Roundtable Discussion: A Boundary-Breaking Dialogue on AI Application Trends | WAVES New Wave 2025
This is a new era for China's venture capital and investment. Currently, the Chinese venture capital market is not only at the turning point of the cycle bottoming out but also in the in - depth period of structural transformation. In the new ecosystem dominated by policies and with high concentration of state - owned assets and capital, only by conforming to the trend and making flexible adjustments can we seize certain opportunities in the face of uncertainties.
From June 11th to 12th, at the Liangzhu Culture and Art Center in Hangzhou, the 36Kr WAVES New Wave 2025 Conference, themed "New Era", gathered top investors in the venture capital field, founders of emerging enterprises, as well as scientists, creators, and scholars deeply involved in technology, innovation, and business. They jointly discussed cutting - edge issues such as AI technological innovation, the wave of globalization, and value re - evaluation, dissected their views on business ideals and the future world, and together explored, searched for, and headed towards the "New Era" of China's venture capital and investment.
On the afternoon of June 11th, a cross - boundary dialogue on AI application trends was held in the investors' venue. The guests participating in the discussion were Duan Jianghua, the founder & CEO of Daimeng (Shenzhen) Robot, Li Qian, the founder of Zadig, Xu Chong, the founder of Conghua Investment Research, Zan Zhongyang, the head of AI marketing in Greater China at AMD, Pang Dawéi, the founder & CEO of ChatExcel, and the host Jin Haibo, the general manager of Huatai Innovation Investment.
The scene of the round - table discussion
The following is the transcript of the dialogue, sorted out by 36Kr:
Jin Haibo: Thank you very much for listening. First, let me introduce myself, and then please introduce yourselves one by one. Huatai Innovation Investment is the equity investment platform using Huatai Securities' own funds. We have been engaged in technology investment for five years. This year, we specially launched the Huatai Chuangxing CEO Global Acceleration Camp program, inviting a group of CEOs in the AI entrepreneurship field to participate. The aim is to deeply empower entrepreneurs with Huatai Securities' resources and help everyone achieve practical results in development. As a platform with the background of a securities firm that combines technology investment and empowerment, we always carry out our work with such a positioning. Next, please introduce yourselves one by one, starting with Mr. Duan.
Duan Jianghua: Hello, everyone. I'm Duan Jianghua, the founder and CEO of Daimeng Robot. Thank you, Mr. Jin, for the invitation. I'm very honored to participate in today's round - table discussion. Daimeng is based in Shenzhen. We mainly focus on human sensory information, such as touch. We collect human sensory and movement information during operations and use this information to train robots' fine and dexterous operation abilities. So our robot company focuses on how to enable robots to have generalized and universal dexterous operation abilities. Different from making robots run and jump, we focus more on how to make them do work.
So how can we achieve robots' dexterous operation and make them truly become our "helpers"? Daimeng has released several core products to achieve this goal.
First, we need to enable robots to obtain human - like sensory information (tactile information). We have a very good vision - based visual - tactile sensor, the world's first multi - dimensional, high - resolution, high - frequency visual - tactile sensor DM - Tac W, which has the characteristics of high resolution, high frequency, and high heat - dissipation efficiency. It can be integrated into multiple execution ends such as two - finger grippers.
In addition, robots also need a very dexterous operation terminal. So we have a dexterous hand with a tactile sensor, the multi - dimensional tactile perception five - finger dexterous hand DM - Hand1, which integrates our millimeter - level visual - tactile sensor on the fingertips.
We also have a data acquisition system containing tactile information, DM - EXton, which is the system that collects human sensory and movement data as mentioned at the beginning. The above are the three core products we are currently researching and selling, which realize the full - link connection of "perception - operation - learning" for robots. Based on the data collected around this product series, we hope to train a data acquisition system useful for certain industry scenarios and truly apply it to various industries. Thank you.
Li Qian: Hello, everyone. I'm Li Qian, the founder of Zadig. We are engaged in the field of an AI - driven, open - source cloud - native DevOps platform. Cloud - native and open - source have been hot topics in the technology circle in recent years, but for us, it's more important to be closer to the customers' business. For example, the milk tea you order, the new energy vehicles you drive, and even many daily life scenarios are actually silently supported by Zadig. Now that AI has come into everyone's view, our team has always been concerned about how to truly implement AI in the DevOps scenario.
I'm very glad to have this opportunity to talk with you about this topic today.
Xu Chong: Hello, everyone. I'm Xu Chong, the founder of Conghua Investment Research. Let me briefly introduce our company. We are probably the most niche in the AI field today because we focus on the emerging REITs field, which is the current real - estate infrastructure fund. Since REITs in China only started around 2021, this emerging financial product may not be very familiar to everyone. We also noticed that there are a large number of real - estate infrastructures in traditional industries. It's a very traditional industry, and our company focuses on investment research in this industry. We apply a lot of AI technologies, including some reg extraction, etc., to greatly improve the efficiency of traditional financial investment research with AI and conduct some quantification. This is our company's current main business.
Zan Zhongyang: Mr. Xu just said their company is the most niche. AMD, as a company in the processor ecosystem, is probably at the most upstream. AMD is a leader in high - performance and adaptive computing. We have a well - known Chinese CEO, Dr. Lisa Su. We Chinese people affectionately call her "Aunt Su", and she has a strong influence in the entire industry. We not only have a lot of experience in the graphics cards in the well - known game field but also have corresponding products in handheld game consoles, PCs, edge - computing workstations, and the CPUs and graphics cards in our data - center servers. AMD is one of the few companies in the world that can cover all computing - power scenarios from the cloud to the edge to the end with its computing engines. We hope to, from the perspective of our processors, cooperate with upstream and downstream ecological partners in the AI field to achieve mutual success. Thank you!
Pang Dawéi: Hello, everyone. I'm Pang Dawéi from ChatExcel. Our team is from Peking University. Our product focuses on spreadsheet processing and can solve Excel and data - analysis problems through chatting. Since most of you here often use the three major office software in your work, such as writing documents and making spreadsheets, the biggest challenge when making spreadsheets is that you can't remember the functions and formulas, which makes it very troublesome. Our product was launched in 2023. We are the first domestic product to process Excel using AI. Currently, our product has the highest usage volume in China. We are a completely native AI intelligent - agent product. This month, you will see that our product will be integrated into Huawei mobile phones and Lenovo computers. Our product is called ChatExcel, and you can directly access it at ChatExcel.com. So we process data in an intelligent - agent way, which is very different from plug - in products. We process data through a thinking - chain model, so we can handle data from Excel, databases, and external sources. Therefore, our positioning is to be a team focusing on data intelligent agents. Thank you.
Jin Haibo: Thank you for the sharing of the five guests. As you can see, today's guest lineup is very diverse and representative. From processor R & D, basic infrastructure software, to agent technology, and key components of embodied intelligence, each of you has unique industry insights from your practical exploration in different fields.
Just now, you elaborated on your current business layouts in your self - introductions. Next, we want to discuss in two steps. First, we want to know your judgments on the future direction in this wave of technological trends and your specific next - step plans. Then, we also want to talk with you about the actual difficulties and challenges you face in business promotion.
Let's classify it. First, please let Mr. Zan share from the most basic level of technology, and then the guests in the software - technology direction can continue.
Zan Zhongyang: First, let me talk about what I've seen from the perspective of a chip manufacturer. What kind of hardware do the forms and user groups of AI applications need? For example, for the AIPC we are promoting, the situation in 2024 was different from that in 2025, and it has developed very rapidly.
Let me give a simple example. In 2024, we held the first AIPC Innovation Summit in Greater China. Many application partners came, and most of their presentations were in the form of WEBUI demos at that time. Just one year later, in 2025, when they stood on the stage, they all presented mature application forms with relatively mature business models, which shows a very rapid development. Also, I want to share with you that ChatExcel is also our good partner. We are in a state of mutual progress. Our products are constantly being optimized and evolved. Based on our processors, for example, on the terminal side, we can now run 70 - billion - parameter and 235 - billion - parameter models, compared with running 7 - billion - parameter, 8 - billion - parameter, and 14 - billion - parameter models before. We are evolving. Application manufacturers are also moving towards us. They are thinking about what kind of computing state and AI algorithms can be more efficiently placed on the terminal side to deliver the most complete and perfect applications to consumers in combination with their applications. I see this trend of mutual progress.
Li Qian: Technologists have always been the most daring and the first to try, but they are often the first to "fail". When the wave of large - language models just emerged, many people were engaged in application development, and it seemed very lively at first. However, soon, it was found that few could succeed. What we do is the "software behind the software", that is, the platform behind engineers, which makes us very sensitive to technological trends. I have a very deep impression of the release of GPT - 4o. It was a turning point for our team, and we achieved our first usable product, Pilot. However, it was the launch of the DeepSeek large - language model that pushed AI to an unprecedented height. The development and changes in China in the past year or so have been very significant. What we mainly see is the enthusiasm of technologists and the anxieties of clients and investors.
However, the reality now is that it is still not easy to implement AI applications. Especially in the enterprise - level software field, there are still not many truly "amazing" applications. The C - end, marketing, manufacturing, and consumer fields are a little better. For example, some of our clients, such as Bawang Cha Ji and Geely Auto, are also actively exploring how to use AI.
Now, there is a consensus that AI is indeed a good tool, not just a "sexy but useless" toy as before. However, we need to jointly explore with upstream and downstream partners where to apply it, find suitable scenarios, and verify its value.
Jin Haibo: Please let the two companies focusing on agents share their views.
Pang Dawéi: Regarding the development trend, since our product was launched in March 2023, we are probably one of the first domestic companies in the application field. Since GPT emerged in 2022, our user volume has been very high after the launch. In the past two years, we have experienced the transformation from a free C - end product to commercialization at the end of last year. As a small team of less than 10 people in the Chinese C - end AI application field, the process from free to paid has made us think about the product positioning and understand user needs.
Moreover, from last year to this year, we have also experienced the transformation from C - end to B - end applications. This year, there has been a large number of B - end clients. At first, users just tried it for fun, but now they really need it to solve problems. In my opinion, the real trend is established when users are willing to pay, especially when B - end clients really make purchases. At this time, we can tell that users actually need tools that can truly solve problems, rather than just for entertainment.
Since our product focuses on Excel data, users require 100% accuracy in problem - solving. If the processing is inaccurate, users will not use it. Secondly, users need a wide range of processing types, from file - type data to data - type data. Thirdly, security is also important. For example, when dealing with hardware manufacturers like AMD, users want to solve data - security problems. From our perspective, the trend is very clear. Users' needs are becoming more specific, and they require accurate and secure solutions. This is the trend we see and the direction our data - intelligent - agent track is taking. Thank you.
Xu Chong: Hello, everyone. Our company is the only one that is finally involved in very practical application scenarios. I'll briefly talk about some problems our industry has encountered. Since our industry, such as real - estate infrastructure, used to have relatively little data and relied more on resources, there was less tangible data. Also, since the REITs industry in China only started in 2021, there is no solid data foundation. In this process, the problem we face is how to make this traditional and physical industry more data - driven and AI - enabled for valuable investment research and analysis.
In this process, as Mr. Jin mentioned before, traditional investment research was very labor - intensive. However, now, with the continuous emergence of new AI applications, we are leveraging these AI tools. Our end - user tools are like floating on the wave of AI, constantly updating with new AI intermediate - layer tools. For our end - user software, I think this is both a trend and a challenge, that is, how to make good use of these rapidly evolving new AI tools. In addition to traditional general - purpose tools, for end - user clients in niche fields, there is a need for companies like ours that can understand clients' needs and make good use of AI tools. I think this is both a difficulty and an opportunity. Thank you.
Duan Jianghua: The robot industry has been very popular in the past one or two years, which has aroused public enthusiasm. People really hope to see humanoid robots enter the physical world to serve tea, do housework, cook, etc.
However, after more than a year of development, people are starting to ask what robots can actually do. When we visited clients, they asked if robots could run stably in their factories for 8 hours. This is a very harsh problem we need to face.
As the industry develops to this stage, we can see a very promising future, but there will definitely be many problems in the implementation process. We can see that most of the entrepreneurs in this wave of embodied - intelligence field are very young. The key to the implementation of robots and embodied intelligence lies in having a group of energetic, knowledgeable people who are willing to work with industrial and investment parties to promote the industrial implementation. Only in this way can we really achieve our goals.
As an industry practitioner, I've noticed that despite different voices from the outside world in the past two years, both the market and technology are evolving very rapidly, which makes us very happy. I often tell my colleagues that we should take a long - term view. Although this industry may face some doubts, it doesn't matter. We have seen a feasible path, even if it is long or tortuous, it will eventually come. The past two years of being in the spotlight and then facing doubts are all experiences, and we need to be more patient. Thank you.
Jin Haibo: We just discussed the topic of commercialization and are glad to find that there are indeed many practical implementation opportunities and scenarios in the AI field. Everyone also mentioned the key issue of "commercial implementation". I wonder if you have noticed that the pace of entrepreneurship in this wave of AI is very different from that in the previous mobile - Internet era. At the beginning of the wave, everyone was full of enthusiasm and wanted to make a big start. However, after two or three years of actual involvement, we can see that the market now has very high expectations for commercialization and monetization. In the past, mobile - Internet startups could tolerate years of losses, but now, if a startup still cannot monetize after two or three years, investors will be hard to satisfy. Both entrepreneurs and shareholders are more eager for rapid implementation.
Based on this, I'd like to invite you to share your commercialization experiences from the cold start to getting the first order, including the pitfalls you've encountered and how to relieve commercialization anxiety. Please let Mr. Duan share first.