Embodied intelligence funding cools: September funding down 40% month-on-month, old shares hard to sell even at a discount

"(Embodied AI projects) I still look at them, but I won't invest anymore."

In October 2026, when asked about the investment heat around embodied intelligence, Chen Yuan, a well-known domestic investor, told Tencent Technology this. He then offered an even more pessimistic judgment: "The primary market AI investment has now entered garbage time."

Chen Yuan's institution is one of the most aggressive early-stage funds investing in embodied intelligence in China. At the end of 2022, he judged that the capabilities of large models would extend into the physical world, and since then he has bet continuously on robot bodies, embodied brains, and world models. In the past, his team might focus on discussing several embodied intelligence projects in a single week; now there is not even one on average per week, and sometimes only one every few weeks. More energy is spent taking care of companies already invested in.

According to NE Times statistics, in September 2026, 46 financing deals occurred in China's embodied intelligence primary market, with a reference amount of about 11.624 billion yuan, down 4.2% and 40.1% respectively from August. This decline does not fully reflect the changes in the market—some projects whose investment decisions were completed as early as May or June often only announce closing in September. Chen Yuan estimates that if only the investment amounts truly newly approved by investment committees that month are counted, it may be only 20% to 30% of the level during the hot period in the first half of the year.

The changes in the old-share market are even more direct. An investor told Tencent Technology that before August, old shares of some leading robot companies "sold in minutes," and buyers could be found at about a 10% discount; recently, in some transactions he has been involved in, quotes as low as 50% of value have appeared.

According to an October 29 report by Securities Times, in the past two months, transfer quotes for old shares of some embodied intelligence companies have clustered at 65% to 80% of the previous round's valuation, and the targets for sale include leading companies. Even at a discount, some early shareholders can still obtain several times their returns, but buyers are no longer as active as before. An investor said that only institutions that believe star companies will eventually go public might be willing to take over.

A striking contrast is Zhiyuan. Several investors recalled to Tencent Technology that at the beginning of the year, Zhiyuan did not accept any investment, and a large number of investment institutions could not get in even with money and connections. In July this year, Zhiyuan announced the start of preparations for a Hong Kong stock listing. But by October, Zhiyuan began advancing a new financing of 1 billion to 1.5 billion yuan at a valuation of about 50 billion yuan. A person close to the transaction told Tencent Technology that in this round Zhiyuan required investors to complete closing as soon as possible.

A pessimistic view has begun circulating in the industry: there may be no new embodied intelligence star company going public this year.

The market seems to have suddenly turned cold—as everyone had long expected, yet so unexpectedly.

A group of investors who once competed to enter can no longer wait to exit.

But stopping betting is easy; getting back the money invested over the past few years is not.

01 Too much invested, too slow to sell

The bullets in the primary market have not been used up, but the previously smooth capital relay is failing, first because investment institutions have found it difficult to find enough new projects with sufficient differentiation.

Zhou Heng is responsible for AI investment at a well-known industrial capital. In his personal view, by May this year, embodied intelligence investment had already begun to cool, mainly because the institutions charging ahead had invested about enough. Robot startups continue to emerge, but many mainstream funds already hold three or four robot companies.

He told Tencent Technology that some people in the industry have even invested in 30 robot companies. "That means you have to persuade the investment committee why exactly a new project must be invested in?"

Project prices are also getting more and more expensive. Information obtained by Tencent Technology shows that both Zhipingfang and Zibianliang Robotics achieved a doubling of valuation from 10 billion yuan to 20 billion yuan in about four months. Xinghaitu's valuation rose from 10 billion yuan to 20 billion yuan in less than two months. An investor told Tencent Technology that in 2024 he looked at a project whose angel round was quoted at 300 million yuan, and he did not invest. By 2026, this project claimed a valuation of 40 billion yuan.

The speed of valuation growth has made it increasingly difficult for later-round funds to calculate returns.

Over the past few years, the high valuations of a large number of robot startups have been built on the expectation of future entry into the public market and obtaining higher pricing. Unitree's listing provided the primary market with a new pricing reference. On August 19, Unitree Technology officially landed on the STAR Market with an issue price of 150.80 yuan; on its first trading day, it once reached 1,100 yuan intraday and closed at 845 yuan, but after listing, Unitree's stock price quickly experienced severe fluctuations. By October 15, its intraday low touched 404.03 yuan, a pullback of more than 60% from the first-day high; afterward the stock price rebounded quickly and closed at 647.05 yuan on October 30, still about 41% lower than the first-day intraday high, with a total market value of about 261.7 billion yuan.

"Before Unitree went public was the best time to exit," Xu Zheng said. But he himself failed to seize this opportunity.

Xu Zheng is responsible for exits at a well-known industrial capital. Over the past year, his institution invested in two robot companies, both at valuations of more than 2 billion yuan when entering, and both now exceed 15 billion yuan. Calculated by book valuation, these are already two investments with very considerable returns, but a few months ago Xu Zheng was persuading colleagues to consider selling some shares, while colleagues felt they could wait a little longer, hoping the company's valuation could continue to rise.

What Xu Zheng worries about is that by the time everyone wants to sell, there may be no buyers left. "When you want to withdraw, everyone wants to withdraw."

What worries some institutions more than secondary-market volatility is whether robot companies can go public as scheduled.

On October 13, Caijing reported that Zhiyuan Innovation is advancing a new financing round of approximately 1 billion to 1.5 billion yuan, at a latest valuation of about 50 billion yuan. Zhiyuan previously announced in July that it had begun preparations for a Hong Kong stock market listing, but under Hong Kong Stock Exchange rules, if pre-IPO investment is completed less than 28 full days before the first submission of the listing application and the investor still holds shares at the time of submission, or if it is completed on or after the day of the first submission and before listing, the HKEX will generally require that the first trading day of the applicant's shares be deferred until 120 full days after the completion of the last pre-IPO investment, except in extremely special circumstances. This means that after Zhiyuan continues to raise funds, completing an IPO this year will already be very difficult.

A person engaged in Hong Kong listing services told Tencent Technology that in the first quarter of this year, among the three companies with the highest valuations in the embodied intelligence industry at the time, apart from Unitree, the other two were already pushing forward with their HKEX listing submissions. But by October, neither company's listing progress had met previous expectations. "Both are stuck."

Another Hong Kong listing transaction lawyer told Tencent Technology that Chapter 18C of the HKEX provides a listing path for eligible specialist technology companies, under which companies that have not yet commercialized do not need to meet the revenue threshold of HK$250 million in the most recent audited financial year applicable to commercialized companies, but they need to disclose a credible commercialization path, including how they will reach that revenue level in the future. She told Tencent Technology that if robot companies currently have relatively little revenue, the Stock Exchange will further focus on the number of customers, orders on hand, commercialization plans and future profit paths. The high-growth forecasts that companies previously presented to investors must be supported by more actual operating evidence at the IPO stage.

More sensitive is revenue quality. The above-mentioned lawyer told Tencent Technology that one project was found during the due diligence stage to have circular contracts, and the sponsor ultimately stopped advancing it. For robot companies, if a company that submits its listing first is found to have major problems in customer authenticity, related-party transactions or revenue recognition, companies that subsequently adopt similar business models may also face stricter verification. In particular, procurement and data transactions formed around data collection centers and local industrial platforms—whether the revenue can be sustained and whether customers are independent—will become issues that need to be explained.

She said that some robot companies are eager to go public not only because the founding teams themselves want to speed up progress, but also because of pressure from existing investors. Some investment agreements stipulate listing deadlines, for example requiring companies to complete an IPO within 18 months. Well-known investment institutions behind star companies often have stronger bargaining power and will also impose stricter requirements on the company's listing timetable.

02 A former consensus

Over the past three years, embodied intelligence can be called the track with the strongest wealth concentration effect in the primary market.

Lin Chuan initially was not optimistic about embodied intelligence. The RMB fund where he works previously focused mainly on semiconductors and AI, and he himself came from a semiconductor investment background. In 2024, he felt that the enthusiasm for robots was "completely incomprehensible."

But by 2025, after more and more institutions began to make arrangements, he found that if his fund still had not invested in robots, he would instead need to explain the reason to LPs. "Everyone around me is investing in embodied intelligence. If I don't invest in even one embodied intelligence company, it will be too difficult."

He ultimately found a reason that could convince himself: if robots really become a huge market, what kind of chips, computing power and controllers will they need? For a semiconductor investment institution, studying the development of downstream robots is also studying future semiconductor demand.

FOMO sentiment enveloped the entire industry. Investors' confusion was that regardless of whether they understood it or not, if this was the next-generation AI platform and they had not positioned themselves in advance, they might be left behind by the times.

Annual statistics released by IT Juzi show that in 2023 there were 42 domestic embodied intelligence financing deals, amounting to about 8.676 billion yuan; in 2024 the number increased to 103 deals, amounting to about 9.355 billion yuan. In 2025, financing events reached 325, with an amount of 39.832 billion yuan, up about 216% and 326% respectively from 2024. In this statistic, early-stage investment still accounted for 62%. A large number of startups had already obtained hundreds of millions of yuan or even larger financing while still in the R&D and technology validation stage.

Leading companies began to attract large industrial capital. Robot investment gradually expanded from early-stage institutions' technology bets into large-scale capital competition involving internet giants, manufacturing leaders, and national and local industrial funds.

Funds did not flow evenly to all startups. Statistics from CVSource show that among 168 embodied intelligence companies that received investment in 2025, the top ten companies including Varifold, Tashi Zhihang, Leju Robotics, Xingdong Jiyuan, Galaxea, and Zhongqing Robotics together obtained 13.472 billion yuan in financing, accounting for 40.95% of the total financing amount for the year under that statistical scope.

By 2026, some companies had already grown large enough. Statistics from Zhidongxi on June 29 this year show that in the domestic embodied intelligence field there were already at least 26 companies valued at more than 10 billion yuan, of which 16 newly entered this tier in the first half of 2026. At present, at least seven leading companies—including Zhiyuan, Galaxy General, Galaxea, Qianxun Intelligence, Varifold, Zhipingfang and Lingxin Qiaoshou—have reached valuations of more than 20 billion yuan.

A project may have been valued at 5 billion yuan a month ago, become 10 billion yuan the next month, and reach 15 billion yuan another month later.

An early-stage investor told Tencent Technology that she had once been in contact with a robotics startup. The founding team initially gave a financing valuation. After evaluation, the investment institution believed the project was worth investing in, but the valuation could be even higher, so it proactively suggested that the founder raise the quote and helped push the company's valuation up.

In ordinary equity investment transactions, investors try to negotiate a lower price and obtain more shares. But when embodied intelligence was at its hottest, the financing price itself also became part of a project's value. A higher quote could make a company appear more attractive among a crowd of star projects, and the price of the next financing round in turn became proof of the reasonableness of the previous round's valuation.

In the eyes of many people at the time, primary-market capital was abundant, and the listing performance of some AI model and GPU companies gave robotics startups an imaginable exit price. Even though some people had already begun to question valuations, new buyers kept appearing in the market.

03 A valuation of ten billion yuan needs new proof

Capital's enthusiasm was in turn changing entrepreneurship itself.

An entrepreneur from the autonomous driving industry told Tencent Technology that in the first half of this year, he had intensive contact with investment institutions and tried to prepare to found a robotics company. At the time, friends in VC generally told him that if he could not raise 100 million yuan in the first round, he might as well not start the business at all.

At that time, he had not even determined a specific entrepreneurial direction. In order to obtain the "entry ticket," he spent several months looking for partners, hoping to invite a top technical expert to co-found the company, but in the end he was unable to build an ideal team.

During that period, a listed company proactively proposed cooperation to him. He recalled that out of market value management needs, the other party was willing to support him in setting up a robotics company, and had even prepared a product that could be used for display. "The product of a certain robotics company that their company had already purchased—if I was willing, they could change the shell, change the logo, and give it directly to my new company for use."

An industry consensus is that the capital market's pursuit of the robotics concept has already given rise to some operations unrelated to technology R&D and real orders.

In September 2026, Mech-Mind founder Shao Tianlan publicly questioned some of the industry's data collection centers and related-party transaction models, triggering a debate in the embodied intelligence industry about the authenticity of revenue. This kind of dispute over revenue quality also came into the view of investment institutions. After financing speed declined, investment institutions began to re-examine robotics companies' commercialization capabilities. Companies that in the past could obtain high valuations relying on founder backgrounds, technology demonstrations and financing progress now must answer questions about order delivery, customer payments and sustained revenue.

More and more people began to feel the industry's shakeout trend. On October 23, National Business Daily reported that Nanjing embodied intelligence company Wujing Dynamics had fallen into a wage arrears and work stoppage crisis. The company was founded in February 2026. Local media previously reported that about four months after its founding, Wujing Dynamics had announced that 110 million yuan in financing was in place, had signed framework orders worth nearly 500 million yuan with a number of central state-owned enterprises, and had another nearly 500 million yuan in local market reserve orders.

But by October, some employees told the media that the company had been in arrears on wages since August and later stopped paying housing provident fund and social insurance, involving about 30 to 40 employees, with cumulative wage arrears of nearly 8 million yuan. Some employees had begun to seek unpaid wages through labor arbitration. The media contacted Yuan Haihui, who had previously publicly identified himself as CEO. He said he could not reach the company's actual controller and it was not convenient for him to respond to the relevant situation.

Beyond commercialization, progress in general-purpose models also began to change investors' judgment of robot brain companies. Over the past two years, a large number of robotics startups chose technical directions such as VLA and world models, hoping to build general-purpose robot brains capable of understanding environments, planning tasks and controlling actions. An important premise for their high valuations was that robots need independent basic intelligence systems, and companies that master this capability could become new platform-type companies.

On October 22, Unitree Technology founder Wang Xingxing showed a video in a public speech of a robot controlled using GPT-6 Astra. According to his introduction, without additional training for specific tasks, the robot was already able to complete a painting according to prompts and autonomously complete operations such as changing paint and cleaning brushes. Wang Xingxing believes that general-purpose large models have already shown signs of "breaking through" physical AI.

Chen Yuan believes that general-purpose models can already become a new capability benchmark. If a company claims to be developing a general-purpose robot brain or world model but cannot demonstrate capabilities beyond those of systems driven by general-purpose models in related tasks, it needs to rethink its technical direction. The intelligence capabilities that robotics startups once needed to invest large amounts of money to build may rapidly lose their scarcity as general-purpose models advance.

Robot body companies have also begun to face another set of valuation standards. Chen Yuan believes that in the past, investors mainly priced companies according to the market size robots might form in the future. Even if a company did not yet have much revenue, as long as it had a chance to become a future general-purpose robotics platform, the market was willing to give it a price-to-sales ratio of 100 times or 200 times. But when robots truly enter the stages of manufacturing, delivery and large-scale production, companies' competitiveness will increasingly depend on supply chain management, product costs, production efficiency and customer repeat purchases. Growth in hardware sales revenue needs to be measured together with manufacturing costs, R&D spending and profit margins.

In other words, as robots gradually move from cutting-edge technology products toward large-scale manufacturing, the 100-times and 200-times price-to-sales ratios that the market was willing to accept in the past may fall to 50 times, 20 times or even 5 times in the future.

04 Who can hold on until the next round

After the capital tide receded, the situations of leading robotics companies and mid-tier companies are diverging.

Lin Chuan remains optimistic about the long-term prospects of embodied intelligence. He believes that a group of leading robotics companies already hold more than 2 billion yuan in cash reserves, which, even without new financing in the short term, would be enough to support R&D investment for one to two years. In his view, leading companies will not quickly collapse because of this round of capital cooling; for leading companies that still have large amounts of cash on their books, the market cooling may even create new opportunities. Acquire excellent teams and key technologies at lower prices.

He believes that the ones truly struggling are the mid-tier companies valued at 6 billion to 8 billion yuan that lack technological and product distinctiveness. These companies have often already gone through several rounds of financing and built relatively large R&D teams, yet still do not have enough commercialization revenue to support operations. If they continue to raise funds, investors may demand a lower valuation; if they insist on the original price, new capital is unwilling to enter. Some companies will have to cut R&D investment or even choose a new product direction.

Zhou Heng has observed a similar divergence. In his judgment, embodied intelligence still presents plenty of opportunities, but financing is becoming an increasingly obvious polarization: the earliest star teams can still attract capital, and leading companies that have already entered the IPO preparation stage also have opportunities to continue raising funds; the most difficult position is that of the middle rounds.

For excellent teams newly founded this year, the first round of financing may still reach a very large scale; by the second round, the backgrounds of star founders and their technical directions are still attractive; but by the third round of financing, investment institutions begin to require companies to demonstrate capabilities and progress different from those of existing leading companies.

Zhou Heng has already invested in a leading robotics company valued at more than 15 billion yuan, and news of its executives leaving to start businesses has been constant recently. Zhou Heng said he is not in a hurry to take the company public. In his view, the robotics industry needs to be judged over three years, five years or even longer. A RMB fund may have a lifespan of seven or ten years, and investors must work backward from the final exit time to determine their investment strategy. If a technology needs five years to achieve commercialization, then the fund should be prepared to wait five years when it enters.

However, when asked whether the relevant investment agreement includes a buyback arrangement, he said that of course it does.

"The best era of financing has passed," Chen Yuan said. "The kind where you write a 10-page PPT, assemble a lavish team, and raise three rounds in half a year just on a narrative—that era is over."

After we finished talking, he hurried away to attend a mobilization meeting for business matchmaking among the embodied intelligence companies they have invested in.

(Chen Yuan, Lin Chuan, Xu Zheng, Zhou Heng and others in this article are pseudonyms.)

This article comes from "Tencent Technology," author: Gu Lingyu, editor: Xu Qingyang, published by 36Kr with authorization.