A dialogue about the present and future of artificial intelligence

In Lin'an, Hangzhou, a profound dialogue about the future of artificial intelligence took place at the 2025 Fortune Capital Entrepreneurs Summit and Industry-Finance Conference. It not only gathered top practitioners in the Chinese AI field but also brought the most core issues in the industry to the forefront —

1. On this innovative hotbed in Hangzhou where the "Six Little Dragons" of AI were born, what exactly is the secret to success?
2. In the large model race, how can pioneer startups like Zhipu AI dance with tech giants like Alibaba Cloud?
3. In the highly anticipated field of embodied intelligence, are we creating a more efficient "tool" or nurturing a truly "human - like" partner?

The dialogue was hosted by Wu Xi, a partner of Fortune Capital. The guests on the stage were impressive, including: Pan Xiaohui, the chairman of Hangzhou Industrial Investment Group; Zhang Peng, the CEO of Zhipu AI; Zhao Tongyang, the founder of Zhongqing Robotics; Li Chao, the co - founder of DeepRobotics; and Yang Bin, the person in charge of the product solutions of Alibaba Cloud's Tongyi large model.

This discussion not only revealed the innovative ecosystem of "governing by non - interference" in Hangzhou but also delved into the calm thinking and firm belief of the AI industry in terms of technical routes, business models, and even ultimate risks.

The following is the transcript of the round - table dialogue —

The Hangzhou Code: Why Were the "Six Little Dragons" of AI Born Here?

At the beginning of the dialogue, host Wu Xi raised a question widely concerned in the industry: Amid the fierce competition in first - tier cities, why was Hangzhou able to stand out and give birth to the so - called "Six Little Dragons"?

Pan Xiaohui: We didn't do anything. The "Six Little Dragons" emerged on their own. It wasn't intentional on the part of the government or state - owned enterprises. Analyzing this matter, there is both an element of "chance" and some "inevitability".

During the Spring Festival, the sudden popularity of companies like DeepSeek and Unitree did have an element of "luck", but it greatly boosted the confidence in national scientific and technological innovation. However, behind the chance lies an inevitable foundation. Hangzhou has adhered to the market - oriented path for many years, clarifying the boundaries between the government and the market — "The government does what the government should do, and the market does what the market should do." The government's role is to optimize the ecosystem, provide public services, and follow the principle of "not disturbing without cause and responding to every request". Especially when the confidence of social capital was insufficient in previous years, state - owned capital represented by Hangzhou Capital firmly injected crucial liquidity into innovative enterprises in the form of mother funds and industrial investments.

An academician from Zhejiang University once summarized Hangzhou's competitiveness as follows: "The university president doesn't interfere with the deans, and the deans don't interfere with the professors." This culture of trusting professionals and empowering individuals is the core for innovation to grow freely.

Li Chao: The emergence of the "Six Little Dragons" is closely related to four characteristics of Hangzhou. Firstly, it has a unique talent structure. Hangzhou has gathered talents in multiple fields such as algorithms, software, and engineering, deeply influenced by the "talent atmosphere" created by large companies like Alibaba. Secondly, there is tangible government support. When enterprises saw no hope, the government's confidence and policy support were important factors for them to persevere. Thirdly, it has a unique urban rhythm. Hangzhou offers "both entrepreneurship and a good life". Compared with the hustle and bustle in first - tier cities, this relatively relaxed rhythm is more conducive to hard - tech innovation that requires long - term commitment. Finally, the strong academic atmosphere created by universities such as Zhejiang University and Westlake University provides an endless source of soil for innovation.

Yang Bin: Hangzhou has always been very open. Essentially, Hangzhou has a good ecosystem, with the government having a relatively open mindset and capital being able to provide support. Many enterprises have grown here. What Alibaba Cloud or Tongyi does is to create the best models and then open - source them for everyone to use. Through cloud services, we can help startups like DeepRobotics grow faster and reduce the difficulty of their "last - mile" success. Alibaba Cloud serves everyone well, tries to create the best models and open - source them to speed up everyone's success.

The Battle of Ecosystems: How Can Large - Model Startups Dance with Giants?

When the topic shifted to the highly competitive large - model field, the complex relationship between startups and industry giants became the focus. Host Wu Xi first affirmed the "amazing combat effectiveness" demonstrated by Zhipu AI — it has performed well in technological innovation, B2B business development, and financing capabilities. However, he also pointed out that there are several "mountains" in front of it, such as Alibaba and ByteDance. He directly asked Zhang Peng, the CEO of Zhipu AI, about the "way to break the deadlock" in the face of future direct competition.

Zhang Peng: As a startup, what we essentially compete on is "conversion efficiency", that is, how to convert capital and investment into technological innovation and market - promotion energy more quickly. The advantage of large enterprises is that they have a "deep pool" and sufficient depth. The advantage of startups is their "small - scale, quick - response, and flexibility". There is an old Chinese saying, "A small boat is easy to turn around." We can adjust our strategies in a timely manner based on market feedback and our predictions of future technological trends.

Since its establishment six years ago, Zhipu AI has "basically never stopped and has been constantly iterating". This flexibility is the core competitiveness of a small team. In addition, technology is the entry ticket, and the ecosystem is the moat. Another advantage of Zhipu AI lies in accurately positioning and building its own ecosystem. This includes establishing relationships with customers, partners, developers, the academic community, local governments, and even expanding the ecosystem overseas. This ability to "make friends" has helped us overcome many difficulties.

Yang Bin: This large - model revolution is a huge industrial revolution. The market is large enough to accommodate many players. Alibaba Cloud's ecosystem is built on its inherent advantages: a large cloud - customer base (about three million), a natural connection with customers, and a full - stack technology layout from "chips to MaaS to PaaS to API to SaaS".

If there was a story about computing power when we started cloud computing ten years ago, now it's about the intelligence model. We hope to sell intelligence to more customers so that they can create more business value, and this value - creation chain will be shortened. This requires a large and multi - level ecosystem to support it.

The "Cold" Thinking of Embodied Intelligence: Towards a Tool or Towards a "Human"?

As the discussion delved into the cutting - edge hot topic of embodied intelligence, a philosophical speculation about the ultimate goal of robots quietly began. Host Wu Xi first presented an observation: Current mainstream technologies such as the VLA (Vision - Language - Action) model have insufficient capabilities in solving complex - scenario interactions, resulting in problems of "insufficient generalization" and "insufficient cooperation between the brain and hands".

Zhao Tongyang: The current results are not satisfactory. But we must first define the ultimate goal of humanoid robots. If we only regard humanoid robots as a tool, then its competitor is the robotic arm in the factory, which is a market with a ceiling of only "20 billion". The real future of humanoid robots lies in the "trillion - dollar or larger" consumer market. In the family scenario, the tool attribute is not the most important. Instead, the "interaction attribute" and "companion attribute" are more crucial. Therefore, in addition to the VLA, the world model and the ability to understand the world are extremely important. "Just like when we raise a child, we don't teach him how to work right away. We must teach him some social knowledge... What we need is the understanding of the world." Our goal is to create the "human" part of the machine, making the robot have a "human touch", which is its core value.

Li Chao: DeepRobotics positions itself as "technological innovation and industry leadership". We divide the capabilities of embodied intelligence into four levels: embodied walking, embodied navigation, embodied operation, and the embodied brain. The entire industry has only just solved the walking problem.

We have proposed a development philosophy of "laying eggs along the way", that is, "maintaining a route that allows technology to quickly generate value along the way". This means that we won't wait for all technologies to be perfectly mature. Instead, we will prioritize applying the mature technologies to vertical industries to solve practical problems and generate commercial value. For example, our technological accumulation in robot movement and navigation itself represents an opportunity dozens of times larger than the current market. Whether the final form is humanoid or quadruped is not crucial to customers. The key is whether the problem can be solved. Our approach to developing humanoid robots will continue the successful experience with quadruped robots — we will first apply the technology to specific fields such as power line inspection and emergency firefighting.

The views of Zhao Tongyang and Li Chao clearly show two fundamental innovative strategic paths in the field of embodied intelligence. Zhao Tongyang is pursuing a "moon - landing - style" revolution, aiming to achieve a cognitive breakthrough close to AGI, and its success depends on fundamental progress in basic research. Li Chao, on the other hand, adopts a "gradual innovation" strategy, building a sustainable business by solving current problems while accumulating resources and data for future technological breakthroughs.

These two paths are not simply a debate between B2B and B2C. They reflect a deep - seated debate in the entire AI field about the development path: Should we aim directly at the ultimate goal of general artificial intelligence, or should we first build powerful specialized AI tools and hope that they will eventually converge into more general intelligence?

At the Peak of the Wave: Risks and Beliefs

Amid the lively discussion, the guests did not avoid the huge risks and uncertainties under the AI wave.

The host first threw the risk question to Pan Xiaohui, who represents state - owned capital. After all, investing in early - stage AI projects with high investment and long commercialization cycles undoubtedly brings great pressure.

Pan Xiaohui: There is definitely pressure and risk. In investment, risk cannot be avoided; it exists objectively. "The coexistence of risk and return is the real charm of investment." The way state - owned capital manages risk is not to avoid it but to disperse and balance it through investment portfolios. There are two specific approaches: one is to invest in top - tier VC funds like Fortune Capital; the other is to form its own large - scale direct - investment fund and invest in 30 - 50 projects to hedge the risk of failure of a single project through a portfolio approach.

Zhang Peng: If we talk about whether there is a bubble in this field, "there must be some". But I think the ceiling of AI technology is high enough, and it may even be "the last general technological revolution of humanity". Therefore, "you can't over - exaggerate it". It may take time to see when the bubble will burst and correct. What I'm really worried about are two deeper - level risks: The first is the endogenous risk, that is, the subversion of the technology itself. The second is the exogenous risk, that is, security and ethical issues. In human history, we have never encountered something with an intelligence level comparable to our own. How to get along with it, how to use it, and whether it will cause harm are all unknown. Well - known foreign scientists have also called on everyone to jointly sign a statement and suspend relevant research until the dangers of super - artificial intelligence are identified. We can see that there has been a lot of discussion in the industry, especially in the academic community, recently.

Yang Bin: I hold a "cautiously optimistic" attitude. Commercially, I see clear demands for large models in three aspects: automation of knowledge work, popularization of professional services, and personalization of entertainment content. I believe that more unexpected applications will emerge in the future. However, I also agree with the severity of security risks. Alibaba's laboratory is investing in research on "model interpretability" to address this challenge.

Zhao Tongyang: The risks in the field of embodied intelligence are more specific and real, especially in terms of hardware bottlenecks: The battery life is still far from meeting the 8 - hour work system, and the wear and lifespan of machine joints are far inferior to those of humans. But within two years, its usage cost can basically be reduced to 1/3 - 1/4 of the cost of hiring a human. In this regard, it's just a matter of cost - effectiveness.

Li Chao: For a steadily developing company like DeepRobotics, the bubble and hype in the industry "bring more benefits than drawbacks" because it can attract more excellent talents, supply - chain partners, and market attention, thus promoting the healthy development of the entire industry.

This discussion about risks clearly outlines a risk hierarchy: At the application and hardware levels, risks are related to physical laws, engineering implementation, and market acceptance; at the platform level, risks are related to business models and known security management; and at the most basic model level, risks rise to the philosophical level of scientific paradigms and the future of humanity.