Embodied intelligence has crossed the watershed, and investors who want to sell their old shares are everywhere across the market.
A group interview at this year's Bund Conference was extremely impressive. On that day, all the interviewees were AI-focused investors from well-known institutions. During the self-introduction session, investors from at least three institutions voluntarily stated that embodied intelligence is one of their institutions' core investment tracks.
Until the host of "Funtalk" brought up recent news that regulatory authorities have tightened the listing approval for embodied intelligence companies.
The atmosphere on the spot turned awkward almost instantly.
The investors then began to review the development of embodied intelligence, acknowledging that the industry is still in a very early stage and its technology has not yet converged.
It is fair to say that this question became the watershed of the entire interview, and that day was later regarded as the watershed of the whole industry.
Indeed, on the very same morning, news broke out that regulators had issued informal "window guidance", which included the requirements of having recurring revenue capacity, being on a track of narrowing losses, or having achieved real technological innovation into the scope of IPO approval for humanoid robot enterprises.
Prior to that, the industry had experienced at least a year of overheated boom, with startups springing up overnight and financing valuations rising steadily.
In the first half of 2026 alone, the total financing amount of China's embodied intelligence track exceeded 900 billion yuan, 5 times that of the same period in 2025. The "100 billion yuan valuation club" in the embodied track continued to expand, with more than 20 companies valued at over 100 billion yuan emerging within this half year.
For the embodied track where technology is still in the early stage of development, the pursuit from the primary market is largely based on the imagination of the industry's future. But when market sentiment kept rising, two incidents took place.
In May, Ubtech launched a companion robot "only for adults", which blurred the boundary between humanoid robots and adult-oriented products, breaking people's imagination of embodied intelligent products. In August, after Unitree went public on the A-share market, its market value nearly halved after five trading days, further breaking the imagination of the valuation of embodied intelligent enterprises.
These two incidents seem to be independent and unrelated to each other, but when the internal tension of the foam becomes too large, their successive occurrence may become inevitable.
With expectations continuing to decline, the primary market for embodied intelligence financing began to cool down. Recently, multiple investors confirmed to us that after the watershed, there are a large number of investors in the primary market who want to sell their old shares.
This is easy to understand: after the tightening of listing rules, the cycle to realize exit through IPO has lengthened and the risk has increased, so many people want to cash out and exit earlier.
Investor Zhang Xin (pseudonym) is one of them. Earlier this year, he invested in a world model startup with a post-money valuation of 3 billion yuan, but by October, the company's public valuation has risen to 22 billion yuan.
Zhang Xin recently wanted to sell his shares at a valuation of 12 billion yuan. Unfortunately, he has not found a buyer so far.
01
A team of 40 people, more than half of whom are part-time workers, only has a DEMO to show, but is seeking financing at a valuation of 10 billion yuan.
Do you think this is absurd? Sun Bo, the founder of Lingjing Intelligence, also saw this scene at this year's WRC.
Although Lingjing Intelligence is also an embodied intelligence startup, mainly engaged in the research and development of intelligent brain, this year Matrix Partners became the investor of Lingjing Intelligence's angel round, angel+ round and PRE-A round.
But Sun Bo told the above story, which of course expresses his point of view: this industry does have a tendency of foaming at a certain stage.
He also forwarded a news about Shao Tianlan to us. The news wrote that several Moments posts of Shao Tianlan, the founder of "the first stock of embodied vision, brain and hand" Megvii Robotic, tore off the mysterious veil of the commercialization of embodied intelligence
Not long ago, Shao Tianlan posted a number of consecutive Moments, pointing directly at the problem that embodied intelligent enterprises use so-called data collection centers to generate false revenue.
Shao Tianlan deduced the operation logic behind it: local governments or state-owned platforms invest in the construction of data collection centers, robot enterprises sell equipment to the centers and confirm revenue, then the data collection centers sell data back to the enterprises to make the capital flow back.
The value of the existence of data collection centers is self-evident. Different from the data of large models, which are mainly texts including various text content already deposited on the Internet that can be used as ready-made learning materials for large models, embodied intelligence has more types of data collection and a longer link.
The extreme scarcity of high-quality real-world data has become a bottleneck restricting the development of the industry.
According to Zhao Fuheng, Investment Director of Ninege Capital, even if all Egocentric data are collected now, it may not be enough for large-scale pre-training in the embodied field. It can be said that the data collection center is the key to solving the data bottleneck restricting the development of the industry.
The model of data collection center itself is not problematic. The problem lies in whether embodied enterprises should recognize this part of revenue as commercial order revenue, especially when using it as reference data for investors, whether it can reflect the real operating status of the enterprise, and whether it can prove that the enterprise has achieved a closed commercial loop.
After all, this kind of "gathering-type" data collection center order through the related party transaction of "transferring from the left hand to the right hand" does have the risk of inflating revenue.
In the field of embodied intelligence, many consensuses have not yet been formed, including of course the consensus on commercial orders. In addition to the orders from the data collection centers mentioned above, some embodied enterprises also recognize letters of intent as commercial orders.
There are many gaps between a letter of intent and actual revenue.
A set of reference data shows that in January 2026, the number of letters of intent for domestic humanoid robots has exceeded 120,000 units, but the actual global shipment of humanoid robots in 2025 was only 18,000 units. Moreover, a considerable part of these orders flow to the scenarios of technical verification and display interaction.
If all the moisture in the commercial orders in the embodied field is squeezed out, the commercialization of many embodied enterprises will become a pseudo-proposition.
In Sun Bo's view, if the embodied intelligence industry wants to achieve real commercial landing, it must simultaneously solve three core problems: reliability, real-time performance and cost control.
But for now, ontology manufacturers have not yet come up with good solutions, so revenue is still the biggest challenge they face.
Without revenue, enterprises cannot go public. If they cannot go public, investment institutions will find it difficult to exit, and they will not dare to invest if exit is hard. Under such a cycle, the financing boom of the embodied track will gradually cool down.
At present, the enterprises that have gone halfway are facing greater challenges: their valuation from the last round is very high, and the difficulty of the next round of financing is not small.
But the enterprises already have a considerable scale: they need to invest in R&D, pay employees' salaries, pay office rent, and the repurchase clauses in the financing contracts are approaching the trigger point as time goes by. They are walking on a thorny road.
02
The main reason why the embodied industry cannot achieve commercialization is that the technology has not yet converged.
Yan Qianhang, Vice President of Frees Fund, told "Funtalk" that the technology is still in the early stage of R&D, the development of embodied intelligence will not grow linearly every year as expected, and it is too early to talk about large-scale commercial landing.
He used autonomous driving technology as an analogy: back in 2012 when people faced autonomous driving, no one knew exactly where it would land, whether it would be used in mines or logistics. But with various technologies maturing and going online, there will be huge changes when people rethink the landing scenarios.
Looking back at the embodied field, many technical routes are still controversial.
For example, in terms of the underlying paradigm, there are several independent main lines now. In the past, VLA was the mainstream paradigm, but now people begin to pay attention to the world model, trying to use it to supplement the causal reasoning capability that VLA lacks.
For another example, there are constant discussions on whether the ontology needs to be humanoid, or whether the humanoid robot should be biped or wheeled, and all manufacturers are presenting their own diverse technical routes.
Many early landing practices are half-baked, and are easily crushed by stronger technologies in the later stage. Therefore, Yan Qianhang believes that after the technology has the basic maturity for commercialization, it is necessary to observe how to achieve landing.
Relatively speaking, the commercialization path of upstream component enterprises is relatively certain. Because no matter where the technology finally converges, or which ontology enterprise will win the competition in the future, they all rely on core components such as dexterous hands, sensors, and joint modules.
For example, Lingjing Intelligence is also doing the "shovel selling business". Its self-developed intelligent brain base has connected to more than 300 embodied enterprises, with an industry coverage rate of over 70%.
Multiple investors also confirmed to us that the current embodied intelligence industry is not uninvestable, but investors tend to invest in key components with high technical thresholds and know-how.
Judging from the financing situation of the primary market in the first half of this year, more than 70% of the capital flows to the midstream of the embodied industrial chain. Although the upstream only takes 20% of the total financing, 30 component enterprises in the upstream have raised nearly 9.6 billion yuan of financing, with an average financing amount of 320 million yuan per enterprise, which is far higher than the average financing level of midstream enterprises.
As for the ontology that received the most attention before, with the strong supply chain foundation in China, the prospect of product strength is certainly very optimistic, but precisely because of the consistent empowerment of the socialized supply chain, the differentiated competitiveness between ontologies will be narrowed.
There is nothing new under the sun. At the end of last century, PC manufacturers once fell into the Wintel dilemma. With the deep binding of Windows operating system and Intel processor, which jointly defined a set of unified open standards, downstream PC manufacturers could only do assembly and appearance design on the basis of this set of standards.
During this period, most of the profits of the industrial chain were taken away by the Wintel alliance. PC manufacturers not only had no right to speak at the underlying level, but also could only earn very low processing gross profit.
In fact, the ontology manufacturers in the embodied industry may face similar crises in the future.
For example, Jia Yueting relied on FF Robot to tell a new story in the capital market, and received 70 million US dollars of financing in just two months. But after digging into the product details, the outside world questioned that the FF Robot was manufactured in China, shipped to the United States for re-registration of qualifications and sold under the brand, and FF did not give a positive response to this.
In FF's official promotion, from March to June, 242 units have been shipped, and the annual shipment target has been raised to 2000 units.
How did Jia Yueting get his robots? Of course, they are OEM products. At present, China's robot OEM industrial chain is very mature. In Huaqiangbei, you can find almost all parts and OEM factories. Many factories even provide ODM services, supporting small-batch, rapid proofing and iteration.
Just like traditional PCs, robots have also entered the "1-meter cabinet" era.
03
Although it is necessary to demystify the embodied industry, this has not affected the consensus on the future of the embodied industry.
But at the moment when the technology has not yet converged, we need to face up to the existence of foam, and at the same time figure out what kind of foam this is.
Generally speaking, there are two types of foam. One is value-type foam, specifically for embodied intelligence, which refers to the gap between what people expect it to solve in the future and what it actually solves in the end. The other is scale-type foam, which refers to the expectation that there will be 100 excellent embodied companies in the world in the future, but in fact there may only be 10.
If you can't understand the difference between these two foams, just look back at the 100-group war in 2014. At that time, the judgment on local life services was not wrong, but it was too exaggerated to have 100 companies doing group buying and food delivery.
Obviously, the foam of embodied intelligence today is also this kind of scale-type foam. Everyone knows very well that many companies will disappear, but as Zhao Fuheng said, investors are still willing to continue to invest money to see if the embodied track can produce a giant like large language model.
The more implicit meaning of this sentence is the current mentality of the primary market: they would rather invest wrongly than miss the opportunity. This mentality to a certain extent adds fuel to the embodied intelligence boom.
The secondary market is of course much better, because targets can be bought at any time. So although the secondary market will also be affected by market sentiment, it will always return to the mean after a period of time.
This is why the inversion between the primary market and the secondary market is emerging in the embodied industry at present.
Many investors cited the example of dexterous hands. The current star company in the market is Lingxin Qiaoshou, a startup engaged in dexterous hands. Market sources say that the valuation of the company in the last round was about 20 billion yuan, and its target valuation in the next round exceeds 40 billion yuan. Another source says that Lingxin Qiaoshou is already preparing for Hong Kong IPO, and optimistic investors expect its market value to exceed 100 billion yuan after listing.
Interestingly, the capital market is not short of companies making dexterous hands. For example, Zhaowei Electromechanical (003021), which is recognized as the most representative manufacturer of dexterous hands, its latest dexterous hand has 20 degrees of freedom and it is a supplier of Tesla. However, the market value of Zhaowei Electromechanical is only about 17 billion yuan.
However, how much a company is worth is a later story. First of all, it has to be able to go public. An investor judges that after the new regulatory rules, Lingxin Qiaoshou's listing plan is likely to be greatly postponed.
Of course, more than one company has been affected by the new regulatory rules, and the investment recovery cycle of this industry is being greatly extended. In this process, various possibilities will increase: first, valuation decline; second, failure in competition; and of course the most terrible one is startup failure.
This is the reason why investors like Zhang Xin want to sell their old shares. The good news for this group of sellers of old shares is that they entered the embodied industry relatively early, usually in last year or early this year. With more than a year of capital carnival, even if the value of these equity has fallen, it is enough for investors to make profits.
The more uncomfortable ones are the investors who entered around the middle of this year. At least in a relatively short period of time, they are the takers of this passing-the-parcel game. To a certain extent, this group of investors will become the determined accompanying runners of this industry until the industry ushers in a real inflection point.
As for those embodied intelligence startups, of course no one is more painful than them. The whole industry objectively still needs to go through large-scale technology and commercial verification, so these companies will need capital support for a long time.
But facing the cooling market, the difficulty of their financing is much greater than before. The equity circulation in the primary market also adds difficulty to them. Take the company that Zhang Xin invested in as an example: when there are old shares being sold at a valuation of 12 billion yuan in the market, who will inject capital into the company at a valuation of 22 billion yuan?
An investor told us that in order to restrict the transfer of old shares, many embodied intelligence companies are now unwilling to promote share reform.
From a business perspective, this is of course a rational choice. As Yan Qianhang said in the interview, before the arrival of the industry inflection point, everyone should adjust their expectations and stay relatively rational.