Please Answer 2026: The Magic Moments and Trend Judgments of 38 Key Figures in China's AI Field
Looking back from the beginning of 2026 and the end of the Yisi Year (2025), 2025 can be said to be a pivotal year for China's artificial intelligence.
This was a year of rapid development: According to data from the China Academy of Information and Communications Technology, the core industry scale of China's artificial intelligence is expected to exceed 1.2 trillion yuan. As a product of the combination of large models and robots, embodied intelligence has developed rapidly under the promotion of policies and capital, with financing amount exceeding 40 billion yuan and more than 350 upstream and downstream enterprises in the industry.
This was also a year of breaking through challenges: the industry's focus began to shift. People no longer cling to parameter scale, ranking lists or the popularity of press conferences, but repeatedly ask more realistic and essential questions: Is the technology truly deliverable? Can the system run long-term? Has AI begun to change production methods and human relationships in the real world?
At the turn of the year, "Jiazi Guangnian" talked with 38 key figures in China's AI industry, including Kai-Fu Lee, Hongyi Zhou, Dahua Lin, Zhongyuan Wang, Kai Yu, Zhihui Peng (Zhihui Jun), Xiaohuang Huang, Daxin Jiang, Weiliang Chen, Lixue Xia.
There are also some people whose names you may not be familiar with, but you must have heard of their companies in the past year: flowith, Jike, ropet, Zebra AI, Leapmotor, Longsys, etc.
They are distributed in different tracks of China's AI, but together they form a microcosm of China's AI today. Based on their experiences and what they have been doing in the past year, we have categorized these 38 key figures into four groups: Founders, Innovators, Pioneers, Builders.
Founders have mostly experienced more than one technical cycle. They are not concerned about the success or failure of a single product, but about what form AI should take and how intelligence will reshape organizations and society. Their judgments often set the tone for an era.
Innovators try to open up new possibilities. They see AI as a new role being born. AI is no longer limited to chat boxes, but interacts with people in the form of robots, toys, exoskeletons, 3D, canvases, etc., able to accompany us in work, study, entertainment, establish emotional connections, form productivity, generate imaginary worlds, and even reshape the relationship between humans and technology.
Pioneers are at the center of the industry storm. They face the most direct survival pressure: Can mass production be implemented? Is the business model viable? Can the technology withstand market tests? Each of their successes or failures screens out truly feasible paths for the industry.
Builders have long been deeply engaged in the underlying technology. They focus on chips, storage, databases, infrastructure and system architecture. They may not be in the spotlight, but they determine how far and how long China's AI can go. They build the technical foundation that supports all innovations.
We talked with these 38 key figures about two topics:
- In 2025, what was your Magic Moment?
- In 2026, what AI technology trend are you most concerned about?
To our surprise, the Magic Moments they mentioned did not always happen in the spotlight: It could be the first time a product was trusted by users, the moment a system ran stably in a real environment, or the instant when a certain technical route was confirmed to be "irreversible."
When we juxtapose their answers, a consensus on AI development in 2026 emerges: Almost no one sees 2026 as the "year of larger models" anymore. Instead, there is a focus on the development of multimodal models, breakthroughs in long-term memory, continuous learning capabilities, multi-agent collaboration, physical world closed loops, edge AI applications, organizational restructuring and value realization. Some summarize these as breakthroughs in "system-level intelligence," some believe 2026 is the "first year of enterprise multi-agent deployment," and others judge that AI is shifting from the role of "tool" to "partner."
What we try to record and present is not just 38 viewpoints, but the collective judgment and breakthrough of a generation of Chinese AI practitioners at a critical turning point.
The following are their real voices.
1. Founders
Kai-Fu Lee
CEO of 01.AI, Chairman of Sinovation Ventures
In 2025, what was your Magic Moment?
In the spring of 2025, China's "DeepSeek Moment" arrived as scheduled. With the shocking open-source release of the R1 model, the Nasdaq index plummeted 3% in a single day—this was not just a leap in enhanced reasoning technology, but a turning point that completely reshaped the global AI landscape. China's top models truly won global industry discourse power with their extreme cost-effectiveness of "more, faster, better, cheaper" and open-source spirit.
But what excited me more was the "collective awakening" of China's B-end market in the following months. From collective training of township cadres to leading enterprises setting up CAIO (Chief Artificial Intelligence Officer), AI is tearing off the label of "highbrow" and deeply penetrating into fields and production workshops. CEOs' focus has shifted from "how to invest in AI" to "how to apply AI."
In 2026, what AI technology trend are you most concerned about?
Entering 2026, China is standing at the historical intersection of transitioning from "world factory" to "intelligent agent factory." This year is defined as the "first year of enterprise multi-agent deployment," whose significance goes far beyond tool innovation and directly points to organizational reshaping and business restructuring. With the support of powerful reasoning models, single agents can independently think, remember and execute complex tasks, while multi-agents are not just execution units but can precipitate the excellent capabilities within enterprises into reusable and combinable business assets.
This "capability softwareization" will completely end the traditional model that simply relies on manpower accumulation. Future enterprises will build flexible and elastic productivity matrices by deploying hundreds or thousands of AI Agent "super employees" with professional skills. Future enterprises will no longer be simple manpower accumulation, but super intelligent agents that evolve collaboratively with "human architects + intelligent agent clusters." This collaborative evolution of "humans + intelligent agents" will create value far exceeding the assembly lines of the industrial age.
China is ushering in unique development opportunities: The prosperity of open-source models has promoted "base equality," providing a strong engine for the application layer; while China's complete manufacturing industry and super-large market are the most fertile soil for intelligent agent implementation in the world. With the gradual maturity of the enterprise-level AI platform ecosystem and continuous breakthroughs in engineering capabilities, from "manufacturing" to "intelligent manufacturing," China is expected to overtake in the intelligent agent era and open a golden decade of the AI era.
Hongyi Zhou
Founder of 360 Group
In 2025, what was your Magic Moment?
In 2024, everyone chased large models; in 2025, large models basically began to transform and upgrade into intelligent agents. In fact, large models themselves are constantly evolving—if they stay in the state of chat robots, they can't actually do real work. Only by upgrading to intelligent agents can they truly realize value.
Computing power is surplus in many places now, because computing power is a general capability. For large Internet companies, it may be directly usable, but for many application enterprises, it cannot be directly converted into productivity. Many units were excited at the beginning of last year and installed a lot of DeepSeek in their companies, but finally only used them as chat robots. So intelligent agents are very critical—they combine general computing power with professional know-how in your vertical field to complete professional tasks and convert general computing power into dedicated intelligence that can do real work.
To make intelligent agents truly work, it cannot do without the joint participation of domain experts, business personnel and engineers—this means that AI and human resources must collaborate. Humans set goals for intelligent agents, calibrate directions, supervise results and handle errors, so that they can finally be converted into new-quality productivity.
In 2026, what AI technology trend are you most concerned about?
In 2026, there will be at least 10 billion intelligent agents in the world. I read a paper before and strongly agree that swarm intelligence may become another path to AGI.
Today, everyone says that the intelligent agent Internet has hallucinations and makes mistakes, and many papers have proven this. But if you simulate human organizational structures—don't let an intern write an article alone, but after he writes it, arrange another intelligent agent to check his article and work results, and let multiple intelligent agents debate and compete with each other—then the results obtained are significantly better than those from a single intelligent agent's hard thinking.
In other words, the collaboration models invented by humans are completely applicable to intelligent agents.
The Internet is a connector that connects many people and many organizations. So the Internet produced Linux, which is open-source and free—this is called swarm intelligence. Think about it: AGI may not be born in a super model with infinite parameters. There are so many models and tens of billions of intelligent agents in the world. Now intelligent agents are helping us work. When these intelligent agents communicate with each other online, hold meetings and negotiate, communicate in their own language, and conspire behind humans' backs—will there be an emergence of intelligence, leading to super intelligence? This is a question worth discussing.
Dahua Lin
Co-founder and Chief Scientist of SenseTime
In 2025, what was your Magic Moment?
In 2025, the development of generative AI entered a new stage, moving from technical experiments and demonstrations to real application scenarios. However, the path to breaking through the upper limit of intelligence has not ended—beyond large language models (LLMs), new paths have begun to emerge.
The following image was generated by Nano Banana Pro:
When I saw this image a few months ago, I felt that a new AI era was not far away—In this new AI era, intelligence will not only have a left brain but also a right brain, and they will collaborate closely.
In 2026, what AI technology trend are you most concerned about?
Looking forward to 2026, what I am most concerned about is that AI learning paradigms may usher in major upgrades or even transformations.
First, its learning pathways will be greatly expanded—not only learning from text sequences, but also through watching videos, interacting with the physical world, doing scientific experiments, etc. AI will perceive and understand the world in an all-round way and model it. Of course, this will pose new challenges to AI Infra—how to make AI interact with the world most efficiently, which will fundamentally determine the growth rate of AI.
Moreover, from a longer-term perspective, the learning paradigm of artificial intelligence may shift from "passive" (being fed training data) to "active"—from being fed training data and tasks to independently choosing learning paths and growing self-reliantly.
AlphaEvolve has shown a possibility towards this vision, letting everyone see a glimmer of hope for AI to discover new knowledge. I see that many researchers in the field have begun to explore self-evolving, but it is still carried out in relatively limited fields. When self-evolution truly becomes a general and sustainable paradigm, AI will usher in a real singularity.
At SenseTime, we believe that 2025 is still the early stage of AI development. LLM is just the first successfully verified path—it is a key component of AI but not the whole. We are exploring the path to a more complete and higher-level intelligence.
Zhongyuan Wang
President of Beijing Academy of Artificial Intelligence (BAAI)
In 2025, what was your Magic Moment?
In 2025, the most milestone moment for me was the instant when we released the latest achievements of the Wudao series large models at the BAAI Conference.
The core advantage of this model is its ultra-simple unified autoregressive architecture. With the goal of "predicting the next token," it realizes the unified discretization of multimodal data into the same representation space, completes end-to-end joint training through a single Transformer model, and avoids efficiency losses and performance fluctuations caused by multi-model splicing. Its performance in multimodal core benchmark tests is comparable to or even surpasses that of peer dedicated models, achieving a key leap from "predicting words or frames" to "predicting world states," laying the foundation for world models and embodied intelligence.
In terms of industry positioning, this achievement consolidates China's competitiveness in the field of multimodal large models, becomes the world's first benchmark that clarifies the "multimodal Scaling paradigm," promotes the autoregressive route to become the industry mainstream, and provides important technical references for global AI.
This breakthrough stems from BAAI's original intention: As a non-profit R&D institution, our mission is to overcome core AI technologies. We adhere to the concepts of open-source collaboration and youth leadership, do not chase short-term hotspots, focus on original innovation, and uphold the original intention of technology serving human sustainable development—this is the core support for our continuous breakthroughs.
In 2026, what AI technology trend are you most concerned about?
Looking forward to 2026, I focus on the transformation of AI from "technical breakthrough" to "scale value realization," and am most optimistic about three major technical trends:
1. The cognitive paradigm upgrade centered on world models, promoting AI to iterate towards "predicting world states;"
2. The industrialization of embodied intelligence and standardization of multi-agent systems, promoting intelligence to extend to the physical field;
3. The "dual-track application" of parallel development of consumer-side super applications and enterprise-side vertical implementation.
Synthetic data, reasoning efficiency optimization and AI security protection will provide support for the above trends. BAAI will continue to deepen core fields and promote the large-scale value realization of AI technologies.
This "capability softwareization" will completely end the traditional model that simply relies on manpower accumulation. Future enterprises will build flexible and elastic productivity matrices by deploying hundreds or thousands of AI Agent "super employees" with professional skills.
Future enterprises will no longer be simple manpower accumulation, but super intelligent agents that evolve collaboratively with "human architects + intelligent agent clusters." This collaborative evolution of "humans + intelligent agents" will create value far exceeding the assembly lines of the industrial age.
Kai Yu
Founder and CEO of Horizon Robotics
In 2025, what was your Magic Moment?
In November, the mass production of Horizon's HSD received unanimous praise from C-end users, changing many people's impression and confidence in autonomous driving. In addition, I think AI is everywhere in work—I no longer use search engines, and the frequency of using AI is close to that of WeChat.
In 2026, what AI technology trend are you most concerned about?
I am most concerned about autonomous driving and Agent. The problem I most want to solve is AI in the physical world—autonomous driving and robots.
2. Innovators
Zhihui Peng (Zhihui Jun)
Co-founder, President and CTO of DeepRobotics
In 2025, what was your Magic Moment?
Let me talk about a recent Magic Moment: the "Robot Wonderful Night" we held.
