Glasses as a New Entry Point: VITURE's glasses can transform the Android system into a 3D experience, making watching movies and playing games more immersive. IDC data predicts that global smart glass shipments will reach 13.6 million units in 2026, as they offer convenience by freeing users' hands and allowing them to use the devices with just a glance.
- Robots Entering Mass Production: Zhiyuan's robots are already being used in flat-panel factories and have collaborated with JD Logistics. JUNPU Intelligence's production line is the world's first to utilize multiple robots working together. Critical Point's dexterous hands can grasp objects just like human hands, indicating that robots are moving from the laboratory to practical applications.
4. Global Competition and Cooperation: Facing Tesla's Robotics Division, Chinese Enterprises Are Open
In the face of international competition, Chinese companies are not isolating themselves but are embracing collaboration.
- Foreign Companies Participating in China's Ecosystem: Firms like Seagate are contributing storage technology to help solve AI's data challenges, working together to expand the market.
- The Impact of Tesla's Robotics Division: Tesla is preparing for mass production of robots. Chinese companies see this as a positive development, similar to how the entry of electric vehicles into China spurred the upgrading of the domestic supply chain. Qiling Intelligence believes that the industry is large enough to accommodate many players, and Zhiyuan also welcomes competition, as it will drive overall progress in the field.
Conclusion
China's AI industry chain is transitioning from a follower position to one where it takes the lead in certain areas. The foundation of computing power is becoming more solid, models are exploring the physical world, and terminal applications are becoming more user-centric. Through partnerships and global cooperation, the industry is developing rapidly. In the coming years, AI will become even more integrated into our work and daily lives, evolving from being able to "speak" to being capable of "performing tasks."