Li Mingshun: 70% of the AI ​​market will be open source in 5 years

From the "Hundred Model War" in 2024 to the robot competition in 2025, the iteration of technology is becoming increasingly rapid. Behind this, the underlying logic that determines the discourse power of technology - open source and closed source, the struggle between the two major camps is accelerating.

Taking the large model as an example, ChatGPT-4 under OpenAI has 1.76 trillion parameters and built a high wall with a closed-source model, while Deepseek, which became popular in 2025, reconstructed a new AI order with an open-source ecosystem. There is no right or wrong in the two models, but they are different strategic choices for enterprises. So in the eyes of investors, what are the advantages and disadvantages of the two models? What are the business opportunities? For this reason, the reporter of "Business School" interviewed Li Mingshun, chairman of Xingxing AI and managing partner of Shunfu Capital.

Li Mingshun: When we look at projects, we mainly base them on three core dimensions: people, market size, and product positioning. In terms of people, in the current wave of AI entrepreneurship, almost all members of the entrepreneurial team are highly educated talents with high technical levels and a global vision. At present, more than 80% of the teams in the invested projects have doctoral degrees, and I have personally invested in more than 30 doctoral entrepreneurs. This type of team has a strong technical gene, and most of the members have studied or worked abroad. The current team size is not large, and there are not many people, but everyone is very capable. In the AI ​​era, everyone can act as a generalist and is a compound talent.

In terms of market size, only a large market can achieve great success. In a small market, even if one has strong personal abilities, the development space is extremely limited. Therefore, I generally prefer to invest in entrepreneurial teams in large markets. This is the "track thinking". Currently, artificial intelligence is transforming thousands of industries, and every industry will be reshaped. Only by starting with a large track can you have a chance of success.

In terms of product positioning, we examine whether the core entry point of the project is the industry's "pain point" rather than the "itch point". The way to determine whether a project is a real pain point is to see whether customers are willing to pay for it. For example, our requirements for many entrepreneurs are whether there will be customers paying for their products or services within six months, and whether there will be income and cash flow within six months.

Li Mingshun: Over the years, I have always had a clear attitude - I will not invest in the field of large models, because large models are not something that ordinary startups can do. Even if large companies do it, success is not guaranteed. This is because the global monopoly in the field of large models is very high, and the development of large models requires a lot of capital investment, which is definitely not something that ordinary startup teams can afford.

Therefore, we pay more attention to the development of artificial intelligence in the application field and look for good application scenarios. This wave of artificial intelligence has a wide range of transformations in the industry, just like the Internet more than 20 years ago, it will be well implemented in every industry. In this case, the strongest basic model should be selected. It is very important to have good technology to fine-tune these models and combine them with the data of your own scenarios. 

I think AI applications are China's opportunities. Because AI applications serve scenarios, and China has two outstanding advantages in scenarios: one is manufacturing capabilities, and the other is engineering capabilities. China is the country with the most complete manufacturing industry categories in the world and has the most industrial scenarios. AI has huge room for increasing the added value of manufacturing. At the same time, China is also the country with the largest number of engineers. After integrating AI technology, China's engineering and software capabilities will create a higher level of intelligence. Therefore, AI applications are definitely China's biggest development opportunity and a manifestation of China's new quality productivity.

Li Mingshun: In fact, open source and closed source are not simply business choices, but a reflection of cost and technical implementation. In certain situations, AI models cannot be used in many scenarios. For example, in industries involving privacy data and high attention to security, customers are reluctant to provide data to large models because once provided, data sovereignty will be lost. Open source solves this problem perfectly because open source models can be deployed on local devices to build private models, which is an important reason why many commercial applications can be implemented.

In terms of cost, open source technology has greatly reduced costs. For example, one technology has reduced costs by 20 to 30 times. This has enabled many industries that were originally difficult to calculate in terms of return on investment to now be able to calculate reasonable returns. Industries with extremely high cost requirements, such as e-commerce and toys, would rather choose manual labor than adopt new technologies if the costs are too high. But now, as open source technology has reduced costs to consumer levels, the demand for inference has grown rapidly, which is exactly the opportunity brought by open source. Therefore, open source and closed source are not just based on technology preferences, but a comprehensive consideration from the perspective of data sovereignty and cost. It is more of a business choice.

Li Mingshun: In the early days, many large companies were keen on making closed-source models. For example, Baidu in China and OpenAI in the United States. Large companies believe that the closed-source model can create the greatest commercial returns and benefits. To a certain extent, closed-source is a way for companies to calculate costs and benefits more easily. It is relatively easy to calculate how much investment can be made and how many future interface calls can recover the cost. Therefore, companies are more willing to invest in this. Especially now, when many people make vertical models and small models, they are sometimes reluctant to open source, because in their unique fields, closed source helps to maintain better monopoly.

However, open source has a different logic. Taking Deepseek as an example, I think it is an ecological company rather than a product company. It ranks among the top in the app store rankings in more than 190 countries around the world. If it were a closed source company, it would be difficult to achieve such results. On the other hand, its open source ecology can not only attract many large domestic companies, such as Tencent, Alibaba and Baidu, but also overseas, such as Amazon, Microsoft, Meta, etc. are also using it. This means that both ecological partners and competitors are using its open source results, which greatly enhances the openness of the ecology. And after open source, there will be more community developers to provide support in technology research and development, which will help the continuous improvement and progress of products. So, open source and closed source have their own advantages, but I personally prefer open source.

Li Mingshun: Open source is essentially a commercial behavior, and it should not be simply understood as selfless dedication. Open source is actually a very effective business strategy. More than 20 years ago, when we were doing Discuz! (Community Power), we chose free open source, which enabled 70% to 80% of Chinese websites to quickly use Discuz! to build communities. On the surface, it seems to be doing charity, but in fact it has gained a high market share. With this market share, we have obtained more financing and built a new business model.

For China's AI industry, I think it is a diversified balance, and the two ecosystems of open source and closed source coexist with each other. This is a good ecological model for the richness and diversity of the AI ​​industry in a huge country like China.

At present, more Chinese companies choose open source, which I think is appropriate. China, like the US market, is a large market that can accommodate diversified models, which is extremely beneficial to the development of the open source model. Open source helps China and European and American countries to carry out more extensive cooperation under ecosystems such as DeepSeek. Taking the current situation as an example, the West has launched the so-called "Democratic AI Alliance" in an attempt to marginalize Chinese AI and let everyone use its designated products (such as OpenAI-related products). However, because we adhere to the open source concept and open up the code and model operation methods, now even some European and American countries have begun to adopt our technology, which not only avoids the blockade of technology, but also expands the coverage of the technology ecosystem, which is extremely valuable.  

From the perspective of application richness, the open source model has more advantages than the closed source model. The closed source model easily leads to application homogeneity, because closed source products have low plasticity and changeability, and developers often just call simple interfaces. Under the open source model, developers have more room for fine-tuning, and products can adapt to more diversified forms. For example, we deploy some private models locally, which can better meet the application needs of personalization or specific scenarios. In the future, in the field of Internet of Things (IoT), the application of local models will also make many devices more intelligent. Therefore, open source can promote the diversified development of applications.

For entrepreneurs, I strongly recommend that they embrace the open source model. The goal of entrepreneurs is to create unique products. If they rely on traditional large-scale closed-source models, it is difficult to achieve differentiation. Developing products based on the open source model can better reflect the richness and uniqueness of the products. Of course, the closed-source model also has its application scenarios in the financial, medical and other industries.

Li Mingshun: Before this wave of AI technology development, there were already some financial and medical scenarios that cooperated with closed-source models. At that time, relevant institutions would desensitize the data, and many data would not be directly transmitted to the server of the large model, but would be transferred after data "cleaning". This was a feasible solution. However, with the emergence of open source models, this situation has become less and less common. Nowadays, people prefer to build completely autonomous models locally, which has also prompted the financial and medical industries to apply large models more boldly. It is obvious from recent news reports that many hospitals and banks no longer prefer traditional closed-source models.

Regarding the competitive situation between open source and closed source models in 3-5 years, I believe that the open source model will occupy a larger share, possibly reaching more than 70%, and the closed source model will have a relatively small share, showing a "70:30" situation.