Roundtable Discussion: Investment Guide for Embodied Intelligence: Bridging the Gap from Technology to Market | 36Kr Future Industries Conference 2025
On September 10th, 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 "In the Era of In - depth Cultivation, the Tides Surge in the Land of Abundance", it has made every effort to create a high - standard, high - value, and high - impact industry event that combines national vision, 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 large - scale consumption. It brings together top industry players to discuss development paths and outline the future of the industry. In the two - day agenda, the conference takes the "industrial cooperation chain" as the main logical line, focusing on the collaborative mechanism among the government, capital, and industry. It delves deep into how to break down barriers, integrate resources, and precisely address the pain points, bottlenecks, and constraints in industrial development.
36Kr organized a round - table discussion titled "Investment Guide for Embodied Intelligence: The Leap from Technology to Market" at the Future of Industry Conference. Huang Zhuxi, a senior analyst at 36Kr [Host], Ji Xing, a partner at Light Source Capital and the managing partner of Light Source Innovation Fund, and Wu Changzheng, the president of Magic Atom, these industry experts jointly explored the investment directions and paths for embodied intelligence.
The following is the content of the round - table discussion, edited by 36Kr:
Huang Zhuxi: Hello everyone, I'm Huang Zhuxi, a senior analyst at 36Kr. I'm very honored to invite Mr. Ji Xing from Light Source Capital and Mr. Wu from Magic Atom to share their insights on the embodied intelligence sector with us. First, let's talk about a hot topic. Embodied intelligence is undoubtedly the hottest sector this year, but the popularity also comes with controversies. A somewhat contradictory phenomenon is that while the industry is not yet mature and the technology is still evolving, the valuations of enterprises are soaring in both the primary and secondary markets. There seem to be some unreasonable aspects worthy of discussion here. So, I'd like to ask Mr. Ji Xing first. How do you view these current phenomena in the embodied intelligence industry?
Ji Xing: Thank you, host, and thank you to 36Kr and the organizers. I'm very glad to have this opportunity to share our views on embodied intelligence, which is indeed one of the hottest sectors in the current market. Regarding the "valuation bubble" issue mentioned by the host just now, I'd like to start with a set of data.
First, the "popularity" we're talking about is relative. As of the first half of this year, the total financing in the field of AI and embodied intelligence in China was approximately over 6 billion US dollars. In contrast, the financing in the United States during the same period exceeded 100 billion US dollars. Even after excluding the huge financing of OpenAI, the financing in the United States was still 60 billion US dollars, about ten times that of China. But the question is, who can assert that the future scale of the AI market and application prospects in China will be ten times smaller than those in the United States? I believe no one dares to make such a claim. Therefore, I think the so - called "bubble" or the amount of financing needs to be viewed in a relative context. In recent years, the popularity of the embodied intelligence field in China has indeed been rising continuously, but in terms of absolute amount, there is still an order - of - magnitude gap compared with the United States.
Second, let's talk about the essence of the bubble. Usually, the formation and bursting of a bubble correspond to the downward adjustment and even the collapse of the market's future expectations. However, in the field of embodied intelligence, the situation we're seeing is exactly the opposite. With the maturity of technology, the decline in costs, the continuous expansion of application scenarios, and the enhancement of generalization ability, the market's future expectations for this field are continuously rising. I believe that embodied intelligence, as the key node connecting AI and the physical world, will be a trillion - level market in China. Behind this is China's most complete and complex supply - chain system in the world and the most diversified application scenarios as support.
Third, historical experience also provides a useful reference. Looking back at industries such as new - energy vehicles and photovoltaic in China, which are now leading the world, they all went through a stage where they were regarded as a "bubble" by the outside world in their early development. However, facts have proved that the early capital enthusiasm not only greatly promoted the rapid development of these enterprises but also, more importantly, the influx of capital accelerated the maturity of the entire upstream and downstream supply chain, effectively reducing the comprehensive cost and ultimately contributing to the rapid rise of the entire industry. It can be seen that this "early investment" is a necessary path for any strategic emerging industry to reach maturity. In general, as long as the technology remains leading and the market space is broad enough, the current popularity in the industry is by no means the prelude to the bursting of a bubble but the real beginning of a great industry.
Huang Zhuxi: Light Source Capital has invested in many representative projects in the embodied intelligence sector, such as popular enterprises like Lexiang and Guangxiang. I'm curious to know what, in the view of Light Source, are the key factors for these projects to successfully raise funds in the early stage? Or rather, what kind of projects will be favored when Light Source evaluates projects?
Ji Xing: First of all, in a sector as early - stage as embodied intelligence, the team is always the top priority.
Although the so - called "Product - Market Fit" (PMF) we often mention is important, there is actually an earlier and more easily overlooked prerequisite, which is "People Mission Fit", that is, the entrepreneurial direction is highly compatible with the founding team itself. We've observed that the top - notch teams often have a unique quality: they not only have a world - class "academic" technical background but, more importantly, they also have a deep understanding of product logic and industrialization paths and have accumulated rich implementation experience.
Projects like Magic Atom and Galaxy Universal, which we've invested in and served, have fully demonstrated this excellent ability structure and resource diversity in their founding teams. This interdisciplinary "dream team" can quickly build real market - competitive barriers, rather than just staying at the technical R & D level.
In addition to the characteristics of the team itself, the team's ability to deeply integrate with scenarios and industries is also worthy of attention. This year, we've indeed seen many robots "doing the yangge dance" or "running marathons". These technical demonstrations have brought extremely high popularity to the embodied intelligence sector. However, as an investment institution, what we care most about, and what we believe every entrepreneur should focus on, is the ultimate question: Who will pay for your (technology or product)?
This requires the project to have a clear ability to define scenarios. Specifically, the team needs to precisely find a landing scenario that is "technically achievable and commercially viable" and be able to clearly convey the core value of this scenario to investors and the market. In our view, the actual verification of scenario implementation always carries more weight and is more convincing than the "concept packaging" in the capital market. Take Galaxy Universal as an example. Since its establishment, the company has firmly followed the industrialization path. It has deeply cooperated with industry giants such as Mercedes - Benz, ZEEKR, and CATL, continuously verifying the value of its technology and products in industrial scenarios such as the automotive and energy sectors.
Finally, I'd also like to talk about the role of capital in this sector and our perspective on valuation. We always believe that the competition in the field of embodied intelligence is essentially a triple coupling of "technology × scenario × capital". Capital should not be a tool for chasing inflated valuations but an important driving force for accelerating technological breakthroughs and promoting scenario penetration. The financing rhythm of a good project should be closely tied to the goals of industrialization breakthroughs, and every penny of funds should be used at the right place. Take Galaxy Universal again as an example. In the past two years, it has raised over 2 billion yuan, but the use of funds in each round has been highly focused on key technological and industrialization milestones.
From the perspective of valuation, the current valuations of embodied intelligence enterprises seem high, but this is actually the market's recognition of the value of such scarce teams and high - quality assets that deeply integrate technology and industry. When looking at such valuations, we need to use an "end - game thinking" rather than just focusing on short - term numerical fluctuations.
Of course, the development of any emerging industry will have its ups and downs, but we believe that the development of embodied intelligence is definitely the mainstream trend in the future.
Huang Zhuxi: We'll delve deeper into the question of "who will pay" later. Continuing on this topic, I'd like to ask Mr. Wu from Magic Atom. Since its establishment in 2024, in over a year of development, what key nodes and challenges has the enterprise encountered, and how has it dealt with them?
Wu Changzheng: Thank you, host. I'm Wu Changzheng from Magic Atom. Since its establishment in January 2024, Magic Atom has gone through a year and a half of development. During this period, the company has indeed experienced several key points and major decisions. I'll briefly share a few:
First, in terms of technical routes and product development. Right after the company was founded, we finalized our core products and technologies, covering two product lines: quadruped robots and humanoid robots. Especially for humanoid robots, which evolved from quadruped robots, the technology was finalized around May last year, and mass production started to be promoted in October.
Second, in August, we began to promote humanoid robots to enter real - world scenarios for exploration and application. Although the robots' ability to adapt to scenarios was still in its infancy at that time, we firmly chose the strategic direction of solving specific problems in vertical scenarios.
Third, in March this year, we started to officially release products. So far, we've released 4 products, including full - size general humanoid robots, consumer - grade quadruped robots, four - wheeled quadruped robots, and a small - size bipedal humanoid robot with high dynamics, initially forming a relatively complete and rich product matrix.
Fourth, in April this year, we launched the "Thousand - Scenario Co - creation Plan". In the process of industrialization, we found that in addition to products and technology, we also need corresponding partners to open up relevant scenarios for us and jointly overcome the practical application problems in these scenarios. Currently, we've contacted over a thousand enterprises, screened out over a hundred deployable scenarios, and are promoting in - depth POC projects with more than 50 enterprises, enabling robots to perform tasks such as handling, loading and unloading, quality inspection, and tray - loading on production lines.
In general, we've made significant progress in product richness, maturity, and mass - production capability. Currently, we're in the stage of mass production and delivery. In the future, we'll continue to focus on scenario - based implementation, promoting humanoid robots and embodied intelligence technology to become truly usable new productive forces.
Huang Zhuxi: Entrepreneurship is indeed a step - by - step process of "overcoming one difficulty after another". Just now we talked about scenario adaptation and commercial implementation. The second question is for Mr. Ji Xing. From the perspective of an investment institution, from which aspects are current enterprises in the embodied intelligence sector achieving commercial implementation? What have you observed in this process, and what suggestions can you give to startups?
Ji Xing: This is a great question.
Currently, there is a general consensus in the industry that this year can be regarded as the first year of commercialization of embodied intelligence. This judgment is mainly based on two positive signals: on the one hand, the cost of upstream core components is continuously decreasing, providing a feasible cost basis for commercial implementation. After all, if the BOM cost of a solution remains high, commercialization is out of the question; on the other hand, the technology itself has crossed the key maturity threshold, making it possible for products to enter real - world scenarios and solve practical problems. Against this backdrop, we've noticed that many companies are now moving from the technical demonstration stage to in - depth exploration of more specific implementation paths in real industrial and commercial scenarios. We'd like to share two suggestions for these startups:
First, product development should be truly driven by scenario needs.
The starting point of R & D should not be "what technology can be achieved in the laboratory" but must return to the commercial essence of "what practical problems customers need to solve in specific scenarios". "Who will pay" is the fundamental question that determines whether an entrepreneurial path can succeed. Therefore, the team needs to first build the ability to customize and refine products in specific scenarios, and on this basis, further explore the possibility of generalizing and migrating this ability to more scenarios.
Second, great importance must be attached to the closed - loop of the business model, especially the actual performance of ROI (Return on Investment).
No matter how attractive the application prospects of a scenario seem, if the economic account doesn't add up, commercialization will be difficult to sustain. The solution provided by our ecosystem enterprise Galaxy Universal for Meituan's smart pharmacies is a very typical positive example: traditional 24 - hour pharmacies require three - shift manpower support, resulting in high labor costs; while a robot priced at over 200,000 yuan (and the subsequent cost will continue to decrease) can effectively replace this part of the manpower, bringing clear and quantifiable cost savings and returns to customers.
This quantifiable value is precisely the cornerstone for the commercialization of embodied intelligence to be established.
Huang Zhuxi: From your perspective, Mr. Wu, in which fields is the commercialization of the entire industry mainly concentrated? What commercialization attempts has Magic Atom made so far?
Wu Changzheng: As Mr. Ji Xing just mentioned, this year, people have indeed been very concerned about the implementation and application of embodied intelligence robots, especially humanoid robots. From the perspective of Magic Atom, current commercialization still focuses on industry and commercial services. This is consistent with Mr. Ji Xing's view. We're taking a three - step approach: starting with industry, then moving to commercial services, and finally entering the home.
We expect that the comprehensive ability of robots will increase tenfold in three years and a hundredfold in five years. Only after their capabilities mature will they gradually enter the home scenario. Currently, we're more focused on some vertical industrial scenarios. For example, we're now concentrating on industrial scenarios, performing tasks such as handling and loading and unloading on the production lines of the home - appliance manufacturing industry.
Why did we choose the production lines of the home - appliance manufacturing industry first? There are mainly two reasons: First, the automation rate in this industry is not very high. Since the home - appliance manufacturing industry needs to iterate and upgrade products every 2 - 3 months, the switching cost of highly automated production lines is very high, so it still relies heavily on manual labor; second, humanoid robots can solve the generalization problems that traditional automated equipment or industrial robots cannot solve. For example, the same workpiece may be in different states in five processes. Relying on the data - and - model - driven technical paradigm, our robots can have end - to - end vision - language - action (VLA) capabilities and gradually adapt to the changes in such scenarios and tasks.
In addition, we currently adhere to using 80% of the data from real - world robot scenarios. Because there is a relatively large difference between the data collected in the laboratory and that in real industrial scenarios. For example, the lights used in industrial sites are different from those in the laboratory, which may cause stroboscopic effects in the pictures seen by the robots. Only real - world deployment can enable robots to deeply understand and adapt to such practical challenges.
In terms of commercial services, since 2016, there have been commercial - service robots, but they haven't been widely deployed. The main reason is that the tasks they can solve are relatively limited, only able to complete 10% - 20% of the work. However, the current humanoid robots can increase this to 40% - 50%. In addition to greeting, guiding, and explaining, robots can also perform multi - modal interactions and more naturally understand user intentions, thanks to the intelligence improvement brought by large - scale models. At the same time, the current humanoid robots are different from the previous commercial - service robots. They have a two - arm structure and can perform operational tasks. For example, when there is a shortage of goods in a store, they can conduct inventory checks or replenish goods, significantly expanding their application potential in commercial scenarios.
Huang Zhuxi: From industry to commercial services, and then to the personal consumption scenario. How long will this process take? How does the industry view the possibility of embodied intelligence truly entering the consumer - grade market and realizing scenarios such as companionship?
Wu Changzheng: This question has been repeatedly asked in