The Ultimate Game Between Carbon-Based and Silicon-Based: Hurdling and Sprinting of the Robotics Industry

While AI is advancing by leaps and bounds in the digital world, capable of writing poetry, generating programs and codes in the blink of an eye, the robotics industry seems to have fallen into a state of "stagnation".

What is holding back the "AI" in the physical world?

Why is it so difficult for robots to steadily pick up a cup of water in the real world?

What is the next key breakthrough node for the robotics industry?

Can we still place high hopes on the robotics industry?

This summer, in the opening in-depth dialogue of the "Nande Deep Talk" session, Zhang Yanmin, fund manager of Southern Fund, Wang Wenjie, founder of the robotics startup Qixuan Technology, and Zhou Hong, deputy executive chief editor of Wallstreetcn, jointly discussed robotics - a key industry that spans between technology and reality.

This is not only an in-depth discussion that directly hits the core technology of the robotics industry, but also a hard-core collision over industry, operation and business models, and even a philosophical discussion on the game between "silicon-based and carbon-based beings".

When thinking is set free, the discussion results will naturally become extremely wonderful.

Golden Quotes:

1. The outbreak of large language models is driven by the massive amount of text data on the Internet; but we do not have the multi-dimensional data in the physical world required for robotics development, such as the force and tactile sensation when hitting a nail.

2. You cannot train robots by watching videos on Bilibili or YouTube, because the physical details of the real world cannot be learned through a 2D screen.

3. Technological bubbles are often a "necessary evil" in the history of human development; without appropriate bubbles, it is impossible to stimulate disruptive innovation to the maximum extent.

4. Welding is not a romantic and ideal pursuit, but a compromise that sacrifices human health in exchange for craftsmanship. The mission of industrial robots is to do the work that is not suitable for humans.

5. General-purpose humanoid robots are just like the Apollo moon landing program. They are the dream at the top of the pyramid, and in the process of climbing towards that goal, technologies will continue to spill over and nourish thousands of industries.

6. It takes two to three months to produce a proof-of-concept PCB board in the United States, while it only takes three days in Shenzhen. This speed of hardware iteration is the Chinese advantage that Silicon Valley envies the most.

7. The current robotics industry actually follows two paths: one is to transform the environment to adapt to machines, which is called automation; the other is to transform machines to adapt to humans, which is called embodied intelligence.

8. When robots can do everything, even build machines and make inventions on their own, they will become the ultimate productive force. At that time, the meaning of human existence will face the ultimate interrogation.

9. If the value of a single robot in the future is as high as that of an automobile, which is 100,000 RMB, the future robotics industry will be a huge market, not to mention the related sectors of maintenance, part replacement, remote assistance and so on.

01

The "Physical AI Dilemma" Facing the Whole Industry

The opening of this discussion focused on a topic that has drawn widespread attention:

Why, with such powerful AI large model technologies at our disposal, can robots still not complete simple fine movements such as delivering a pill?

"The reason why large language models can generate emergent intelligence is that humans have accumulated massive amounts of text and image data on the Internet over the past decades for training. But there is no data of the physical world in libraries." Wang Wenjie, founder of Qixuan Technology, pointed out the core of the problem straightforwardly.

Wang Wenjie said that humans are born with eyes, hands, tactile sensation, perception of the surrounding environment, and hearing, that is, these composite perception systems. But for robots, it is actually very difficult to possess these perception and operation capabilities.

Zhang Yanmin then cited a vivid example: "Pictures can tell AI that this is a hammer that can be used to hit nails. But in the real world, how heavy is a hammer? How much force do you need to use to hit the wall without damaging it? These multi-dimensional sensor data are scarce in current systems. You cannot train robots by watching videos on Bilibili or YouTube."

In public perception, the quadruped robot of Boston Dynamics can do backflips, and robots in the Spring Festival Gala can dance in perfect unison, which seems to indicate that the "cerebellum" (motion control) of robots has been extremely developed. But most of these capabilities are built on traditional industrial servo control. Once out of the preset environment, robots will appear to be "well-developed in limbs but simple in mind".

This is the "Physical AI" dilemma currently recognized by the whole industry. AI in the digital world has no physical boundaries and can carry out unlimited fast trial and error iterations; while embodied intelligence must follow the laws of physics, and the generalization of every movement requires the feeding of massive real-world data. It is exactly this kind of data that is extremely scarce in the real world.

02

The Data Dilemma: How to Make Robots "Grow Up"

After understanding this problem, you can understand why institutions from all over the world are investing in "data collection" and trying their best to break through the bottleneck of physical data.

WorldLabs, a spatial intelligence company founded by Li Feifei, recently announced the acquisition of SceniX, a robotics simulation startup. This acquisition aims to use the "Real-to-Sim-to-Real" technical path to replace the dangerous and expensive real physical data collection with a high-fidelity virtual environment, so as to provide robots with a digital training ground for unlimited trial and error. In China, various "data collection factories" and "data collection bases" are springing up everywhere.

But the difficulties are a little more complicated than imagined.

Wang Wenjie explained: "Humans have adaptive capabilities - if I can't see clearly, I can find ways to see clearly. But cameras and sensors cannot do that, and the dimensions they can collect are very limited. Sound is just sound, tactile sensation is just tactile sensation, color is just color, which is far less sensitive than human senses."

Zhang Yanmin added: The current "brain" of robots has not formed real experience and consciousness, and it cannot improve itself the way humans learn movements. For today's robots, the hardware of their "brain" is already very powerful, but this "brain" does not know how to translate experience into practical solutions.

In other words, the core difficulty of current embodied intelligence is: chip performance is not a problem, the algorithm framework is not a problem, what is really missing is the cycle and process of "growth".

03

Will Robots Have an "Emergence" Moment?

A core rule of large AI models is that when data, parameters and computing power reach a critical point, the effect of the whole model will be improved at a stroke, as if intelligence emerges all of a sudden.

Then, will there be a similar "emergence" moment in the robotics industry - when data accumulates to a certain level, robots will suddenly "emerge" with unexpected intelligence?

Wang Wenjie gave a cautious judgment on this question: The breakthrough of large language models is a process in which quantitative change leads to qualitative change. But this process of qualitative change is so complicated that the people who build this system do not know why it works in this way.

This is like what happens when a neural network has more than 10 billion nodes, where the possibilities reach 3 to the power of 10 billion, which is beyond the scope of human understanding. Therefore, no one can tell clearly when the "emergence" moment of robots will come.

Zhang Yanmin then extended the topic to a more controversial field: Will machines generate self-awareness?

"Philosophically, most philosophers do not recognize that machines can have self-awareness," he said. "The current large language models learn from massive amounts of data, and when a similar situation occurs, they give answers according to probability. But can they really deduce things that are not included in their knowledge base? This is the biggest difference between humans and machines - humans have inspiration, the kind of inspiration that cannot be explained clearly."

He also mentioned the famous "Chinese Room" experiment in philosophy: a person who does not understand Chinese, with a translation manual, can pass the Turing test in certain fields - which does not mean that he has consciousness. There may be an insurmountable gap between the "intelligent performance" of machines and "real consciousness".

Wang Wenjie gave a risk warning from the perspective of probability theory: "The principle of AI is probability theory, which means there is a certain probability of exceeding expectations and extreme situations. This is inevitable, the only difference is whether the probability is one in a million or one in ten. As long as it is based on probability theory, there will definitely be a possibility of unexpected events happening."

04

The Commercial "Landing Point" of the Robotics Industry

Returning to reality from the cloud of philosophy, the focus of the dialogue turned to the most practical question:

Where is the profit model of the robotics industry?

Wang Wenjie's answer comes from practical experience. Qixuan Technology focuses on industrial welding robots, and the logic for choosing this track is clear and firm:

"Welding is the kind of work that is not suitable for humans. Thermal radiation damages eyes, protective gas hurts lungs, and dust harms health. We use robot technology to complete the work that is not suitable for humans."

The improvement of working conditions brought by this positioning is also generating real commercial value: "After enterprises use robots, the weld quality of each product is more consistent, the quality is improved, the manufacturing cost is significantly reduced, the order response speed is increased, and the competitiveness of the factory is enhanced."

Zhang Yanmin outlined a more macroscopic picture from the capital perspective: "If we believe that the intelligence of robots is getting higher and higher, application scenarios will suddenly emerge before you notice."

"At first, robots could only do welding, then they could sweep the floor, water flowers, and pick potatoes in the fields. If we follow the first principle - robots all serve humans - then for the 7 billion people on the planet, theoretically, dozens of robots of different forms will follow each person in the future."

"If the value of a single robot in the future is as high as that of an automobile, which is 100,000 RMB, the future robotics industry will be a huge market, not to mention the related sectors of maintenance, part replacement, remote assistance and so on."

Facing such a grand vision, Wang Wenjie still remained calm: "A grand prospect must be supported by current commercial returns. Relying purely on capital blood transfusion cannot last forever. Only by continuously gaining market recognition and economic returns can we continue to iterate and polish our products, and make the company's business develop faster and faster."

05

Special Robots vs General Robots, Who Will Grasp the Future of the Industry

The most enthusiastic collision in this dialogue comes from a topic that runs through the whole industry:

Special robots or general-purpose humanoid robots, who will truly change the world in the future?

Wang Wenjie's judgment is: from a practical point of view, special robots are more closely matched to customer demands, and can meet the needs of specific scenarios in the first place.

He further analyzed the paradox of humanoid robots being applied in factories: usually, the work that industrial robots cannot do in factories is left to humans to do manually. If humanoid robots are used to replace this part of manual work in the future, the work to be replaced is undoubtedly very difficult.

In addition, in reality, the performance indicators of industrial robots in terms of speed, accuracy and other aspects are currently much higher than those of humanoid robots. If the work that industrial robots cannot complete is to be handed over to humanoid robots, this is also close to a paradox.

Zhang Yanmin responded with an interesting analogy: "It's like launching satellites. Sixty years ago, launching satellites had no commercial value, but in the process, technologies A, B, C and D that were developed for the project could be applied in civil fields, and may burst out other achievements at a certain moment.

"In the process of continuous exploration, capabilities will be gradually built up."

He also maintained the unique dialectical attitude of investors: "When climbing the peak of science, we should do the difficult but right things. We know where the North Star is, so we move towards that direction. It's just that at different stages, we need to find our own positioning in business."

06

China's Advantages: Not Only Cost, But Also Speed

Talking about the competition pattern of the robotics industry between China and the United States, the two guests gave highly consistent judgments: China has incomparable advantages in the manufacturing end, and the essence of this advantage is not only cost.

Wang Wenjie cited a detail: In Shenzhen, if a startup company wants to process a part, it can find a supplier as soon as it goes out. If you are willing to pay, the finished board will arrive at the laboratory in three days. In the United States, the same thing may take two to three months.

Zhang Yanmin quoted a statement from an angel investor: To make the same consumer electronic product in Silicon Valley, the iteration time in Shenzhen is about 1/30 of that in Silicon Valley. The difference is mainly not a cost issue, but a time issue.

"And time is the most important cost."

Behind this speed advantage is China's unique innovation ecosystem. Wang Wenjie described a common scene: "If you are the owner of a small startup company, when you go to a small factory in Shenzhen to inquire about the price, they will not treat you coldly just because your company is small. Because there is a possibility that one of their small customers will grow into a giant all of a sudden one day."

Zhang Yanmin believes that the robotics industry may form a global complementary pattern of models and hardware: "Almost all of Tesla's robot supply chains are located in China, which itself speaks volumes."

In addition, there is a factor that is often overlooked: engineer dividend.

"China's excellent reserve of engineers makes American startup companies full of envy. This is not cheap labor, but strong R&D and manufacturing capabilities, superimposed with a complete industrial chain collaboration system. This is the real competitive barrier," Wang Wenjie said.

07

Bubbles and Reality: The "Necessary Process" of the Capital Market

At the end of the dialogue, the topic naturally fell on the issue that investors are most concerned about: What is the value and the bubble of the robotics track respectively?

Zhang Yanmin gave a specific data: "The current TTM valuation of the CSI Robotics Index is about 60 times. If we exclude the stocks with negative PE, the average valuation is about 120 times. Since the beginning of this year, the most popular track in the primary market is neither semiconductors nor large models, but embodied intelligence, with a total investment of about 160 billion RMB." (Data source: Wind, as of June 30, 2026)

Behind this figure is the market's extremely high expectation for this track - as well as the accompanying fluctuation risk.

But Zhang Yanmin also has his own long-term view: Technological bubbles are sometimes called a "necessary evil" in the history of human development. Without appropriate bubbles, there is no way to stimulate innovation. More than 100 years ago, the construction of railways in the United States did lead to overcapacity back then, but in retrospect, it was very useful for social welfare. The same was true of the Internet bubble in the past.

This judgment is certainly not an endorsement for bubbles, but provides a longer dimension for thinking: The key to the problem may not be whether there is a bubble in the short term, but to try to make the bubble fit the direction that society needs.

08

Epilogue: Standing on the Verge of Explosion

As the dialogue drew to a close, the three of them unanimously felt a shared sense of historical significance - what we are experiencing now is a rare moment of technological explosion.

Zhang Yanmin summed up his observation in one sentence: "The robotics industry is still in the ascendant."

Wang Wenjie used the story of his son as a down-to-earth annotation: "When my son does his homework, he takes a photo and sends it to Doubao to ask for the answer directly. The whole process of learning has undergone fundamental changes for this generation."

From the robotic arms in the welding workshop, to the smart assistants on standby at any time in the home, from the text emergence of large language models, to the tireless exploration of the real world by Physical AI. The technological revolution that human beings are experiencing now, although no one can see its full picture, its