The Wuzhen Signal of the Six Rising Stars: AI Entrepreneurship Enters the Era of Competing in Scenarios from Competing in Models
The misty Wuzhen always serves as an excellent backdrop for technological narratives. In 2025, the World Internet Conference kicked off here once again. This year, what attracted the most attention was not the new releases of tech giants, but a round - table discussion called "The Dialogue of Six Rising Stars in Wuzhen".
DeepSeek, Unitree Robotics, Coohom, BrainCo, DeepBlue Robotics, and Game Science - these six companies, known as the "Six Rising Stars of Hangzhou", made a joint appearance. This was their first collective public statement, and it was also the most notable collective debut of Chinese entrepreneurs after the hype of the AI wave subsided.
These six companies span five fields: embodied intelligence, brain - computer interfaces, spatial intelligence, large models, and game culture. Although they seem to be on different paths, they actually share the same logic: to make AI understand the world anew.
From the data stacking in the Internet era to the cognitive construction in the intelligent era, and from the traffic - driven logic to the intelligence - driven logic, this is the real turning point for Hangzhou and the domestic AI startup ecosystem. Hangzhou, known as the "Origin of Digital China", is experiencing its second leap from the Internet to intelligent manufacturing.
Data shows that as of September 2025, the core AI industry in Zhejiang Province achieved an operating income of 494.4 billion yuan, a year - on - year increase of 22%. The R & D expenditure reached 39 billion yuan, a year - on - year increase of 14%, and key technologies continued to make breakthroughs.
This joint appearance of the "Six Rising Stars" is not just about six companies. It is a collective portrait of a city, a generation of entrepreneurs, and an innovation mechanism in the AI era.
Six Paths Converge in Wuzhen
The story of Unitree Robotics is almost a microcosm of the rise of embodied intelligence in China. When the company was founded in 2016, it only had three employees. Now, it has grown to over a thousand.
Wang Xingxing, the founder and CEO of Unitree Robotics, recalled the rapid evolution of robots: "This year, especially in recent months, many robots can dance or perform. The reason behind this is simple. The global community has made many excellent achievements in the robotics field through joint efforts."
The breakthrough in the robotics industry comes from "global co - creation". Wang Xingxing pointed out that China's strong manufacturing capabilities have provided a solid foundation for the development of the robotics industry. Now, Unitree Robotics' products are sold worldwide, and many top - tier laboratories, universities, and companies have started to develop software on its robots.
Industry data confirms the booming development of China's robotics industry. In the first three quarters of 2025, the revenue of China's robotics industry increased by 29.5% year - on - year. A total of 595,000 industrial robots and 13.5 million service robots were produced, both exceeding the full - year totals of 2024.
Wang Xingxing believes that "AI has accelerated the development of the robotics industry". In the field of embodied intelligence, "it's already like a dream, as many things from science - fiction movies have become reality". Looking to the future, he confidently predicted: "It's imaginable that there will be even more surprises in the robotics industry next year or the year after."
Han Bicheng, the founder of BrainCo, started his entrepreneurial journey at Harvard University. "I often worked until two or three o'clock in the morning," he said. He and his team "spent a long time researching brain - computer interfaces and later found that this technology could really bring about many changes to human life, such as helping mute children speak again."
In 2018, Han Bicheng decided to move the company back to Hangzhou from the United States, and the enterprise has since developed rapidly. He started by addressing the most urgent needs and planned the products along a technical route from "solving pain points" to "scaling up".
"Solving pain points" means targeting groups such as the disabled who are in the greatest need of brain - computer interface technology. BrainCo's neural - controlled prosthetic technology can already help limb - disabled people control prosthetics to perform delicate actions like writing and playing the piano through "thoughts". "Scaling up" means expanding into larger fields. For example, the company will launch sleep products to help insomniacs sleep better.
China's 14th Five - Year Plan (2021 - 2025) has included "brain - computer interfaces" in the six future industries to be vigorously developed. Han Bicheng said: "I am becoming more and more confident in doing this well."
Looking forward to the future of brain - computer interfaces, he said: "Many well - known artificial intelligence theories are inspired by neuroscience. Now, we are using AI to help humans solve difficult brain - science problems, which forms a wonderful closed - loop."
Huang Xiaohuang, the co - founder and chairman of Coohom, saw that spatial intelligence will be an important field after large language models and is an important infrastructure for robots to operate in the physical world.
"The data scale of the Internet will reach its limit, and AI represents a new stage after a certain amount of data accumulation," Huang Xiaohuang believes. In the future, the world will be full of robots, and each person may even have multiple robotic servants. "At that time, there will be a large number of robots not only in living environments but also in work environments. Spatial intelligence technology will be needed to manage and command these robots to better serve humans."
It is understood that the Coohom team discovered a few years ago that the scaling laws of large language models also apply to spatial reasoning, but there were not many application scenarios and products at that time. The development of the robotics industry is opening the door for the accelerated application of spatial intelligence technology.
Zhu Qiuguo, the founder of DeepBlue Robotics, felt the most about the "changes" in the past decade. He recalled: "It was a novelty to bring a robotic dog to an exhibition ten years ago, but now robotic dogs are everywhere at exhibitions." Now, they not only manufacture robotic dogs but also start to develop humanoid robots. Zhu Qiuguo said: "This year, we launched a humanoid robot. Our ultimate goal is to replace and assist humans in some dangerous scenarios."
The product R & D follows the concept of "starting with the end in mind" and is closely integrated with the development of specific application scenarios. This year, DeepBlue Robotics launched the world's first industrial - grade humanoid robot that can operate outdoors.
Zhu Qiuguo emphasized: "Humanoid robots are ultimately meant to replace and assist humans in dangerous, harsh, and complex scenarios. In the future, robots will become more protective and move outdoors more often. They should be able to withstand bad weather and even falling into ditches."
Feng Ji, the founder and CEO of Game Science, created the globally popular game "Black Myth: Wukong", which became a cultural phenomenon in 2024. However, in his view, Game Science's success was built on the "shoulders of giants".
Feng Ji mentioned that looking back at the development of the Chinese game industry, over the past decade or more, China has developed into the game market with the largest number of users and the largest market size in the world. This industry has nurtured a large number of users and talents.
He specifically shared a set of data he observed: Among the top 10 most profitable games globally last year, 4 were developed by Chinese teams, and another 3 had Chinese investment or participation. "We are just a'splash' on this foundation," Feng Ji said. Maybe Game Science chose the right theme, the right time, and the right business model, but we should not forget that there are many other successful game companies in China. It is the collective innovation of all enterprises that drives the industry forward.
Since its establishment, DeepSeek has chosen the open - source route, aiming to achieve AGI (Artificial General Intelligence). Through its open - source strategy, DeepSeek provides a reproducible and affordable "software - compensating - for - hardware" technical solution for global developers.
Chen Deli, a senior researcher at DeepSeek, emphasized the value of open - source in the dialogue: "In our communication with technical developers in the community, we received a lot of meaningful feedback, which in turn promoted the development of DeepSeek itself. Therefore, we always believe that open - source technology can bring better cooperation and sharing."
Regarding the future of AI development, Chen Deli believes: "We are still in the first half, or even the early stage of the first half, of this AI revolution." He pointed out that there are many problems and bottlenecks in current artificial intelligence, but technological development often has an "acceleration effect". "If we look ahead 10 - 20 years, in the process of moving towards AGI, once certain key nodes are crossed, AI technology will experience an explosion."
Challenges after the Technological Frenzy
In the past few years, Chinese AI entrepreneurs have achieved a difficult breakthrough from scratch in the highly competitive environment. According to Tianyancha Professional Edition data, the number of AI - related enterprises in China has exceeded 4.243 million. From January to October 2025, there were 139 financings of over 100 million yuan in the AI field, with a total amount exceeding 60 billion yuan.
However, when the technological glow began to hit the ceiling of reality, at this year's Wuzhen round - table, the leaders of the "Six Rising Stars" stopped talking about parameters and computing power. Instead, they started to think about what is the real power to survive the cycles after the technological frenzy?
The investment logic in AI is also changing. According to IT Juzi data, in this year's financings of over 100 million yuan, embodied intelligence has surpassed large models as the most attractive sector. 73 enterprises received over 20 billion yuan in financing, accounting for one - third of the total financing amount. Capital is shifting from blindly chasing concepts to favoring enterprises with revenues, orders, and mass - production capabilities.
The head - effect is becoming more prominent. Yuezhianmian, Zhipu, and Minimax have received over 2 billion yuan in financing and have a combined valuation of about 20 billion yuan. More startups must prove their commercialization capabilities.
Zhao Zhanxiang, the founding partner of IO Capital, said: "Both the mature routes that can be implemented immediately and more cutting - edge directions like in - memory computing and photonic computing are worthy of attention."
During the Wuzhen dialogue, Wang Xingxing, the founder of Unitree Robotics, mentioned that the rapid development of the robotics industry not only benefits from China's strong manufacturing capabilities but also from the open - source algorithms of robot large models.
This represents a shift in the positioning of startups, from technological isolation to ecological co - construction.
The transformation of DeepBlue Robotics is more representative. Its founder, Zhu Qiuguo, admitted that around 2018, the team was obsessed with solving technical problems such as the wear and heat of robotic dog rubber pads. Now, they are more concerned about adapting to real - world application scenarios. "We launched a humanoid robot, and our ultimate goal is to replace and assist humans in some dangerous scenarios."
"Laying eggs along the way" has become a common development philosophy for many companies. This means not waiting for all technologies to be perfectly mature, but applying the already - mature technologies to vertical industries to solve practical problems first.
The rise of the "Six Rising Stars" is not accidental. Pan Xiaohui, the chairman of Hangzhou Industrial Investment Group, said, "We didn't do anything; the 'Six Rising Stars' emerged on their own." An academician from Zhejiang University once summarized Hangzhou's competitiveness as "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.
Open - source co - creation is the key to ecological construction. Since its establishment, DeepSeek has taken the open - source route. Chen Deli admitted: "In our communication with technical developers in the community, we received a lot of meaningful feedback, which in turn promoted the development of DeepSeek itself."
Coohom's experience also confirms the value of open - source. The company open - sourced a spatial dataset in 2021, which opened up new space for its development in the field of spatial intelligence.
Yang Bin, the person - in - charge of the product solution of Alibaba Cloud's Tongyi large model, expressed a similar view on the ecosystem: "What Alibaba Cloud does is to create the best models and then open - source them for everyone to use."
Despite the promising prospects, the development of embodied intelligence still faces many challenges. Taking data collection as an example, Wang Xingxing pointed out that compared with large language models, which have abundant Internet data, the robotics field faces problems such as insufficient model structures and data volume. The differences between robots from different manufacturers lead to inconsistent data collection, and there is no unified consensus on aspects such as tactile and visual applications and camera installation positions.
Commercialization is another bottleneck. The NANDA project of the MIT Media Lab released a report titled "The Generative AI Gap: The State of Business AI in 2025". The report pointed out that although enterprises have invested 30 - 40 billion US dollars in generative AI projects, 95% of organizations have not obtained any commercial returns.
Facing the rapid development of AI, technology - innovation enterprises are also clearly aware of the importance of technical regulations. Feng Ji, the founder of Game Science, analyzed that on the one hand, the AI revolution empowers individuals; on the other hand, we should be vigilant against the risk of excessive concentration of technological advantages due to technological development.
Chen Deli analyzed from a time - dimension perspective: "In the short term, human - machine collaboration can achieve a '1 + 1>2' effect. But in the medium - to - long - term, some jobs will indeed be replaced by AI. Technology companies should play the role of guardians."
Zhang Peng, the CEO of Zhipu, admitted that there is a bubble in the industry, but he believes that the ceiling of AI technology is high enough, and it may even be "the last general - purpose technological revolution for humanity". What he really worries about are two deeper - level risks: endogenous risks, i.e., the subversion of the technology itself; and exogenous risks, i.e., security and ethical issues.
Despite these challenges, the AI entrepreneurs at the Wuzhen round - table are still confident about the future. "Once technology crosses the key nodes, it will experience leap - forward development," many participants said. The development of AI is irresistible, and promoting trustworthy AI requires the joint efforts of technology and regulation.
When AI Moves from Understanding Language to Understanding the World
At the Wuzhen dialogue, the "Six Rising Stars" talked about their company stories, but their paths point to the same direction: AI should move out of the text - based world and enter the physical world.
Shortly after the Wuzhen Summit ended, Fei - Fei Li, a professor at Stanford University and known as the "Godmother of AI", published a long - form blog post. In the post, she elaborated on the concept of spatial intelligence and stated that the development direction of AI in the next decade is to build machines with true spatial intelligence.
The post also discussed the essence of the shift of AI development from the information space to the physical space. AI should not only understand language; it must understand the world. It should not only generate content; it must be able to act. If GPT - like language models have driven the revolution in the "cognitive layer", then the world model points to the full - chain ability of "understanding - predicting - acting".
The "Six Rising Stars" are particularly symbolic at this moment. Fei - Fei Li proposed that the key to the world model consists of three parts: spatial understanding, action ability, and environmental interaction. If we apply this framework to the six Hangzhou - based companies, we will find an unexpected but reasonable correspondence.
Coohom is building AI's spatial understanding, Unitree and DeepBlue Robotics are building action abilities, BrainCo is building the perception and decoding of human intentions, Game Science is creating a virtual world as a training environment, and DeepSeek is providing a knowledge base through open - source models.
These six paths may seem scattered, but they actually form a more physical - oriented future AI framework. Chinese AI entrepreneurs seem to be approaching the ability boundary required by the "world model" in a bottom - up way. The industry is shifting from "understanding text" to "understanding the world".
From a technical perspective, the concept of the world model is not new. However, three changes have truly pushed it towards industrialization:
First, the data in the text - based world is approaching its ceiling. As Huang Xiaohuang, the founder of Coohom, said: "The data of the Internet will eventually reach its limit." To continue to make breakthroughs, AI must enter the physical and spatial worlds.
Second, the large - scale commercialization of embodied intelligence is opening up data sources. The robots of Unitree and DeepBlue Robotics are developing the ability to collect "real - world interaction data", which pure NLP models can never achieve.
Third, China's intelligent manufacturing and game industries provide a natural "dual - track environment" for the world model. On one hand, there is a highly mature