Roundtable Dialogue: Embodied Intelligence, the "Marathon" Has Just Begun | 36Kr Future Industry Conference 2025
On September 10, the 2025 36Kr Future of Industry Conference, hosted by 36Kr, grandly kicked off in Xiamen, China. This conference has joined hands with the "China International Fair for Investment and Trade" hosted by the Ministry of Commerce. With the core theme of "The Era of Intensive Cultivation, the Surging Trend of Jiahe", it has made every effort to create a high - specification, high - value, and high - influence industrial event that combines national height, industrial depth, and market popularity.
The conference closely aligns with national strategic guidelines and the forefront of industrial development. It focuses on five core sectors: artificial intelligence, low - altitude economy, advanced manufacturing, new energy, and mass consumption. It brings together top forces in the industry to discuss development paths and outline the future of the industry. In the two - day agenda, with the "industrial collaboration chain" as the main logical line, the conference focuses on the tri - party collaborative mechanism of "government, capital, and industry", and deeply explores how to break barriers, integrate resources, and precisely solve the pain points, blockages, and bottlenecks in industrial development.
36Kr organized a round - table discussion titled "Embodied Intelligence: The 'Marathon' Has Just Started" at the Future of Industry Conference. Lin Haozhuo, the founding partner of Joy Yuan Asia, Ding Hao, a partner of Boyuan Capital, and Lin Wenxin, a partner of Yiwei Capital, these industry experts jointly discussed one of the hottest topics in the investment and industrial circles - embodied intelligence.
The following is the content of the round - table discussion, edited by 36Kr:
Host:Good afternoon, dear guests! Today, we are very honored to invite Mr. Lin Haozhuo, the founder of Joy Yuan Asia, Mr. Ding Hao, a partner of Boyuan Capital, and Mr. Lin Wenxin, a partner of Yiwei Capital.
We are very glad to have the opportunity to discuss the topic of embodied intelligence with friends in the investment and industrial circles today. It is also one of the hottest topics in both the investment and industrial circles at present.
Now, please let each guest briefly introduce themselves. Let's start with Mr. Lin Haozhuo.
Lin Haozhuo:Joy Yuan Asia is an investment institution focusing on the industrial chains of artificial intelligence, robotics, and semiconductors. It mainly makes layouts around three major directions:First, strengthening and supplementing the upstream of the artificial intelligence industrial chain, including data centers, infrastructure, core components of robots, and downstream scenario - based solutions;Second, AI foundation models;Third, core semiconductor equipment and key basic elements such as computing, transmission, and switching.
Ding Hao:Hello, everyone. I'm Ding Hao from Boyuan Capital. Boyuan Capital is a market - oriented hard - tech investment platform established by Bosch Group in China. Bosch is one of the world's largest automotive parts suppliers. Last year, it achieved a global revenue of over 9 billion euros, of which more than 60% came from the automotive and intelligent mobility - related fields. It is worth mentioning that China has been Bosch's largest single revenue and profit - contributing country for four consecutive years.
Therefore, in the Chinese market, we are actively investing in the hard - tech field and coordinating with Bosch's local and overseas products and strategies. Against this background, artificial intelligence and robotics have become our key investment directions in the past two years. Bosch has a large number of artificial intelligence and embodied intelligence application scenarios in the supply chain and manufacturing processes and has accumulated rich industry know - how. As the "chain leader" of the industry, this is the core reason why we attach great importance to embodied intelligence and actively participate in its layout.
Lin Wenxin:Hello, everyone. I'm Lin Wenxin from Yiwei Capital. We are an FA institution established at the end of 2017. In the eight years since our establishment, we have successfully helped more than 400 startup clients raise over 70 billion RMB in total, mainly focusing on technology - related industries. Among them, embodied intelligence is an industry we attach great importance to. We initially served relatively traditional robot companies, including service robots, industrial robots, AGVs, etc. In the past three years, we have spent a lot of time in the field of embodied intelligence. Our current clients include more than a dozen active embodied intelligence enterprises in the market, such as Independent Variable Robotics, Sweet Potato Robotics, Songyan Power, and Daimeng Robotics, covering all aspects of the upstream and downstream of the embodied intelligence industrial chain. We have always believed that embodied intelligence is indeed a huge opportunity and one of the few industries that may give rise to very large companies in a relatively short period in recent years. So we have invested a lot of time and energy. I have worked in both the United States and Europe, engaging in both investment and entrepreneurship. I have been doing FA for the longest time. I really think that Chinese entrepreneurs, especially those with a technical background, need in - depth services to connect with the capital market. That's about the situation.
Host:Thank you, Mr. Lin. Through the self - introductions of the guests, we can see that today we have investment institutions from Tsinghua University, CVC and FA institutions, as well as local state - owned assets like us. The first question I want to ask today is: The industrial chain of embodied intelligence is somewhat similar to that of new - energy vehicles, and even has greater imagination space. From upstream chips and sensors to the middle - stream robot bodies and large models, and then to downstream application scenarios, this value chain is very long, and there are still variables and uncertainties. So at the beginning, I want to ask all the guests a common question. If we were to invest 100% of our capital in the embodied intelligence track, how would you or your company allocate the investment ratio among the "brain", "body and limbs", and "scenarios"? What are the investment logic and judgment criteria for this ratio? Let's first invite Mr. Lin Haozhuo to share his thoughts.
Lin Haozhuo:The ratio is roughly 4:4:2 or 4:5:1. The scenario side accounts for 40%, the body and components (including sensors, electronic skin and other perception devices) account for 40% - 50%, and the brain accounts for 10% - 20%. The core logic is as follows: The body forms vary greatly (the requirements in ports, mines, and household scenarios are different), and components are the foundation for industrial implementation; scenarios are the key to value realization and need to be prioritized in the layout; the brain is still in the early stage, and its generalization ability needs to be improved, so we don't invest heavily in it for now.
Host: Thank you, Mr. Lin. You will focus your investment on the scenario side, followed by the body and components, and allocate a smaller proportion to the model and the brain. Now, let's invite Mr. Ding Hao to share his views.
Ding Hao:In our view, the development stage of embodied intelligence is very similar to that of autonomous driving (L4) seven or eight years ago: The financing rhythm is fast, but the commercialization process is still relatively slow.
Corresponding to the three directions of "brain - body/limbs - scenarios", it can be compared to the development path of intelligent driving: The brain is like the intelligent driving algorithm of L4 back then, the hardware products such as controllers and sensors correspond to the limbs, and the scenario classification is similar to that from L2 to L4. The development cycle of autonomous driving in the past may provide a reference for embodied intelligence.
First of all, the brain is the most indispensable part in the early stage of the industry. Without the brain, the entire track cannot move forward. Although the ability of the brain at this stage is only at a level of 40 points or even lower, the same is true for the development of embodied intelligence. The "brain" is very worthy of attention and investment in the early and middle stages of the industry's development and will run through the entire development process of the industry. However, the maturity of models and algorithms requires a long - term process. From a level of 40 points to 60 points and then to 90 points, each step depends on long - term data and scenario accumulation and iteration. Just like today's autonomous driving, based on the end - to - end model and a large amount of data precipitation, the current L2+ product form has gradually been realized. Future robots will also go through a similar long process, starting with small models and small data sets in specific scenarios and then developing into more general large - model products such as VLA models. Therefore, we will continue to pay long - term attention to the "brain" end.
Secondly, from the perspective of industrial capital, we believe that the scenario end is crucial. Without real scenarios, it is impossible to accumulate data to iterate products, verify business models, and achieve scale. Embodied intelligence has great application value and market potential in the general industrial field. In the past year, we have carried out POC and application pilot explorations with several leading robot companies in the factory end, including industrial logistics, complex assembly, and complex process operations, and identified the first batch of implementable industrial scenarios. Although it will take some time to promote commercialization, as long as replicable and economically viable scenarios are found, the industrialization and scale - up of embodied intelligence can be expected in the future.
Finally, the "limbs" correspond to the components and supply chain links, and Chinese players have significant advantages in this aspect. When the brain reaches a basic level and scenarios are initially available, everyone will flock to various links of the supply chain, including joint motors, dexterous hands, reducers, lead screws, and sensors. In fact, we have seen many listed companies, automotive parts manufacturers, and numerous startups enter the various component links of robots, including both general - purpose components and components unique to robots. The development of core components and the local supply chain is the key for China to become the largest market for embodied intelligence and is also our key investment direction.
Overall, our investment ratio in the three parts is roughly: 40% for scenarios, 40% for the supply chain, and 20% for the brain.
Host: Now, let's invite Mr. Lin Wenxin to share his views.
Lin Wenxin:Our choice may be different from those of the previous two guests. I think the core driving force of this wave of embodied intelligence is AI. So actually, the brain is the biggest variable. However, I don't think the brain and the body can be completely separated. Because if we really want to implement intelligent robots in scenarios, it must be combined with the body. For many brain manufacturers, the body may be self - developed. For me, I will consider these two parts together, accounting for about 50%.
Regarding the core components in the supply chain, if they overlap with the current automotive industrial chain or other advanced manufacturing supply chains, the investment value is not that great. Because whether they are listed companies or enterprises that have done well in these fields before, as long as they invest enough capital and attention, they can make it. However, if they are some relatively new sensors, it's different. In this link, I will mainly invest in relatively new sensors and chips.
Finally, scenarios are categories that can bring good investment returns in the relatively short term. However, due to my personal preference for technology, I won't allocate a large proportion to scenarios. So for us, the ratio is probably 5:3:2.
Host:Thank you, Mr. Lin. The previous question was quite interesting. From the answers of the guests, we can see that investment institutions have different views on which links to invest in the embodied intelligence track and the proportion of investment in each link. There is no consensus yet. This also represents the current development status of the embodied intelligence industry. Just as today's theme is "The Marathon of Embodied Intelligence" has just started, at the beginning of the marathon, the bets and the segmented tracks of the players have not fully converged. We can wait and see. As the entire track develops dynamically in the future, this ratio may be adjusted.
Joy Yuan Asia is an institution with a global vision and is very strong in the hard - tech field. It has invested in many star projects in recent years. The first question I want to ask is, in the current embodied intelligence value chain, which link do you think will present the most globally competitive investment opportunities?
Lin Haozhuo:There are mainly four directions.First, technology - based companies that solve the problem of "mismatch between pre - trained data and robot bodies".Currently, the robustness of models is insufficient, and upgrading the body (such as replacing sensors) will greatly affect the inference effect. If this problem can be solved through AI optimization and hierarchical model design, the potential is huge.Second, body companies focusing on industrial scenarios.China is the only country that covers all manufacturing categories. In scenarios such as shipbuilding and ports, where the work is dirty and tiring, there is a large space for robot substitution, and it is easy to achieve global leadership. The Tsinghua team we invested in focuses on the ship spraying and cleaning scenario and quickly entered the market relying on China's shipbuilding industry advantages.Third, weight - reducing materials for robot bodies.Traditional metal materials have limitations in terms of weight and cost. New materials and metamaterials are important breakthrough points.Fourth, edge - side AI chips.In the future, as the number of robots increases sharply, it is impossible to rely entirely on cloud computing power. Ultra - low - power edge chips are an inevitable trend. The Qingwei Intelligence we invested in has accumulated rich experience in this field.
Host:Mr. Lin has shared the investment opportunities in the embodied intelligence "marathon" from a global perspective. From your sharing, you mentioned four opportunities, including pre - training, industrial scenario applications, weight - reducing solutions, and even edge - side AI, which all seem to have good investment prospects.
I would like to ask Mr. Ding from Boyuan Capital. Not long ago, Boyuan Capital announced a strategic cooperation with Galaxy General, which can be regarded as a key strategic bet on the development of humanoid robots by Boyuan Capital. Currently, the market may think that the hardware of humanoid robots is not yet perfect. There is room for improvement in terms of performance and cost. In addition, in specific scenarios, such as industrial scenarios, there may already be special - purpose robots. So in the special - purpose field, embodied intelligence also has to face the cost - performance challenge of existing special - purpose robots. So I'd like to ask Mr. Ding, do you think the first "atomic bomb" to detonate the robot industry is a breakthrough in the cost of the supply chain or the most explosive point in downstream application scenarios?
Ding Hao:As mentioned before, in the past year, we have tried to use embodied robots in various links of the production lines in several Bosch factories, including industrial assembly and industrial logistics. In the factory demand research, we collected hundreds of demands. By disassembling each scenario and clarifying all processes, links, rhythms, and action requirements, we found that more than 90% of the current scenarios cannot be effectively covered by existing embodied intelligence products. There is still a huge gap between demand and products.
Many dark factories have achieved a high degree of automation. However, at the same time, there are still a large number of manual operation links in the production