At the banquet, a coal boss told the “Contour” reporter that he did not want to miss the opportunity given by the times - compared with coal mines, the mining and output of artificial intelligence (hereinafter referred to as “AI”) will only increase, and it will benefit children and grandchildren more… He was desperate to know where there was an opportunity to get involved.
In Beijing’s North Third Ring Road, an inconspicuous office building is always packed on Tuesdays for the “Kunlun Nest” entrepreneurship salon—most of the attendees are in their thirties and forties, carrying backpacks, and the topics of discussion invariably focus on AI and entrepreneurship—here, the opportunities that the coal boss is looking forward to are in abundance.
Some call this place the “barometer of AI entrepreneurship in China,” while others jokingly refer to it as the “last bastion of idealists.” From technical backbones fleeing big firms, serial entrepreneurs, traditional industry players transforming their careers, to freshly graduated university students, they all carry their enthusiasm and confusion about AI, trying to find their own answers amidst the waves of algorithms and capital.
Unlike the previous wave of internet entrepreneurship, there are no suited investors here, nor are there grand IPO declarations. Instead, there is a pragmatic and anxious atmosphere—how to quickly monetize using AI tools? How to survive in the crevices of giants?
This is a microcosm of the AI entrepreneurship wave in China. Since the global AI frenzy ignited by ChatGPT in 2023, AI technology has iterated at a “leapfrog” pace over the past two years, and the entrepreneurial ecosystem has gradually returned to rationality from its early fervor. At this time, capital is more cautious, technical thresholds are higher, and opportunities are more segmented.
In the Kunlun Nest, entrepreneurs are discussing the prospects of AI entrepreneurship. Photo by Liu Wang from the magazine “Deng Shen Xian”.
AI reshapes the boundaries of creation
Input a description, and the AI tool directly generates an image or video; tap a few notes on the keyboard, and the AI composition automatically generates a piece of music… Such products are no longer a novelty in the current cultural industry.
“AI will change the cultural industry,” said writer and screenwriter Fu Yao to reporters. Fu Yao was once a sales consultant for the telecommunications industry at IBM China, also responsible for sales in the northern region for Dell, and later provided consulting and training services for several large enterprises. His novels “The Era of Startups” and “Win or Lose” have both been adapted into popular TV dramas.
“In the past, adapting a novel into a script required 3 to 5 screenwriters and took six months or even two years. Now, AI might be able to generate a first draft in just 10 minutes.” Fu Yao spoke with a hint of astonishment about the AI-driven process in the film and television industry.
He told reporters that the traditional film and television industry chain is lengthy, and the adaptation from novel to screenplay is particularly complex— a 200,000-word novel needs to be expanded into a 600,000-word script, with psychological descriptions needing to be transformed into dialogue, where both logic and emotion are indispensable. The intervention of AI is attempting to solve this problem.
Fuyou mentioned that not long ago, he helped a large entertainment company train an AI large model. The company gathered 100 screenwriters to “feed” the model, analyzing the narrative structure of classic dramas to train the AI to generate script frameworks. “It excels at strong logical plots, such as murder cases. Input ‘bathroom murder’, and the AI can generate different variants based on the training data, even able to design segments like ‘how to make the police unable to find the murderer,’” Fuyou explained.
But AI is not omnipotent. “Romantic dramas? They are still not well-written. Emotions are quite subjective, and the traces of AI imitation are obvious,” Fu Yao said.
In the film and television industry, the application of AI has penetrated areas such as animation rendering and costume design, but core creativity still relies on humans. “For example, in script generation, AI can quickly provide a framework, but how to make the story resonate emotionally still requires the screenwriter to control the emotional core,” Fu Yao emphasized. The essence of AI’s “thinking” is the logical connection of code to “tokens” (such as “murder”), rather than true creativity.
Even so, the improvement in efficiency is enough to be disruptive. “In the past, preparing a 30-episode series took several years, but in the future, it may be compressed to a few months. Costs can be significantly reduced, but the artistic quality remains uncertain.”
Compared to adapting scripts, AI generating video animations directly based on instructions is currently more convenient and effective. Han Lei, the founder of Smart Manufacturing Future, deeply feels this.
Han Lei’s company has created a mecha sci-fi story called “Spirit Core World”, which tells the story of the birth of an autonomous electronic life form called “Spirit” in the human Internet between 2015 and about 2400 AD, and has had a huge impact on the entire world. In this story, the “spirit” is like an AI with autonomous consciousness.
The journalist saw a more than 300-page album at Han Lei’s place, as thick as a wedding photo album, detailing the faction settings, biological and mecha characteristics, military strength, and character relationships of “Spirit Core World,” depicting an extremely vast mecha world.
“Core World” Setting Collection “Isohyet” Reporter Liu Wang Photography
Transforming a massive, sci-fi and futuristic IP story into various aspects is not a simple task. Recently, a friend used AI animation to produce a teaser for “Linghe World,” which boosted Han Lei’s morale.
A few years ago, the technology was not mature, and it would have required a huge cost to produce a movie like “Spirit Core World.” But now, finding an AI team can probably make it happen, and the cost can be controlled. "Han Lei bluntly stated, "AI can change the cultural industry. Just wait and see, there will be many more powerful IPs than before.
In fact, more and more entrepreneurial teams are focusing on the application of AI in the entertainment industry. The reporter saw in the Zhongguancun Artificial Intelligence Large Model Industry Cluster located in Dongsheng Building that products like Hexinpai and Shenyandayi have already taken shape.
Taking the Harmony faction as an example, it is a one-stop, low-threshold AI music workstation that can perform AI lyric writing, AI composition, AI arrangement, song synthesis, and more. The music works created by core users using this tool have garnered over 100 million plays across the internet.
Finding landing scenarios in the gaps.
In fact, the main direction of current AI entrepreneurship has been most concretely displayed on a wall on the first floor of the Dongsheng Building. The reporter observed that a map of the Haidian Artificial Intelligence Innovation District industrial chain was posted on the wall, covering the entire upstream, midstream, and downstream industrial chain.
In the downstream, entrepreneurs are focusing on AI+ application products, which include voice terminals, smart cars, visual products, robots, as well as smart education, smart healthcare, smart manufacturing, smart finance, and smart security; in the midstream, there are basic open-source frameworks and technology development platforms, as well as various algorithm models and general technologies; in the upstream, there are data services and hardware devices.
Haidian Artificial Intelligence Innovation District Industry Map. Photo by Liu Wang, reporter of ‘Isobath Line’.
However, the majority of startups focus on AI+ application products. A relevant person in charge of the Dongsheng Technology Incubator told reporters that in 2024, the incubator added more than 100 AI startups, most of which are based on AI large models for application development.
Twenty-year internet veteran, Zhou Shangjinhang, the head of Kunlun Nest AIGC Open Laboratory, was once the early technical director of Youku Tudou Group. As the AI wave surged, he chose to start a business consulting service. He mentioned: “The competition threshold for large models is too high; small companies and even medium-sized listed companies can’t handle it. On one hand, training large models is expensive, and on the other hand, the talent requirements are high. Therefore, we can see that there are very few players in large models around the world; others can only create applications on top of large models.”
As a result, one can see a scene where, while people gaze at the starry sky in the AI world, another group of entrepreneurs is bending down to pick up coins from the real world.
“Ten years ago, everyone wanted to be Lei Jun; now, earning 200 yuan a day with AI might be the ‘success’ of an ordinary person.” said an AI entrepreneur.
Mo Xiaoyi is a serial entrepreneur who is applying AI in the financial sector to assist banks in managing non-performing assets. “The scale of any non-performing asset package in banks is quite large now. Previously, non-performing asset packages had to rely on manual analysis of debt composition and default situations to form an evidence chain, requiring auditors to review each case one by one. However, by leveraging AI for batch processing, efficiency can be improved, at least forming a complete data structure and data chain, ultimately creating a whole set of processes.”
Wang Yongwang previously worked as a developer at a metaverse game company. Observing the rapid development of AI, Wang Yongwang decided to join the AI entrepreneurship wave. He is currently applying AI in restaurants, establishing local knowledge bases for each restaurant, interacting with consumers, chatting with consumers, and helping consumers plan meal sets.
Another entrepreneur, Gao Shan (a pseudonym), previously worked in software development and his last job was in overseas trade. In this wave of AI, he is focusing on the silver economy track. “Helping elderly people create a memoir and then building a local knowledge base based on the information in the memoir. The final product will resemble a desktop robot that can narrate their stories in the tone and style of the elderly, interacting with the younger generation,” Gao Shan stated.
The Path of “Breaking the Wall” in Thinking
More people are aiming for the AI education track. Li Shanming is a continuous entrepreneur who, in the past year, has hit two waves at the intersection of anti-addiction and AI enlightenment: on one side is parents’ anxiety about “mobile phone addiction,” and on the other side is the educational revolution brought about by the popularization of AI technology.
At the beginning of 2025, DeepSeek became a sensation, and the public fell into “instruction anxiety”: people could ask questions but could not truly harness AI. During interactions with parents, Li Shanming discovered that the most frequently asked question was: “What can AI really help children learn?” The enthusiastic market response made Li Shanming realize that whether it is children or the elderly, the core of technological inclusivity lies in “breaking the wall” – bridging the cognitive gap between users and AI.
So, Li Shanming designed a set of “gamified AI courses”: children use “Doubao” to input commands, generate tank battle game codes, and then debug to make the tanks fire shells or design a Tetris game. “Hundreds of children have verified that they can get started in just 1 hour.”
Li Shanming showed the AI Enlightenment business group to the reporters. Photo by Liu Wang from “Deng Shen Xian”.
Similarly targeting AI education, but unlike Li Shanming’s popular education, Ke Qiang (pseudonym) is working on helping primary and secondary schools build AI courses and write textbooks, focusing on the development of teenagers’ thinking and logical abilities.
Ke Qiang has his own views on AI education - humans are responsible for strategic thinking, while AI executes tactical details. To this end, he has abandoned the previously popular training of “prompt engineers” and instead emphasizes logical expression: “Teaching prompts today may be obsolete with tomorrow’s technological iterations. But logic is eternal; it determines whether a person can have an equal dialogue with AI.”
Currently, the understanding of AI in the education sector is still insufficient, and those engaged in technology development do not have a complete understanding of education. This has made Ke Qiang see market potential: “Now, various schools and education commissions are looking for people who understand both education and AI, and our team happens to have the capability to do this work.”
Although the niche markets targeted by the aforementioned grassroots entrepreneurs are different, they share a common point: their startup costs are not high. Wang Yongwang and Gao Shan are currently both working alone, Li Shanming has also clearly stated that he will not expand his team, while Ke Qiang’s team consists of only four people.
Pan Tongdan, deputy secretary-general of the China Mobile Communications Federation Computing Ecological Alliance, told reporters that among grassroots entrepreneurs, there is a special word, called a one-person company, and one person can complete the business closed-loop of the entire company. On the one hand, AI reduces labor costs and improves work efficiency. On the other hand, AI can also help people quickly produce new products, and then sell them on online marketplaces, which can quickly generate revenue.
Similar scenes are most intuitively displayed in various incubators. Journalists visiting multiple incubators in Beijing found that many startups are basically small teams of two or three people.
Ke Qiang bluntly stated that in the current market environment, the cost of starting a business cannot be high; you must utilize minimal social resources, minimal funds, and a very small team to complete the entrepreneurial process. “This is a prerequisite that tests whether every entrepreneur can get started.”
Digging for “hardware”
In fact, apart from the applications of private deployment of large models and AI education, hardware products are more favored by capital because they lower the threshold for AI usage.
The AI all-in-one machine is in a state of explosive popularity. A report released by Peking University mentioned that the marginal cost in the AI era is significantly higher than in the Internet era (requiring continuous computing power and token consumption), while the all-in-one machine can alleviate this pain point through “software-hardware synergy + localized deployment.”
The reporter learned at Kunlun Nest that the humanoid robot open laboratory has designed an all-in-one product named DeepBook. In terms of appearance, it is a black square box, priced below ten thousand yuan, and has already sold over a hundred units.
DeepBook All-in-One. Photo by Liu Wang from “Isobath Line”.
According to Dr. Guo Chengkai from the Kunlun Nest Human-shaped Robot Open Laboratory, most of the current buyers are state-owned enterprises, universities, or companies that want to avoid the risk of data leakage. They create a knowledge base from the corresponding data of the company, load it into the integrated machine, and deploy the integrated machine in the company’s server room, allowing direct access to interact with the company’s knowledge base.
In addition, there are robotic arm products. Yang Kai, a member of the Kunlun Nest humanoid robot open laboratory team, told reporters: “Our robotic arm is also in a semi-commercial state. We have developed two generations of products and will soon update to the third generation, which will be available for external sales. However, one of the member units of the laboratory, the ‘dexterous hand,’ had a huge sales success last year, priced at over 60,000 yuan, with about two to three hundred sets sold.”
Yang Kai mentioned that with another 6 to 8 months of research time, robotic arms can be applied to related industries, such as the service industry and control panels, and can also be combined with robots.
However, Yang Kai also stated that for embodied intelligent robots to develop into the kind seen in movies, it will take at least 5 to 10 years. “Not only must they reach human-level understanding, but they also need to achieve human-level behavioral capabilities. However, hardware will definitely lag behind the development of large models, as hardware must be based on the computing power provided by large models.”
"Some of the current technologies of robots are still immature, and many of them are experimental. For example, the humanoid robot at the end of the current terminal has a carrying capacity of 10-20 kg, which does not meet the needs in many scenarios. Another example is a service robot, which is not enough to help an elderly person go to the bathroom or pull the elderly person out of a wheelchair. Yang Kai said that this kind of physical problem cannot be solved by software updates, and the entire industry has to iterate.
During the interview, Yang Kai casually showed the reporter a small white square box, about the size of an AirPods case, with a metal button on top. Pressing it allows for conversation, similar to a portable version of “Doubao.” The reporter pressed the button and asked, “How did dinosaurs go extinct?” Soon, a voice provided a relevant response.
Yang Kai introduced that the original intention of this product is to provide emotional companionship and academic guidance for middle school and elementary school students. “There are currently quite a few products like this on the market, but there is still a lot that can be done in the future, such as setting different roles and personalities, and then binding different knowledge bases according to different roles, which would make conversations much more interesting.”
The aforementioned AI small box can engage in conversation by pressing the button on the top. Photo by Liu Wang, reporter of “Deng Shen Xian”.
The Capital Maze of Grassroots Entrepreneurs
According to data released by IT Juzi, in the past decade, the overall scale of financing in China’s artificial intelligence industry primary market has expanded from 30.07 billion yuan in 2015 to 105.251 billion yuan in 2024, achieving a growth of 3.5 times. In the overall financing scale of 2024, early-stage investments account for 42%.
Among the hundreds of billions in capital, the proportion of grassroots entrepreneurs receiving investment is extremely low—market preference leans more towards teams with efficient laboratory transformation, top-notch research backgrounds, or those from large tech companies.
The reporter learned that large-scale business incubators generally provide entrepreneurs with a complete full cycle of services, such as Teli Station, which divides the growth of enterprises into three key nodes, namely the R & D stage, the market and growth stage, and the mature stage, and will provide angel funds, industrial capital, mergers and acquisitions/IPOs and other different supports at different stages of enterprise development.
The Dongsheng Industrial Park will hold entrepreneurship competitions from time to time, and the winning teams will receive certain policy support. In addition, there are regular salons featuring relevant ecological companies, investment institutions, entrepreneurial mentors, and university resources.
The Xingdi AI Application Incubator in Chaoyang District, Beijing, jointly initiated by seven well-known investment institutions, including Innoangel Fund, Plum Ventures, Lenovo Capital, Tsinghua Alumni Fund, Oasis Capital, Xiaomiao Langcheng, Yuanhe Origin, and over 20 ecological resource partners, provides capital market consulting and training for startups, helping them solve the problem of difficulty in raising funds in the early stages.
The “full lifecycle” services of a certain startup incubator. Photo by Liu Wang of “Equal Depth Line”.
However, not all enterprises can successfully settle into the incubator. For instance, the Xingdi AI Application Incubator considers various factors when selecting entrepreneurial teams, including project management teams, technical maturity and innovation, market prospects, business models and operational results, and social responsibility.
In fact, when incubators select entrepreneurial teams, they share a similar perspective with capital institutions, where “people” are their key consideration. A person from an investment institution told reporters: “The most important aspect we look at in projects now is the team. The best combination is to have university research resources, along with talents who understand the market and marketing.”
"In the last round of entrepreneurship, grassroots entrepreneurs can basically get investment, at most tens of millions of yuan, as little as one or two million yuan, which can support an entrepreneurial team to engage in research and development. However, at this stage, many investors’ cash flow is very tight, and they will not invest because of the entrepreneur’s dream, so it is basically difficult for grassroots entrepreneurs to get money. Ke Qiang said.
Su Dian, the founder of Garage Coffee and Kunlun Nest, also mentioned that the current AI application layer is facing a dual dilemma: accelerated technological iteration has led to unclear investment directions, while narrowed exit channels have intensified funding anxiety.
In summary, compared to the investment frenzy in the early days of internet startups, when a single PPT could secure funding, that era is long gone.
The aforementioned investment institution personnel told reporters that currently many investment institutions are investing in hard technology manufacturing, focusing more on upstream investments, such as semiconductor chips, upstream chip materials, and the end products are similar to embodied intelligent robots, involving hard technology-related content in each link of the industrial chain.
"However, from the large model to the application, there are many domestic mainstream institutions waiting and seeing, but there are not many who really start, because the iteration of the large model is very fast, and what is being voted now is not the real technical direction in the future, and the probability is not so large, so everyone is still in a wait-and-see attitude. The same is true at the application level, where AI is changing very fast, and it is difficult to determine which application can actually run out. It is likely that it will succeed in the future, and it has not yet come out at all. The above-mentioned investment institutions said that now it is also a process of learning and running-in for investment institutions.
Liu Yang, a founding member of Baidu Venture Capital and with fifteen years of venture capital experience, shares the same view: “In the mobile internet era, people are very certain that smartphones can completely replace PCs, and the mobile internet can geometrically amplify the world of the previous generation of the internet. Some projects may seem like failures today, but the overall direction is correct. However, today, AI has not clearly conveyed its future direction. Up to now, there is still no clear indication of what can create significant commercial value; rather, the overall landscape is still in exploration.”
The Change in Entrepreneurial Logic
Clearly, unlike the logic of “burning money for scale” in the era of mobile internet, today’s entrepreneurs need to confront cash flow pressures head-on. “Projects without the ability to generate revenue won’t survive beyond three months,” Han Lei stated.
In fact, Han Lei experienced a long period of career growth. In 2015, a mech robot named NK01 made its debut at the Hunan TV New Year’s Eve Gala, creating quite a buzz. This robot was designed by Han Lei’s company, Intelligent Manufacturing Future. Today, you can still find news online about this product’s appearances at various music festivals and variety shows.
In 2017, the Future Intelligent Manufacturing Company launched a large manned mech called “Shenshui Yuan XX21”. This mech stands 3.5 meters tall and weighs 4500KG, capable of carrying 2 people at the same time. Once inside the mech, a person can even drive and fire the cannon.
According to Han Lei, the leasing business of the NK01 prototype in 2015 had quite good revenue. “However, due to a lack of systematic commercial awareness, the business model went off track, focusing on cultural tourism instead of the robot performance market.” In Han Lei’s words, “A ‘big piece of meat’ is right in front of us, but we didn’t cherish it properly.”
Moreover, it is worth noting that many entrepreneurs today are no longer obsessed with “disrupting giants,” but rather focus more on “ecological symbiosis.” Liu Yang stated, “In the era of mobile internet entrepreneurship, people have seen too many stories of giants fighting each other and entrepreneurs disrupting giants. However, in today’s AI entrepreneurship era, more entrepreneurs will choose to explore the future business ecosystem together.”
Zhou Shangjin Hang also mentioned: “Instead of complaining about big companies picking the fruits, it is better to find the niche markets they overlook.” For example, some teams focus on customizing private AI knowledge bases for enterprises or developing specialized tools using open-source models, which not only avoids direct competition with large companies but also creates a differentiated advantage.
“The shift in entrepreneurial logic is not a decline, but a maturation.” Su Lin summarized, “When the era of crazy expansion ends, surviving and creating real value will be the main theme of the new cycle.”
(Source: China Business Journal)
Source: Eastmoney.com
Author: China Business News
