AI News

Embodied intelligence financing cools down—who can survive the industry shakeout?

36Kr · AIUpdated Nov 16 20:39

AI summary

In October 2026, domestic embodied intelligence sector financing amounted to only 33 deals, with total funding just over 10 billion yuan, a marked decline from the monthly average of more than 15 billion yuan in the first half of the year. The article reviews triggering factors including the bankruptcy liquidation of DeepBlue Robotics and Yinghe Robotics, the halving of Unitree's market value after its listing, rumors of tighter IPO rules, and disputes over the authenticity of orders, and notes that capital is further concentrating toward leading players, with money, scenarios, and scarce data becoming the three conditions for staying at the table.

Why read this

AI score 69

From financing, orders, and IPO exits to the impact of general-purpose large models, this article sorts out the multiple triggers behind the cooling of embodied intelligence and the trend toward concentration among leading players.

Topics

Tags

Body · AI translation

Zhang Shuo left an embodied intelligence company a week ago. The company had been founded less than three years earlier, had already reached Series B funding, and its main product was quadruped robot dogs. He was responsible for to-B key account sales, and in the second half of last year he had even signed several large orders.

In July and August of this year, the situation began to change. Final payments on projects were repeatedly delayed, and to obtain an acceptance form he sometimes had to make four or five trips to the customer's factory. The customers' responses were much the same: the product was unstable in operation and had not yet met acceptance standards, so the final payment would have to wait.

At the same time, the company had been continuously cutting expenses over the past few months. Last month, the boss said at an internal meeting that without new financing coming in, the company could last at most another eight months. Zhang Shuo then decided to leave.

The cooling of the embodied intelligence industry has become a consensus among many practitioners and investors.

Not long ago, capital was still pouring frantically into this industry. IT Juzi statistics show that in the first half of 2026, total domestic embodied intelligence financing exceeded 90 billion yuan, and more than 20 companies had valuations above 10 billion yuan. From body manufacturers and data collection companies to dexterous hands, joint modules and tactile sensors, almost every sub-sector was covered by capital.

Entering the second half of the year, a series of changes gradually dispelled the clamor. In October, there were only 33 financing events in the domestic embodied intelligence sector, with a total of just over 10 billion yuan. Compared with the average monthly financing amount of more than 15 billion yuan in the first half, that is a drop of one-third. This is even taking into account the low point in financing scale around the Spring Festival in February and March; if one looks only at June, the month with the highest financing amount in the first half, when monthly financing totaled nearly 25 billion yuan, then by comparison the October figure was more than halved.

Some investors have begun to turn pale at the mention of embodied intelligence. Startups cannot obtain new financing, and some previously agreed financing has still not arrived.

At what point exactly did these changes begin to appear? Some believe it was the news of tighter IPOs, which blocked exit paths; others believe it was the plunge after Unitree's listing, as the inversion between primary and secondary markets forced a reassessment of the entire industry; still others think it was the public debunking of "fake orders" in the industry by industry insiders.

In any case, the chill has already spread through the industry. The particular predicament of the embodied intelligence industry is that it was ripened by capital faster than many previous hot sectors. Even before clear application scenarios and business models had formed, and before technical routes and product forms had converged, large amounts of capital had already placed their bets.

English text is being prepared.

01. More discerning money and orders not yet fulfilled

Li Xiang is a member of the class of 2026. This fall, he wants to find an algorithm position in embodied intelligence. Hearing from seniors in the previous cohort, he learned that such positions offer good pay, have large demand, and are not as fiercely competitive. But after looking around, he found that top embodied intelligence companies have released very few positions.

He also interviewed for algorithm roles at some second-tier embodied companies, but most went nowhere. After talking with other job seekers in his cohort, he gradually came to feel that the problem probably cannot be attributed solely to higher hiring requirements. There really are fewer positions to choose from, and some companies do not seem to have urgent hiring plans. The industry that was seen as a source of job opportunities last year has left him with far fewer options than he expected this year.

Hiring is just one slice of the industry's changes. The chill is spreading gradually through the job market, startups and investment institutions, and those first affected are often the companies that have not yet formed stable revenue but still need continuous investment in R&D.

A CTO at an embodied data services company told "Dingjiao One" that the company had basically finalized a financing round in July, but it is still stuck at the letter-of-intent signing stage. The other party's stated reason is that the process is still underway. Institutions that missed the company's previous financing round and were eager to get into the next round's list of investors have also mostly stopped responding.

If you cannot get money from investors, you can only get it from customers, but for most embodied intelligence-related companies, commercialization has not yet formed stable cash flow. Even if orders are obtained, delivery, acceptance and payment collection may still continue to consume funds.

Investors are no longer keen to discuss embodied intelligence, and the number of related projects going to investment committee meetings has clearly dropped rapidly. A US dollar fund investor told "Dingjiao One" that US dollar funds usually enter earlier than RMB funds. In his view, when investing in bodies in 2024 and in brains and data in 2025, the valuations of the relevant companies were already very high. If they only realized there was a bubble by the second half of 2026, their reaction would be far too slow.

IT Juzi statistics show that in the first half of 2026 alone, at least 19 companies in the robotics sector, including those in bodies, data, brains and dexterous hands, crossed the $1 billion valuation mark. On average, a new unicorn emerged in the industry in less than 10 days.

Even some leading companies, such as Zibianliang, Zhipingfang and Lingxin Qiaoshou, crossed the RMB 20 billion valuation threshold by the end of June. One project held by the aforementioned US dollar fund investor saw its valuation increase tenfold within half a year as of the end of March this year. After prices rose rapidly, new entrants need to pay higher costs for the same technology and team, while potential returns are further compressed.

This kind of capital frenzy has left later investors able to do nothing but stand back. Since the second half of the year, many FAs have stopped taking on embodied intelligence projects, citing either valuations too high to sell or a lack of scenarios that makes it hard to explain to investors where revenue will come from. The long-term narrative that was once widely accepted now requires more real-world evidence.

Changes in the exit environment have deepened this concern. At the end of August, tighter listing rules and higher thresholds for embodied companies made exits through the secondary market harder to assess. A chain reaction quickly followed in the primary market, with some investors beginning to sell old shares of embodied intelligence companies at a discount. Loosening prices for old shares not only affect pricing for new financing but also cause potential investors to slow their decisions; the more a company needs capital, the weaker its negotiating position becomes.

Flood irrigation is over. The valuation and financing path once propped up by industry hype, a few demos and glossy academic backgrounds has reached its end.

02. When did the turning point appear?

The cooling did not start suddenly on any single day. Over the past six months, cracks first appeared in company operations, then spread to market pricing, exit expectations and technology judgments.

In May, DeepBlue Robotics (Shanghai) entered bankruptcy liquidation proceedings. Its parent company, DeepBlue Technology, was once an AI unicorn valued at more than RMB 10 billion and backed by institutions including Yunfeng Fund and CICC. A star-studded background did not prevent risks from surfacing.

More survival problems followed. In August, Yinghe Robotics, which had received investment from Meituan-affiliated entities, ByteDance and other institutions, was accepted by a court for bankruptcy liquidation. This company also had a strong starting point: founded in Shenzhen in March 2020 and incubated by Yingfeng Group, it reached a valuation of $500 million just 10 months after its founding and raised more than RMB 600 million in total. But the large amount of financing ultimately failed to translate into sustained operating capability.

Later came Wujing Dynamics, founded only eight months earlier, which suspended work and owed wages. Of the RMB 110 million in financing it claimed, a key industrial fund did not close on schedule, and its cash flow deteriorated accordingly. These companies faced different situations, yet all exposed a similar problem: external funds did not arrive on time, and the companies lacked sufficient buffer.

A broader shift in sentiment emerged in mid-August. On the 19th of that month, Unitree Technology went public, and its share price then fell sharply, with its market value at one point halved. As one of the most closely watched companies in the industry, its performance in the secondary market provided a new valuation reference, and the primary market's existing pricing began to be tested by the public market.

By the end of August, another piece of news was circulating in primary-market circles. It said IPOs in the embodied sector would be tightened, and loss-making companies could face higher thresholds: they would not only need to be in the industry's first tier but also demonstrate a clear expectation of operating improvement over the next three years, rather than continuing to sustain their story through R&D spending.

Although regulators have not yet publicly confirmed these claims, the rumors have already affected investment decisions. For industries that rely on public listings for exits, changes in expectations often take effect before formal rules do.

The climax of the turning point came in September, when the controversy over the authenticity of orders in the embodied intelligence industry became public. Shao Tianlan, founder, chairman and CEO of Mech-Mind, publicly criticized Galbot and aimed his criticism at "assembled-deal entrepreneurship." He questioned that some companies use related-party transactions with data collection centers and leasing companies to create false and unsustainable revenue; they lack PMF yet keep issuing big announcements and seek to go public just two or three years after founding. In response, Galbot later issued a statement stressing that the company adheres to genuine R&D investment and real application scenarios and does not engage in pointless war of words.

Under the conventional model, data collection centers buy robots, collect data in simulated production environments, and then sell it to embodied intelligence companies or large-model companies. Both equipment procurement and data services can constitute normal transactions; the controversy centers on the flow of funds and related-party relationships in some transactions.

The practice Shao Tianlan criticized is this: an embodied intelligence company forms a joint venture data collection company with a local government, and the data collection company then purchases robot bodies from that company. The funds are mainly borne by the local government, while the embodied company contributes no capital or only a small portion. The data generated by the data collection center is then bought back by the embodied company. Once the cycle is complete, the embodied company obtains orders and revenue, and the relevant funds flow back to the data collection center in the name of data procurement.

As the controversy continued to ferment, whether book orders can represent real demand became an unavoidable question in the industry.

The successive "bad news" did not stop there. In early September, after OpenAI released GPT-6 Astra, a large number of tests using the model to control robotic arms appeared on social media. Some practitioners found that general multimodal foundation models already possess spatial understanding capabilities, and in some tasks their success rates even exceed those of some embodied models.

After the test videos spread, sayings such as "general large models will sooner or later eat embodied models" began to circulate. Wang Xingxing, founder of Unitree Robotics, even said that after GPT-7 is released next year, it may directly break through the embodied intelligence problem.

On November 10, Anthropic further reinforced this judgment. Its released Claude Fable 5.2 achieved an average success rate of 64.9% across 42 tasks in the RoboDojo Sim benchmark, higher than GPT-6 Astra's 29.8% and also better than most robot-specific models.

Companies that disappeared, valuations cut in half, raised thresholds, pierced window paper, and barbarians breaking in from outside the gate—all of these combined to rapidly cool the booming embodied intelligence industry, making it hard for the market to continue using the optimistic assumptions of the first half of the year.

03. Conditions for staying at the table

The industry's cooling has not been uniform—not all companies feel the same chill. As total financing declines, capital is further concentrating in the leading players. This trend was already evident in the first half of the year and is continuing in the second half.

Statistics from iiMedia Research show that in the first half of 2026, industry financing flows followed a progressive logic of "brain > body > scenarios." Financing for the embodied intelligence brain track reached 59.544 billion yuan, with a relatively high average deal size, reflecting capital's preference at the time for software capabilities such as decision-making algorithms and multimodal perception.

Companies engaged in body manufacturing and those with deployable scenarios also obtained relatively high financing. Capital did not bet evenly across the entire industry; more funds flowed to the segments believed more likely to build barriers.

In the second half of the year, this selectivity became even stronger.

In July, Xingdong Jiyuan and Zhuji Dongli each completed financing of more than 1 billion yuan. In August, Paxini and XPeng Robotics received capital support. In September, Digua Robotics and Sudo Technology also completed financing of more than 1 billion yuan. They still largely fall within the popular directions of brains, bodies, and scenarios.

The prices and liquidity of leading companies are also recovering. At the end of October, Unitree Technology's stock price rebounded by more than 70% cumulatively over 12 days, and its total market value recovered to more than 280 billion yuan. Zhiyuan's valuation was about 15 billion yuan in May, and its latest financing round in October valued it at 50 billion yuan. This shows that capital is still providing support to a small number of companies—but this treatment no longer belongs to the entire track.

After entering the clearing stage, the conditions for staying at the table have gradually become clear: money, scenarios, and scarcity have become three unavoidable variables.

The most urgent is money. Most embodied intelligence companies are still loss-making, lack mature cash-generating capability, and require continuous investment in R&D, production, and delivery. After changes in the financing environment, cash reserves directly determine how much time a company has to complete product validation and whether it can survive until the next financing round.

Companies that can hold on include some that have already obtained cash through sales and services, some that can still secure investor support, and others that rely on reserves left from previous financing and extend their survival time by cutting expenses. For the last group, careful budgeting will become a management style they must adapt to.

This may also force some companies to rethink their past operating strategies. When the industry was hot, the faster the financing, the easier it was for companies to simultaneously expand R&D teams, product lines, and business scale; but after the financing environment changed, pushing multiple business lines forward in parallel may turn from a growth advantage into a financial burden.

English text is being prepared.

So a large financing scale does not mean a company will necessarily survive longer. Whether it can promptly shrink its battle lines and concentrate limited funds on the products and scenarios most likely to achieve breakthroughs will become another competitive capability.

The second condition is specific scenarios. Many companies are currently promoting application deployment, but how many products can truly work reliably remains doubtful. Entering a factory, warehouse, or store only means gaining a validation opportunity; whether a scenario can be turned into a business depends on whether it can operate continuously and create calculable value for customers.

An investor in the embodied intelligence industry told "Dingjiao One" that he currently looks only at B2B scenarios when evaluating projects, mainly B2B industrial applications such as 3C electronics factories. In his view, most other B2C scenarios are still telling stories; only after B2B applications mature does it make sense to discuss B2C, and this time window will take at least 2 to 3 years.

Compared with unstructured environments such as homes, some industrial scenarios have clearer task boundaries, and it is easier to measure whether robots create value. Accordingly, the advantages of some companies are not necessarily reflected in being at the technological frontier; as long as they form stable product and service capabilities in a certain niche scenario, they have the opportunity to reduce their dependence on external financing through sustained revenue.

The third condition is scarcity, especially in data. As general large models such as Astra and Fable begin to cross over System 2 and take over some of the capabilities of System 1, it has become more difficult to tell the story of a general embodied brain alone.

For companies that possess high-quality real-machine data, tactile data, and capabilities for training and evaluating complex tasks, scarce data may become a moat. For core hardware manufacturers such as dexterous hands, sensors, and joint modules, if they can form advantages in performance, reliability, or cost, they likewise have the opportunity to become an indispensable link in the industry chain.

There is also another type of company that may not pursue general capabilities but instead deeply cultivates a certain niche scenario over the long term. Their understanding of customers' production processes, accumulated experience in troubleshooting, and delivery capabilities cannot be immediately replaced by an even more powerful foundation model.

From this perspective, the changes brought by general large models may make the situation of some embodied brain startups even more difficult, but may also create new opportunities for companies with advantages in hardware, data, and scenarios.

Embodied intelligence is still regarded by many investors as a huge opportunity, but there is still a long distance from ideal to reality. Technology validation, delivery, and payment collection often take years, while financing and valuations change quarterly or even monthly. When capital loses patience, for an industry that is still at a very early stage, it will inevitably undergo drastic changes. Surviving will become harder, but companies that truly stand up to validation will also have the opportunity to gain greater room for development.

*The images in this article are from Unsplash.

This article is from the WeChat public account "Dingjiao" (ID: dingjiaoone), author: Wang Hanxing, editor: Wei Jia, published by 36Kr with authorization.

Source

36Kr · AI