Summary of Key Points from the 2026 World Artificial Intelligence Conference (WAIC)
The most significant change at the 2026 WAIC was the shift from a focus on “showy displays of technological prowess” to a more practical emphasis on practical applications. The AI and augmented intelligence sectors are no longer competing based on superficial aspects such as the number of model parameters or robot dances; instead, they are concentrating on critical commercial issues like productivity creation, return on investment (ROI), organizational restructuring, data barriers, and innovation in computing power. Consumer companies have also shifted from merely “testing out AI” to carefully calculating how much money AI can save them. Robots have moved beyond the stage of conceptual validation to factory testing, although achieving stable operation remains a challenge.
1. AI Competition: No Longer About Parameters, but About Cost-Effectiveness
Over the past two years, the AI industry has been about competing on the strength of models. This year, however, there’s a shift towards asking “how much value AI really offers.” The reason is simple: model capabilities are becoming more homogeneous, and costs have decreased (for example, the cost of generating AI-generated advertising content has dropped from 120 yuan per piece to 50 yuan per piece), allowing companies to accurately calculate how much labor can be saved and how much profit can be generated by using AI.
- Example 1: A company called Jiaoke used AI for marketing materials, but the cost was higher than using manual methods. It wasn’t until the model costs decreased and efficiency improved that it became economically viable.
- Example 2: The lemon tea brand Linli eliminated 50 positions through AI, all of which were involved in data handling and repetitive approvals.
- Expert Comment: Sun Tianshu, a professor at Cheung Kong Graduate School of Business, describes this as the “second half” of the AI revolution—where the focus has shifted from building infrastructure (computing power, tokens) to determining the value that these technologies can create.
2. AI Agents: More Than Just Tools, They Are Organizational Transformers
AI agents are not just used to help employees create PowerPoint presentations; they are changing the way entire companies operate. While individuals may use AI to improve their efficiency, the overall company efficiency hasn’t increased because the underlying processes haven’t changed.
- How to Change: For instance, a company uses agents for product selection, live broadcast scripting, and data analysis, which also helps optimize future decisions. Schneider Electric has formed cross-departmental teams to integrate agents into operations rather than treating them merely as tools.
- Key Insight: Decathlon realized that AI doesn’t automatically make companies faster; it instead exposes existing problems such as chaotic processes and inconsistent data. These issues must be addressed before AI can truly enhance efficiency.
3. Data Is More Valuable Than Parameters: Those with Real-World Data Have the Edge
In the past, large model companies competed based on the size of their parameter sets. This year, the focus is on “data wheels”—the continuous collection of data from real-world scenarios and using that data to refine models in a cycle that makes them more intelligent over time.
- Example 1: Jieyuexing believes that the success of Claude isn’t due to a sudden increase in model intelligence but because it was integrated into actual work processes (such as code writing) and continuously improved based on user feedback and usage data.
- Example 2: Yunji Technology’s robots have collected service data in hotels and hospitals, enabling them to develop models that understand which actions are more valuable (e.g., delivering water to guests is more useful than dancing).
- Conclusion: Model capabilities will eventually converge, but real-world data is a competitive advantage that cannot be easily replicated. This will become the new barrier to success.
4. Robots Moving Beyond Dances: From Showcases to Practical Applications in Factories
Last year, robot exhibits featured robots performing acrobatics and dances. This year, companies are discussing how robots can be used for order fulfillment and on production lines. However, there’s still a long way to go before widespread adoption:
- Current Status: Most robots are still in the POC (Proof of Concept) phase, capable of working in specific scenarios but not at scale. Zhiyuan Robotics has already deployed a small number of robots in factories for tasks like loading and unloading tablets, which have operated stably for six consecutive days.
- Challenges: Factory customers are concerned about whether robots can work reliably 24/7 without affecting production efficiency. The CEO of Guangxiang Technology noted that the industry is still transitioning from POC to testing phases, with real challenges (such as generalization capabilities and cost control) just beginning to emerge.
- Technological Trends: The debate last year was about whether to use VLA or world models; this year, there’s consensus that any technology that enables robots to complete tasks reliably is acceptable. Integration of different technologies is the way forward.
5. Computing Power Business Evolving
The computing power industry is experiencing new developments. This year, the focus is no longer on selling servers but on “intelligent outputs” in the form of tokens. New models have emerged:
- Token Trading: Companies like PPIO sell idle computing power from China during the night to users in the Western Hemisphere, achieving utilization rates of 75%-80%. Gongji Technology shares idle computing resources from personal computers and internet cafes, with customers hardly noticing the change (even with 1,963 computers involved during the Spring Festival).
- Mobile Computing Power: AI is moving from the cloud to devices (e.g., smart glasses and household robots). Rokid’s glasses can automatically translate languages; Hemo Intelligence’s chips combine storage and computing capabilities, balancing power consumption, cost, and intelligence. Buying such a device is like obtaining a “free token factory” for future use of AI services.
- Future Trends: In three years, 90% of token requests may come from AI agents, and billing will shift from based on the number of tokens used to based on the actual benefits (similar to how the advertising industry charges per click).
Conclusion
The theme of this year’s WAIC is “practicality.” AI is moving from a technological frenzy to value creation, robots are shifting from being showpieces to tools for increasing productivity, and companies are transitioning from experimenting with new technologies to carefully calculating their economic benefits. These changes indicate that AI and augmented intelligence are finally making their way from the laboratory to the real world. However, there are still many hurdles to overcome before they can become a fundamental part of daily life. For ordinary people, AI may not be seen as a high-tech novelty but rather an essential infrastructure, just like water and electricity, becoming ubiquitous in our work and daily lives.