HomeArticle

Who can survive the harsh winter of the embodied AI industry?

定焦One2026-11-16 09:23
From competing for financing to competing with real competence.

Zhang Shuo resigned from an embodied AI company a week ago. Founded less than three years ago, the company has completed Series B financing, with its core product being quadruped robotic dogs. He was in charge of key B2B client sales, and closed several large orders in the second half of last year.

In July and August this year, the situation began to change. The final payment for projects was repeatedly delayed. To obtain an acceptance document, he sometimes had to visit the client's factory four or five times. The responses from clients were mostly similar: the product ran unstably and had not met the acceptance criteria, so the final payment would have to wait.

Meanwhile, the company has continued to cut expenses in the past few months. Last month, the boss said at an internal meeting that without new financing, the company could only survive for a maximum of 8 more months. Zhang Shuo then decided to leave.

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

Not long ago, capital was still pouring into this industry frantically. Statistics from IT Juzi show that in the first half of 2026, the total financing of China's embodied AI sector exceeded 900 billion yuan, and there were more than 20 enterprises with a valuation of over 10 billion yuan. From ontology 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 hype. In October, there were only 33 financing events in China's embodied AI track, with a total amount just exceeding 100 billion yuan. Compared with the average monthly financing of over 150 billion yuan in the first half of the year, it dropped by one third. This figure does not even take into account the low financing scale around the Spring Festival in February and March. If we only look at June, the month with the highest financing amount in the first half of the year, the total monthly financing was close to 250 billion yuan. By comparison, the financing amount in October was more than halved.

Some investors began to flinch at the mention of embodied AI. Startups could not get new financing, and some previously agreed financing also failed to arrive in the account for a long time.

At what exact point did these changes begin to emerge? Some people attributed it to the news of tightened IPO rules that blocked the exit path, some believed it was the sharp drop in Unitree Technology's share price after listing that caused the inversion of valuation between the primary and secondary markets and led to the overall revaluation of the industry, and others thought it was the "fake orders" in the industry that were publicly exposed by insiders.

In any case, the chill has spread across the industry. The special situation of the embodied AI industry is that it was ripened by capital faster than many previous popular tracks. Even before clear application scenarios and business models were formed, and before technical routes and product forms were finalized, a large amount of capital had already placed their bets.

When an industry grows from barbaric expansion to maturity, the reshuffle is often inevitable. Although the financing contraction brings more severe challenges to participants in all links of the industrial chain, it also pushes the industry to shift from capital-driven to technology and business value-driven. As capital returns to rationality, companies that truly have technical barriers, commercialization capabilities and long-term competitiveness will instead have the opportunity to stand out.

01. More Picky Capital and Unfulfilled Orders

Li Xiang is a 2026 graduate. This autumn, he wanted to apply for an algorithm position in the embodied AI field. He heard from seniors and sisters of the previous year that such positions offered good treatment, had large demand, and the competition was not that fierce. But after looking around, he found that leading embodied AI companies released very few job openings.

He also interviewed for algorithm positions at some second-tier embodied AI companies, most of which got no follow-up. After communicating with job seekers of the same year, he gradually realized that the problem could not be simply attributed to higher recruitment requirements. The number of available positions did decrease, and some companies did not seem to have an urgent recruitment plan. For an industry that was regarded as a hotbed of employment opportunities last year, the choices left to him this year were far fewer than expected.

Recruitment is only one aspect of industry changes. The chill is gradually spreading along the job market, startups and investment institutions. The first to be affected are often those companies that have not formed stable revenue and need continuous R&D investment.

The CTO of an embodied data service company told "Dingjiao One" that the company had basically finalized a round of financing in July, but it is still stuck at the intent signing stage. The other party's reason is that the process is still ongoing. Most institutions that missed the company's previous round of financing and were eager to be included in the next round of financing list also gave no response.

Unable to get money from investors, companies can only get it from clients. But for most embodied AI-related companies, commercialization has not yet formed a stable cash flow. Even if orders are obtained, delivery, acceptance and payment collection may continue to consume funds.

Investors are no longer keen on discussing embodied AI, and the number of related projects submitted for review has decreased significantly. A dollar fund investor told "Dingjiao One" that dollar funds usually entered the market earlier than RMB funds. In his view, when investing in ontology in 2024 and investing in the "brain" and data in 2025, the valuations of related companies were already very high. If people did not realize the bubble until the second half of 2026, their reaction would be too slow.

Statistics from IT Juzi show that in the first half of 2026 alone, at least 19 companies in the robotics track covering ontology, data, brain, dexterous hands and other directions have a valuation exceeding 1 billion US dollars. On average, a new unicorn was added to the industry in less than 10 days.

Even some leading enterprises, such as Zibianliang, Zhi Pingfang, Lingxin Qiaoshou and others, crossed the valuation threshold of 200 billion yuan at the end of June. For a project in the portfolio of the aforementioned dollar fund investor, the valuation increased 10 times within half a year by the end of March this year. After the price rose rapidly, new entrants needed to pay higher costs for the same technology and team, while the potential return was further compressed.

This fanatical pursuit of capital made subsequent investors hang back. Many FAs stopped taking on embodied AI projects from the second half of the year, on the grounds that the valuation was too high to sell, or there were no clear scenarios, making it difficult to explain the source of revenue to capital providers. The once widely accepted forward narrative now requires more practical evidence.

Changes in the exit environment have deepened this concern. At the end of August, the tightening of listing rules and higher thresholds for embodied AI companies made the path of exiting through the secondary market harder to assess. A chain reaction immediately occurred in the primary market: some investors began to sell their old shares of embodied AI companies at a discount. The loosening of old share prices not only affects the pricing of new financing, but also makes potential investors slow down their decision-making. The more funds an enterprise needs, the weaker its position in negotiations.

The era of flood-like capital injection is over. The valuation and financing path that used to be supported by industry trends, several demos and impressive academic backgrounds has come to an end.

02. When Did the Turning Point Appear?

The cooling down did not start suddenly one day. In the past six months, cracks first appeared in company operations, and then spread to market pricing, exit expectations and technical judgments.

In May, DeepBlue Robotics (Shanghai) entered the bankruptcy liquidation procedure. Its parent company DeepBlue Technology was once an AI unicorn with a valuation of over 10 billion yuan, supported by institutions including Yunfeng Fund and China International Capital Corporation. Its star background failed to prevent risks from being exposed.

More survival problems emerged later. In August, Yinghe Robotics, which had received investment from institutions including Meituan's investment arm and ByteDance, was ruled by the court to accept its bankruptcy liquidation application. This company also had a high starting point. It was founded in Shenzhen in March 2020, incubated by Infore Group. Only 10 months after its establishment, the company's valuation reached 500 million US dollars, with cumulative financing exceeding 600 million yuan. However, the high financing ultimately failed to translate into sustainable operating capacity.

Later, Wujing Power, which was established only 8 months ago, suspended operations and owed salaries. Among its claimed 110 million yuan of financing, the key industrial fund failed to complete the delivery as scheduled, and its cash flow subsequently deteriorated. These companies encountered different problems, but all exposed similar issues: external funds did not arrive on time, and the companies lacked sufficient buffer.

A broader sentiment shift occurred in mid-August. On the 19th of that month, Unitree Technology went public, and then its share price plummeted sharply, with its market value once halved. As one of the most high-profile companies in the industry, its performance in the secondary market provided a new valuation reference, and the original pricing in the primary market began to be tested by the public market.

At the end of August, another piece of news spread in the primary market circle. The news said that IPOs in the embodied AI industry would be tightened, and loss-making enterprises might face higher thresholds. They not only needed to be in the first echelon of the industry, but also needed to prove that there was a clear expectation of operating improvement in the next three years, and could not rely solely on R&D investment to maintain their business story.

Although the regulatory authorities have not publicly confirmed the above statement, the rumors have affected investment decisions. For industries that rely on IPO exits, changes in expectations often take effect before formal rules.

The climax of the turning point occurred in September, when the controversy over the authenticity of orders in the embodied AI industry was made public. Shao Tianlan, founder, chairman and CEO of Mech-Mind, publicly criticized Galaxy General, and pointed the finger at "gathering-type entrepreneurship". He questioned that some companies used related transactions such as data collection centers and leasing companies to generate false and unsustainable revenue, lacked PMF (Product-Market Fit), but kept releasing big news and sought to go public only two or three years after their establishment. In response, Galaxy General later issued a statement, emphasizing that the company adheres to real R&D investment and application scenario implementation, and will not participate in unnecessary verbal wars.

Under the conventional model, data collection centers purchase robots, collect data in a simulated production environment, and then sell the data to embodied AI companies or large model companies. Both equipment procurement and data services can constitute normal transactions, and the controversy focuses on the capital flow and related party relationships of some transactions.

The trick Shao Tianlan criticized is that an embodied AI company sets up a data collection company in a joint venture with the local government, and then the data collection company purchases the robot ontology from the enterprise. The funds are mainly borne by the local government, and the embodied AI enterprise does not contribute or only bears a small part. The data generated by the data collection center is then repurchased by the embodied AI enterprise. After the cycle is completed, the embodied AI enterprise obtains orders and revenue, and the relevant funds flow back to the data collection center in the name of data procurement.

The controversy continued to ferment, and whether the orders on the books can represent real demand has become an unavoidable problem in the industry.

The successive "bad news" does not end there. In early September, after OpenAI released GPT-6 Astra, a large number of tests using this model to control robotic arms appeared on social media. Some practitioners found that general multimodal foundation models already have spatial understanding capabilities, and the success rate of some tasks even exceeds that of some embodied models.

After the test videos spread, statements such as "general large models will eat up embodied models sooner or later" began to circulate. Wang Xingxing, founder of Unitree Technology, even said that after GPT-7 is released next year, it may directly solve all the problems of embodied AI.

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

Disappearing companies, halved valuations, raised thresholds, punctured window paper and "barbarians breaking in through the door" all superimposed to rapidly cool down the once hot embodied AI industry, making it difficult for the market to continue to use the optimistic assumptions from the first half of the year.

03. Conditions for Staying at the Table

The cooling of the industry is not uniform, and not all companies feel the same chill. While the total financing amount is falling, capital is further concentrating on leading companies. This trend has emerged in the first half of the year and continued in the second half.

Statistics from iiMedia Research show that in the first half of 2026, the flow of industry financing presented a progressive logic of "brain > body > scenario". The financing amount of the embodied AI brain track reached 595.44 billion yuan, with a high average single transaction amount, reflecting capital's preference for software capabilities such as decision-making algorithms and multimodal perception at that time.

Ontology manufacturing companies and companies with landing scenarios also obtained high financing amounts. Capital did not bet evenly on the entire industry, and more funds flowed to links that were considered more likely to form barriers.

Entering the second half of the year, this selectivity has become even stronger.

In July, Starbot Era and ExDynamics respectively completed financing of more than 1 billion yuan. In August, Paccini and XPeng Robotics received capital support. In September, Digua Robotics and Sudo Technology also completed financing of more than 1 billion yuan. They are still mostly within the popular directions of brain, ontology and scenarios.

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

After entering the reshuffle stage, the conditions for staying at the poker table have gradually become clear, with capital, scenarios and scarcity becoming three unavoidable variables.

The most urgent one is capital. Most embodied AI companies are still in a loss-making state and lack mature self-sustaining capabilities, while R&D, production and delivery require continuous investment. After the financing environment changes, cash reserves directly determine how much time a company has to complete product verification, and also determine whether it can wait for the next round of financing.

Among the companies that can survive, some have obtained cash through sales and services, some can still get support from investors, and some rely on the reserves left from previous financing to extend their survival time by cutting expenses. For the last type of companies, careful budgeting will become a business method that they must adapt to.

This may also force some companies to rethink their past business strategies. When the industry was hot, the faster the financing speed, the easier it was for enterprises to expand their R&D teams, product lines and business scale at the same time. However, after the financing environment changes, the parallel advancement of multiple business lines may turn from a growth advantage into a financial burden.

A company that has raised billions of yuan of financing and simultaneously lays out ontology, models, data collection and multiple application scenarios may need to bear high R&D and operating costs. On the contrary, an enterprise with small financing scale, lean team and focus on a specific niche product may have a longer survival cycle if it can control cash consumption.

Therefore, a large financing scale does not necessarily mean that the company can live longer. Being able to shrink the front line in time and concentrate limited