"We know that an era of AI transformation is coming, but we don't know what to do now." This may be the confusion faced by many companies.

After the Spring Festival in 2025, the phenomenal technological breakthrough DeepSeek was born, various AI concepts were flying around, and a large amount of AI information continued to bombard. Many CEOs wondered if the company was too quiet and if the internal staff's thinking had not kept up with the times. The discussion outside was in full swing, so why hadn't the technical staff taken action yet?

Li Kefeng, founder of Shushi Technology, said that he had recently participated in some business school exchange activities and could clearly feel the anxiety of CEOs from all walks of life. Everyone was afraid of missing out on an era. CEOs were worried that the barriers formed in the past would no longer work in the new era. How should they embrace DeepSeek and how can they avoid being eliminated?

Li Kefeng was once the chairman of the JD Group Technology Committee, the technical director of JD Mall, the general manager of Baidu Mobile, and the CTO of Ping An Financial. He is also a person who has always been at the forefront of technology in the Internet era, the mobile Internet era, and the AI ​​era. Because Shushi Technology, which he founded in 2020, is a business based on big data + AI to help companies with digital transformation. So recently, Tencent News' "Altitude 7950" column invited Dr. Li Kefeng, the founder and CEO of Shushi Technology, to talk about many hot topics in the field of AI.

The following is part of the conversation between "7950 Altitude" and Li Kefeng:

What are CEOs worried about?

"Altitude 7950": As far as I know, many CEOs of companies are particularly anxious recently, including CEOs of various industries and CEOs of AI companies. It seems that they are afraid of missing out on an era. I wonder how you feel?

Li Kefeng: Indeed, FOMO (Fear of Missing Out) is a very popular word in the capital circle. Now it has also spread to the entire industry circle. Everyone is afraid of missing out on this era.

These companies fall into two categories: The first category is traditional software companies. People often talk about how the DeepSeek and AI Agent eras may overturn traditional software. Will the original UI-based software be eliminated?

This is one part. Another part is AI companies. The anxiety of AI companies is not whether they will be eliminated in the direction, but now with the birth of DeepSeek, the cost and threshold of doing this are getting lower and lower. A large number of companies have emerged to develop AI applications. Will this field become a red ocean? CEOs are worried about this.

On the other hand, entrepreneurs of some traditional companies in non-technical industries are worried about whether their business models will be disrupted by the new generation of AI technology based on large models.

Will the barriers formed in the past disappear in the new era? Most entrepreneurs are still relatively traditional. The amount of information bombardment every day is very, very large. Various new terms emerge in an endless stream, various new ideas, and various new models are constantly explained through self-media and KOLs. Everyone is wondering how far these things are from me, how I can absorb such technology, and let our companies embrace the new era. This is their anxiety. Indeed, everyone is in a state of anxiety.

"Altitude 7950": How do your entrepreneur friends communicate with you?

Li Kefeng: A lot. Last week I attended a business school event in Shanghai. There were about 20 entrepreneurs at the meeting. I went to share the AI ​​big model and DeepSeek-related technologies. The discussion was very heated and seriously exceeded the time limit.

Every entrepreneur will ask a lot about how to combine big models with their business, including retail consumption, overseas business, manufacturing industry, and new materials business. You will find that no matter which company, the top leaders of the company know that big models are the core of the next generation of technology cycle.

At the same time, we also found that the IT staff in their own companies have not kept up with the latest developments, so there is a big gap. The outside world is discussing this so enthusiastically, but why haven't people in our own company taken action on technology or technology-related positions? This huge contrast is the reason why they are more anxious and constantly seek new solutions and ideas from outside.

"Altitude 7950": What kind of era are we in? Why have such phenomenal technological breakthroughs continued to appear in the past few months, from DeepSeek to Manus? Is it because the technological potential has reached an explosive period?

Li Kefeng: This is a very good question. When we find that something is developing much faster than we previously knew, the fact is that "it" has already arrived. If our actual industry lags behind our expectations, it is often because the problem inside has not been truly solved.

People have had expectations for AI since ten years ago. But now, have the expectations of both AI companies and the companies they serve been met? Actually, no. In the past, reality lagged behind everyone's expectations.

Now, on the contrary, wow, there is something else that is beyond people's expectations and beyond the scenarios that people can think of. Including the birth of Manus, which is regarded by many self-media as the next DeepSeek Moment, but this is also controversial. We can talk more about it in the future.

Including the various industries that I can see now, everyone is talking about it, and there are many new solutions in vertical industries. This potential energy does exist, and its fundamental reason is that we have to go back to its most basic logic, which is the emergence of DeepSeek. From the Spring Festival in 2025 to now, in two months, you will find that many changes have taken place every day, and its foundation is all around DeepSeek. Whether it is the new C-end like Yuanbao, AI search, or other products, it is based on DeepSeek. DeepSeek is the driving force behind all the changes, this is my understanding.

"Altitude 7950": I know that you have a deep understanding of DeepSeek. Currently many companies have connected to DeepSeek. What is your new understanding of DeepSeek now?

Li Kefeng: Since the Spring Festival, I have been communicating with customers non-stop. Our products were also the first to be connected to DeepSeek, and customers are very happy to communicate and cooperate with us. There are several phenomena:

First, after the Spring Festival, the most important thing for almost all companies of a certain size is to fully deploy DeepSeek.

Second, DeepSeek is used in the scene. So we can see that its biggest variable is that everyone has started to use it, and it has gradually been implemented in some scenes.

After ChatGPT came out in late November 2022, various industries were more interested in watching the fun, saying OK and trying to understand the big model, but few actually took action. Only a handful of industries have already used the big model or have started to use it.

But in the past month and a half, at least from what I have learned, 80 to 90 percent of the financial industry, retail consumer industry, high-end manufacturing industry, and even central state-owned enterprises have deployed DeepSeek. Some of them are already using DeepSeek to solve knowledge bases, solve business analysis, and other problems. They have begun to try to implement it. This is a huge change. People have started to use it and embrace it.

Compared with before the Spring Festival, I have communicated with these companies and found that their cognition has improved a lot. Because when you start to use something and embrace it, you can truly understand the value that this big model can bring to the company. The first aspect is to embrace it and try to implement it; the second aspect is that cognition has reached a very big level. These two are new changes in the industry.

"Altitude 7950": What do you think of Manus? The market's enthusiasm actually represents the public's expectations for it. They expect it to be another phenomenal technological breakthrough like DeepSeek, but in fact it is very controversial.

Li Kefeng: Manus is a very good topic. It has occupied a lot of hot topics in self-media and media in the past few weeks.

First, we say that Manus is a very excellent team. It proposed the concept of a completely autonomous agent, which was recognized by everyone on such a large scale for the first time. After Manus was proposed, no matter what people think of it, everyone has recognized what AI Agent is, what Agent is, what it can do, and what its future value is.

So the popularity of Manus in the industry is similar to that of DeepSeek, which is similar to iOS and Android. After iOS and Android came out, more apps like WeChat, Didi, and Taobao were launched. Manus made everyone realize for the first time that the future should be AI Agent. Agent is an app in the mobile era, so what kind of things AI Agent can do in the future is of great value, and it plays a role in popularizing the market.

Second, there are many shells in it, and shells are actually very difficult. Let's take WeChat for example. WeChat is not an Android or iOS app. It is based on Android and iOS to build a very good experience in its scenarios. Although WeChat is a Killer APP, in this field, it is also a shell because it does not do OS things. Shell is not a derogatory term, but how you use the power of a large model to solve complex customer problems and user problems, I think this is valuable. This road itself has been paved, and there are many technical engineering details that are still very advanced. This needs to be viewed positively.

Many people emphasize that it is not the first one, that it cannot be universal, and that many of the experiences are not good enough. This is the early stage of the industry's development. If we look at its ultimate experience at this stage, there is definitely a gap. We should not judge it too harshly. It has indeed created a hot topic and discussion in the market in this field and in AI Agent, and has popularized science. Its value is still quite large.

"Altitude 7950": The most fundamental thing is to see whether it meets the needs of users and whether it is valuable to users?

Li Kefeng: Yes, because the core point, whether it is DeepSeek or other large models, is still a basic capability. Although general large models cannot do many things, because you still lack internal privatized data, know-how in various industries, and professional analysis ideas.

A large model like this is like a very good graduate who is very good in various subjects but still lacks work experience, understanding of the internal situation of the enterprise, and understanding of the user's scenarios and actual practice. This is what AI agents such as Manus need to supplement. Only when something is built as a whole can it truly generate greater value.

"Altitude 7950": I heard a very interesting metaphor in your company to describe the relationship between applications and big models, saying "applications are bottles, big models are water, when the water is about to overflow from the bottle, it actually means that the bottle has become a bottleneck." In other words, the paradigm of applications may limit the performance of big models. How do you understand this?

Li Kefeng: How are products designed in the mobile Internet era? Product managers need to sort out the entire logic of the product. We need to use WeChat, the core of WeChat, how should users use it, what is its value proposition, where should it jump to after clicking on a function, and what kind of communication protocol should it use. These are what traditional product managers define.

When you are dealing with an intelligent agent, think about it. If you give it more frameworks and constraints, just like when we train students, you need to guide it and inspire its own capabilities and power, rather than telling it to do what is written in the textbook. Otherwise, it will not be able to apply what it has learned to other situations. The same applies to this metaphor.

If too many product frameworks are like bottle caps that restrict the flow of water, they will really block it. We should stimulate its IQ more, because it can learn by itself, it has the ability to draw inferences from one instance, it is smarter than your product manager, has a better learning ability, and has a better ability to generalize knowledge. How do we use it? This is what I think the design of the new generation of products should be, instead of restricting it.

"Altitude 7950": Do large models also require unlimited imagination?

Li Kefeng: Yes. Because it absorbs the world's knowledge, it has so much computing power, and the parameters are now hundreds of billions. DeepSeek-R1 has 671 billion, and the human brain has only more than 100 billion brain cells. It has exceeded the number of human neurons, so why don't we think it is smarter than us?

Startups will be more aggressive

"Altitude 7950": I have a question. Many companies have connected to DeepSeek and are actively embracing AI. Why are there so few companies that can use AI to boost their productivity and improve efficiency? What is the crux of this problem?

Li Kefeng: This is a very good question. The world's economic development follows the cycle of technology. Whether it is the PC Internet wave that brought a lot, or the mobile Internet that brought a new wave, such as O2O and mobile payment, which have changed a lot of businesses. We believe that the key point of evaluating a real technology, not a fake technology, is whether you have changed the industry and promoted the development of the industry.

This question returns to a key point. We will verify whether the big model is a real technology and a truly valuable technology in the future. We will look back to see whether our industry and our lives have been changed. This is the only indicator to measure it. This is happening, because we can see that the big model itself is still a very young product. It has only been two months since DeepSeek became popular. DeepSeek-R1, which has reasoning capabilities, was not available until after the Spring Festival.

Let's look further ahead. From the time ChatGPT came out, that is, from 2022 to now, it has only been more than two years. In terms of the entire technology cycle, this is still in the early stages. It took nearly seven or eight years for PC Internet to develop from the United States to the popularization of PCs in China, especially the Internet. We used to connect to ADSL in Internet cafes, and it was not until after 2000 that every family had their own. The development of the Internet in the United States has been seven or eight years. From the perspective of mobile Internet, everyone wanted a ticket at that time. What was the ticket for mobile Internet? It was DAU exceeding 100 million. In fact, QQ and WeChat only broke 100 million in 2012 and 2013. Fortunately, the product I was responsible for, Baidu Mobile, was the third one to get a ticket with DAU exceeding 100 million. It had been five or six years since that era.

It has only been more than two years since the first year of ChatGPT, so we still need to be patient. When we return to its essence, it will indeed have an essential driving force to change the industry and people's lives, because from our bottom, it can solve problems that could not be solved in the era of small AI models. Second, everyone is actively embracing and promoting its progress, rather than just waiting.

So these two underlying variables, the first technical variable, it is indeed a new generation of AI, it can be universal, very low-cost so that everyone can use this scenario, only then will more applications explode, this is its core. Second, in fact, not all industries are waiting, but everyone is building together, and making various attempts around big models and DeepSeek.When many hands make light work, he will push this forward. I am very confident about these two variables.

"Altitude 7950": How to apply AI in the industry is the real problem we are facing now.

Li Kefeng: Yes, this is also a problem that must be solved, and the battlefield has shifted from the development of large model bases to how I can apply it to various industries.

《Altitude 7950》: We know that making large models is very expensive, and the cards are very expensive. So only big companies can afford it. When DeepSeek is open sourced, has the situation changed?

Li Kefeng: It has changed. We can see that Tencent was the first to connect to DeepSeek, and the product experience it provides is really very good. Baidu has also connected to DeepSeek. Now on Github, more than 70% of the AI ​​projects in the entire Github project worldwide have connected to DeepSeek. Including the large enterprises I just mentioned, they have also deployed DeepSeek.

This situation has changed. For large companies, they are facing great pressure. If they make their own closed-source big models, they must be at least twice as good as DeepSeek, so that they can change people's perception. It is not impossible, but it means that when I have a good enough open-source big model, why should I use a closed-source one? This is a soul-searching question. I believe that the big model teams of all large companies are thinking about this issue.

"Altitude 7950": You have worked in many large companies, including Samsung, Ping An, Baidu, and JD.com. You were also the core technical leader in these large companies. After DeepSeek was open sourced, large companies and startups were standing at the same starting line. What are their advantages and disadvantages now?

Li Kefeng: Startups will be more aggressive. Large companies are still thinking, should I give up on myself, or to what extent should I accept DeepSeek? This is a burden. We can see that many large companies have made fine adjustments based on their own product systems and their original large models, which are deeply bound. Now they have to abandon the original ones, optimize their own large models, and incorporate other open source large models. This change is very time-consuming and involves many decision-making things.

For startups, everyone has no baggage. We provide customers with the best solutions based on what is the best right now. Startups often stand from the perspective of customers and users. We only see the pain points of customers and meet their needs. We don’t care what I have to do. Therefore, the agility and customer-centric thinking of representative startups are difficult to achieve in large companies.

Do you think we are all starting from the same starting line? I don’t think so. I think that in every era, small and beautiful startups are often the leaders. Large companies have too many roles to play, too many burdens, and too many teams to prove the value of their existence. On the contrary, from the perspective of the market and users, it is not so pure. We also know that the ultimate product strength and vitality of the product still create value for customers and users. This is my judgment.

"Altitude 7950": When someone in a company proposes a new idea, there will definitely be doubts about its feasibility. How do large companies and startups deal with this?

Li Kefeng: This is a very good question. The idea we are talking about is very valuable, but at the same time it is a distraction because there are some unrealistic ideas. Because the resources of a startup company are very limited, we can't just invest a lot of people to try out an idea. The judgment of the founder or person in charge is very critical, and his judgment has two points:

First, technical judgment and feasibility are very important.

When you think of something, I want to land on Mars right now, this vision is very good and very attractive. We also know that it is difficult to achieve with our current engine technology or the technology in the next five to ten years.

Fortunately, I come from a technical background and I am very passionate about technology. I always say that no matter what I do, I must first jump into it, find the essence of technology, and grasp its rules. Only then can I form a judgment on technology. To be honest, it is very scary when you don’t have this judgment on technology. This is the first point.

Second, what are you good at? We must know that people can only earn money within the scope of their own cognition. When you don’t have focus, when you yourself can’t tell where your core abilities are and what you are good at, your ideas are too divergent.

So I think two points:

First, technical judgment is very important, and this is the most difficult thing for many traditional entrepreneurs who are not from a technical background. They know that this thing is important, but they lack the judgment on how to use this technology. Therefore, they need companies like us to help them plan from their perspective and provide them with good products and services.

Second, what is the scope of everyone's IDEA ? We are a Data+AI company. All innovative ideas are about how to make data more accessible to users and customers. When you focus on a specific field, all ideas are welcome. If you have these feasibility judgments, the feasibility judgment is not now, but whether this technology can be solved in two years, I will support such things. It may be more critical to ensure that everyone actively contributes ideas and it is within a feasible and focused scope.

"Altitude 7950": Is it easier for startups or smaller companies to reduce the cost of trial and error and retain innovative vitality?

Li Kefeng: Because decision makers are close to the front line, when an organization is very large, the decision-making chain to the VP has five or six layers, and only the VP has the ability to make decisions, he is too far away from the actual front line. Who is making such decisions? This is a key point in my opinion .

Why DeepSeek is successful? First of all, it is small

"Altitude 7950": DeepSeek and Manus were able to launch phenomenal technological breakthroughs. They are startups, small companies. What did they do right?

Li Kefeng: First of all, it is small enough. We have always said that there is a culture in Silicon Valley called "one pizza culture". One pizza is shared by 10 or 12 people at most. A team can share one pizza. This kind of organization is actually the most dynamic. Now there are many startups with less than 10 people in Silicon Valley. They are great and mainstream. They also use a lot of big models, write code and do prototype design. Big models can be handled without so many people.

When there are too many people working on anything but the product has not been verified, we call it PMF (product market fit). Do I have a market for this product? Does it solve the problems, pain points and needs of customers or users? When it has not been verified, we must not have too many people. After the verification, because the product needs to be iterated, various experiences need to be improved, and capabilities need to be supplemented, he can expand the team.

We often see in China, especially in large companies, that you may not be verified by the market, but the team has more than 1,000 or 2,000 people. When making products or doing innovation, you must find that one point to penetrate deeply, truly solve customer problems, and create customer value. This cannot be done with too many people, because too many people will lead to divergent directions .

We need to find the most critical point:

First, the team is small enough. I believe that DeepSeek will be a very large team in the future, but at least it has only more than 100 people to reach its current size. The first reason I just mentioned is that it is small.

Second, it is flat. When you are a manager and you think you know everything, or you do everything by yourself, the team loses the ability to innovate independently.

We also know that in the new era, past knowledge is a burden. Young people should be allowed to use their imagination in such a culture. This flat culture and equal culture are very important. For us at Shushi, as I said just now, what I do is to make directional judgments and define directions, but how the real products are developed is a process of rapid iteration every day in the process of constant collisions between young people. We should create such conditions.

Third, it is about the goal we defined. We said why DeepSeek was successful because the goal it defined was: I don’t have that many cards, I don’t have that much money, can I make a big model that is stronger than OpenAI? You can see that it is very critical. I don’t have that many cards, I don’t have that much money, can I use engineering methods to reach or even surpass the best big model.

After setting this goal, the team worked on solving problems around it, thinking about how to achieve it at a low cost. They came up with many brilliant ideas, which were engineering ideas, not inventions like the theory of relativity. They were achieved through continuous experimentation in engineering and we found engineering methods to achieve this goal.

Looking back at some other big companies making big models, I am also very familiar with them. They said that we should follow the path that OpenAI spent so much money to verify, which is their lowest cost. Because verification is very expensive, OpenAI spent 10 times the money to verify how to do GPT3.5, and we should follow others, which is the most convenient. So when he didn't set a goal, I shouldn't take this path, I don't have so many cards, I don't have so many resources, he didn't define this goal, so he can only follow, and he will always be one generation behind others. This is often the goal of some previous big models, I just need to follow. If the goal is set wrong or incorrectly, the result will be different.

We cannot say that the capabilities or talent density of other companies, whether they are large companies or the six small dragons, are not as good as DeepSeek, but these three points are often crucial: first, they are small enough; second, they allow people below to innovate equally and encourage young people to bravely put forward ideas; third, they set the right goals. These three points are the key (factors) for their success.

"Altitude 7950": Therefore, innovation should first be done with a small team. Flatness is for faster decision-making, and the third point is focus.

Li Kefeng: Focus, goal setting, and the direction of your efforts. If the direction is wrong, no matter how hard you work, it will be wrong. How the founder or the top leader defines the team's goals is very critical.

"Altitude 7950": Does your company do the same?

Li Kefeng: We are definitely not as good as DeepSeek, but we are also working hard to move in this direction.

"Altitude 7950": You have experience in several large companies, including Baidu and JD.com. You are a senior executive of a large company who turned to starting your own business. Now you look less like a senior executive of a large company. How do you switch between a senior executive of a large company and an entrepreneur?

Li Kefeng: Actually, to put it bluntly, people always go from the peak of ignorance to the valley of despair, and then to the slope of enlightenment. During this period, I experienced the peak of ignorance. What is the peak of ignorance? It is from not knowing that I don’t know to knowing that I don’t know. This process is quite painful. In the era of mobile Internet and Internet, I was always at the forefront. I have also worked in search, apps, Internet finance, and Internet e-commerce. I have achieved a lot of achievements. Looking back now, I think this "trend" and "trend" are the biggest driving forces.

But in the past, I thought that my talent and contribution were the main thing, but I didn’t realize that I was taking advantage of the situation. In fact, I still have many such problems, including the depth of thinking, the ability to grasp the essence of things, the construction of culture, and the deep understanding of culture.

Although I have become the top technical person and manage a team of thousands of people, in fact, many things have not really solved these problems. When starting a business, you find that there are very few things you can rely on. You can only rely on your friends around you, and you have to be responsible for the results of everything you rely on, because there is no one to back you up. You are the last person, you have to make decisions, and you have to bear everything yourself.

At the same time, we are under great constraints. We are not in a large company where we can recruit unlimited people, have unlimited budgets, and have the company's brand endorsement. In fact, we need a lot of help to do many things, but now we have very few resources to use.

So at this point, we often have to think about what is the essence of a thing or an organization? Because when you have limited resources and time, you must grasp the essence .

Therefore, we will think deeply about the value of our product strength, its unique technical and business thresholds , and how to truly inspire everyone's entrepreneurial culture and mentality in the organization, rather than relying on past management similar to that of large companies to get promotions, salary increases, and status in the industry. Without these, how do you inspire everyone's entrepreneurial mentality? How do you find the real core competitiveness and barriers of the product? This is how we have come along.

When you start to know what you don't know, you will understand it very well and clearly. Including Tencent, because Tencent is our investor, so Tencent's consulting team has been very helpful to me. They will organize exchanges and dialogues between various entrepreneurs and business people, so that you can improve yourself in an organized environment.

Another thing is to constantly learn from the outstanding entrepreneurs around you. We must let go of ourselves and not think that we are executives. If you don’t let go, you will not be able to learn and grow.

The third is to truly return to the first principles. When I consider everything, I consider the most critical thing. Because as we said, without this 1, no matter how many 0s are behind it, it is useless. So I constantly strengthen this thinking and bring this thinking to our core team.

So the core team gradually formed this kind of frequency, and everyone's consumption on people became less and less. Everyone formed an entrepreneurial mentality and a thinking mode, which is what we call the entrepreneurial thinking mode of grasping the essence. You will find that the burden on the organization is getting smaller and smaller, but everyone's 1+1 is greater than 2. There are so many partners, and everyone has played their own strengths. Therefore, this process first requires you to admit that you really don’t know a lot of things, admit that you didn’t do many things well, and then open your heart. So gradually form this kind of thinking, and then synchronize these thoughts through our continuous communication and heart-to-heart talks with partners, such as All-hands Meeting, so that everyone can form such a unified thinking.

"Altitude 7950": Very good. In other words, when senior executives of large companies start their own businesses, it is easiest to use previous successful experiences, but when they become founders of startups, they need to deny their previous self in large companies and run the company in a new and innovative way?

Li Kefeng: The stage of any large company from 0 to 1 is completely different from the stage of tens of thousands or hundreds of thousands of employees. It doesn’t mean that there is a problem with the process of large companies, because the stages are different.

We can see that, for example, in the 0 to 1 stage, Tencent Pony wrote his own code and was a product manager; in the 0 to 1 stage, Liu Qiangdong, CEO of JD.com, was a customer service representative; the same is true for Alibaba. In the 0 to 1 stage, Jack Ma went everywhere, visited so many customers and investors, and was rejected by so many people. We joined later and actually enjoyed the platform behind their success. When it comes to entrepreneurship, we have to go back to the 0 to 1 stage. How do these successful and respectable entrepreneurs make decisions, how do they train their teams, and how do they build such a culture? We should learn from this stage, not the experience you have gained after you have developed to a large platform. It cannot be transplanted to entrepreneurship.

"Altitude 7950": So you have now experienced 1 to 100 or 1 to 10,000 in a big company.

Li Kefeng: Yes, but only entrepreneurs can experience the stage from 0 to 1.

"Altitude 7950": We used to talk about "bottleneck", but does China still have any technology that can be "bottleneck"?

Li Kefeng: The bottleneck is because you don’t have such technology, so we will stop climbing this technology tree. At least now, from my own perspective as a technical practitioner, there are no defects that prevent us from climbing this technology tree.

We can’t say that we are ahead in terms of algorithms, computing power, and other aspects, but at least we have what others have. I think we have reached this point, so we are very fortunate. In fact, in the past few years, we were most worried about Chinese chips, because Chinese chips were not good enough, so Chinese chips could hardly support such large-scale training. However, DeepSeek has very low requirements for chips, so domestic chips can now do super excellent large-scale model training and reasoning.

We think this has solved the bottleneck problem, but it does not mean that China's chips have caught up with the most advanced Nvidia. Now it is enough. So when we have enough, we need to see that we can play the advantages of the use scenarios. This enters a narrative logic of China. China's advantage is more applications. So this solves the underlying problem. I think it will develop faster in the future. So I don't think there are any bottleneck technologies that prevent China from climbing the technology tree.