Written by | Wen Yehao Edited by | Wu Xianzhi
In the second half of the year, the embodied AI sector has experienced continuous turbulence.
WAIC and WRC were held successively, with robots gathering to showcase their operations and the audience bustling; Unitree Robotics went public, with its market capitalization once surging to CNY 444.9 billion on the first day, setting the entire industry ablaze with enthusiasm.
However, as players "exposed each other's skeletons in the closet," the narrative quietly shifted. By mid-October, as a major anchor in embodied AI, Unitree's market capitalization had fallen back to around CNY 180 billion, and the overall discourse on embodied AI became more restrained and rational.
Against this backdrop, AgiBot, another leading player in the embodied AI sector, was reported to have launched a new funding round of CNY 1 billion to CNY 1.5 billion, with a valuation of approximately CNY 50 billion.
Compared to the funding amount, the bigger highlight lies in the subsequent ripples after the stone is thrown—AgiBot, eager to go public, has had its Hong Kong listing postponed, making it almost hopeless to list within this year.
According to IDC's recent statistical caliber, AgiBot has already secured the first place in global humanoid robot shipments in the first half of this year. But at such a juncture, what AgiBot has to face is perhaps far more than just the question of shipment volume.
The True Value of Being "Number One"
The prosperity in statistical calibers is a commonality for all hardware industries when they develop to a certain stage.
The automotive industry has polished this set of expressions to the finest detail: wholesale volume, terminal retail volume, insured volume, and delivery volume, with an interest chain tied behind each figure. Coupled with segmented rankings—number one in mid-to-large SUVs, number one in independent brands, number one in hybrids—almost every player has their own set of narratives.
Smartphone manufacturers are equally adept at this: global shipments, online market share, market share by price segment, activation volume—in short, as long as the modifiers are precise enough, every company can be number one in its own press release.
In this regard, embodied AI players are still in their apprenticeship. Their skill in writing modifiers is far less rich than their predecessors, but they have gradually grasped the essence.
For a period previously, in public data, the number one on the list including wheeled dual-arm robots was AgiBot; if only pure bipedal robots were counted, the number one was Unitree; ranked by revenue, the number one was UBTECH. Further down, there are various titles such as number one in mass production rollout, number one in orders on hand, and number one in cumulative sales.
In response, there has even been some friction among the leading players. Earlier this year, Unitree, which was in the pre-IPO tutoring period, specifically issued a clarification report, emphasizing that its figures were shipment volumes, not order volumes, and did not include dual-arm wheeled robots.
At that time, the number of units was important largely because it served multiple functions. After all, in an industry where real demand has not yet emerged, data such as mass production and shipment volumes are not only a major basis for financing pricing but also a universal bargaining chip for quantification and comparison among all parties in the industrial chain.
In the first half of this year, the divergence temporarily narrowed. Under the SAG caliber, AgiBot's shipment volume was about 8,400 units, with a 44% market share; under the Counterpoint caliber, it was about 9,700 units; and as of September, AgiBot's cumulative rollout had also surpassed the milestone of 20,000 units.
Although the statistical calibers vary, the conclusions tend to be consistent—no matter which ranking is used, the number one is AgiBot.
AgiBot got the "championship" as wished. However, it is a pity that chips are ultimately not currency. When the narrative evolves, the value it carries will also change accordingly.
Since the beginning of this year, as the industry's focus has shifted from the body to the brain, the embodied narrative has gradually moved from dull shipment data to more imaginative rankings. And when the outside world has doubts about embodied value and business models, commercialization capabilities and ROI are brought to the forefront.
Therefore, whether it is AgiBot or other players, selling goods is selling goods; the value depends more on the flow direction than on traffic.
At this stage, the buyers of embodied AI can be roughly divided into three categories.
The first category, continuing the aforementioned brain narrative, is represented by data collection centers, paying for the "future."
According to media reports, the typical structure of this path is to sell the body to a data collection platform, and then repurchase the data through affiliated entities. Data collection should originally be counted as the manufacturer's R&D costs, but after two transactions, it is transformed into book revenue. "Caijing" previously reported that in two projects in Hubei and Zhuhai, there were equity associations between the purchasers and AgiBot.
Of course, these contracts are real, and the deliveries are also real. What the market worries about is whether the same buyer will buy again next year. This is a question not about authenticity, but about revenue quality.
The second category, paying for the "present," mainly consists of lessors and the commercial performance market.
This was once the earliest and most enviable landing method for embodied AI to run through. After the 2025 Spring Festival Gala, the story of robots with "a daily rent of CNY 10,000 and a payback period of a few weeks" spread throughout the industrial chain. When the market was hottest, it was hard to get a single machine. Lessors and commercial performance teams flocked to enter the market, and complete machine manufacturers also took the opportunity to ramp up production capacity.
When everyone rushes into the same business at the same time, supply and demand will quickly reverse. With capacity expansion and platforms flooding in, the daily rental price of robots has been going down all the way.
Staff from Qingtian Zu told Photon Star that the daily rent for Lingxi X2 was CNY 5,499 around the Spring Festival, and now it is about CNY 3,000; the Youth Edition with lower configuration was CNY 3,599 during the Spring Festival peak season, and now it still costs CNY 2,800. The rent merely falling back to CNY 2,800 seems to be the floor price that cannot drop any further.
Although the traffic-attracting prices on some platforms have dropped to several hundred CNY a day, to get such prices, most require monthly rentals, and only deliver the equipment, completely extracting the "human" element from the service.
This means that the most expensive part of the quotation is the engineers traveling with it; their daily salary, transportation, accommodation, and meals—none of these can decrease with the scaling of hardware.
A more fundamental issue is that rentals and commercial performances consume spectacles, and a spectacle is a resource that will be quickly depleted.
In the years when 3D movies were most popular, audiences were willing to pay an extra ticket price for those glasses, but left one after another once the novelty wore off; VR experience halls, which once blossomed everywhere in shopping malls, also closed down in large numbers after customers felt "playing once is enough."
Nowadays, videos of robots dancing and doing somersaults have flooded short video platforms, and onlookers are increasing. How long this business can remain lively remains to be seen.
The third category, paying for "efficiency," mainly consists of industrial capital and factories.
This is the closest to commercialization expectations among the three categories. The accounting logic is similar to AI Coding, which was the first to run through in the past two years. Whether customers are willing to pay depends on whether the accounts can be balanced. Only when the saved money is visible and calculable does the procurement have a justified reason.
The shareholder lists of BYD, SAIC, BAIC, JD.com, and LG have indeed exchanged a batch of testable deployments for AgiBot. It should be noted that, compared with AI, the landing of robots is much clumsier, involving production line transformation, on-site debugging, and months of running-in behind the scenes. Earning stable money from robots may still require time for the product and scenarios to mature together.
The Inescapable Accounts
For a company eager to go public, no one can escape the scale of financial data.
Take Unitree, which handed in its paper first, for example. In the generally loss-making embodied AI sector, it is one of the few companies that can already present a profitable answer. In 2025, Unitree's revenue was nearly CNY 1.7 billion, its net profit deducted from non-recurring gains and losses was about CNY 600 million, its gross profit margin was about 60%, and its net cash flow from operating activities was CNY 670 million.
But behind this rare profitability, in addition to humanoid robots, the more mature quadruped robot dogs also hold up half the sky. Quadruped products are more affordable, have stable buyers in overseas consumer markets, and mature supply chains and economies of scale are sufficient to compress costs and squeeze out profits.
Compared with the already profitable Unitree, AgiBot's posture appears more subtle.
Data shows that AgiBot's revenue increased from about CNY 300,000 in 2023 to CNY 1.05 billion in 2025, and in the first quarter of 2026, a single quarter exceeded CNY 1 billion, with an even faster growth rate; however, from 2023 to 2025, its cumulative loss was about CNY 280 million.
Deng Taihua, Chairman and CEO of AgiBot, has said that they are not in a hurry to pursue positive net profit, and scale is the king.
This statement holds true in the capacity expansion stage. Entering the landing stage, the scale must be an effective scale. The loss itself is not the problem; the loss without a convergence path is.
However, the convergence of the path will not happen out of thin air; it is exchanged for time. And time must be bought with cash.
Just at this moment when money is needed most, it seems that money is starting to become scarce.
On the surface, financing in the first half of 2026 is still setting records, nearly five times that of the same period last year. However, according to data from IT Juzi, stepping into October, the number and amount of financing have declined compared to the previous months. In addition, some investors frankly told the media that while institutions have not explicitly said they will not invest, the investment decision results have basically stopped.
Short-term fluctuations are understandable, but if the atmosphere continues to decline, the logic of this gameplay of exchanging losses for scale may need to be re-examined.
On the other hand, as mentioned earlier, the destination of money has changed since the beginning of this year, with the majority of funds flowing into the R&D of embodied AI models.
On the question of the brain, AgiBot started earlier and bet heavier.
In AgiBot's definition, it is first and foremost an embodied AI foundation model company. Since the model needs a carrier, it naturally proceeded to do hardware; and more than three-quarters of R&D manpower and expenses are pressed on the large and small brains. Unitree is the opposite. In the prospectus at the beginning of the year, the embodied large model was still an item to be broken through, and raising funds from the listing was largely to make up for this lesson.
However, an earlier and heavier bet does not equal a correct bet. In September, OpenAI's GPT-6 Astra has further proved that in the embodied AI track, there seems to be a huge gray rhino lurking, whose direction has not yet been clearly seen.
Paying for an endgame that has not yet arrived inherently carries risks, and how far away this endgame is, no one can give an accurate answer now.
In addition to this, whether it is Mech-Mind Robotics' breaking of the issue price, Huandong Technology's withdrawal of the IPO application, or the ripples stirred by reports of mutual guarantees between upstream and downstream and order inflation, all are continuously increasing the difficulty for embodied AI companies to go public.
According to media reports, a higher-dimensional force is guiding the listing process of embodied AI players through "informal windows," requiring prospective listing enterprises to prove sustainable revenue capabilities, loss narrowing paths, or sufficiently outstanding technological innovations.
Multiple pieces of information point to the same change, which is that the market is systematically squeezing the water out of the embodied AI track. Whether in the primary or secondary market, valuations, financing, and listing narratives are all undergoing increasingly strict tests.
When the window narrows and the threshold is raised, every company may have to re-evaluate itself: whether to squeeze in before the door closes, or to first tell its own story clearly.
IPO Can Wait, but Valuation Cannot Collapse
When the window narrows, the instinctive reaction of most players is to rush ahead, for fear of being one step late and locked out of the door.
Especially for many mid-tier players, the narrowing of financing and listing channels makes them the most impatient.
After all, the fiercely competitive embodied AI track has been burning money from beginning to end. Talents need to be grabbed, computing power needs to be stacked, data needs to be collected, and production lines need to be built. And the channels for players to make money are nothing more than two: financing and commercialization revenue, the latter of which is far from running through.
For them, whether they can go public is a matter of life and death. And at this juncture, AgiBot, instead, has actively slowed down its listing pace with a new round of financing.
According to the rules of the HKEX, new share financing completed within 28 days before the submission of the application or after the submission until the listing date must have its first day of listing postponed to 120 days after the completion of the last financing. This means that if the financing is settled within the year, AgiBot's listing will have to wait until at least the first quarter of 2027.
As for the reasons, there may be multiple factors.
On the one hand, at a time when stock prices in the embodied AI track are falling and cooling down, rushing hard might only get a relatively low starting point. In contrast, if there is still capital willing to come in at a price higher than the target range for Hong Kong stock cornerstones, it is equivalent to anchoring the issue price in the primary market first—the listing can wait, but the valuation cannot be collapsed by emotions first.
On the other hand, as a leading player in the track, AgiBot is not as short of money as mid-tier players, and indeed has the room to "wait."
In addition, in response to many recent industry controversies, this period of time is also enough for AgiBot to clarify its own revenue structure and model, facing the secondary market with cleaner statements and more solid orders.
These reasons may all exist, or they may just be speculations from a bystander's perspective.
Ultimately, AgiBot's biggest advantage at this stage may be its ability to sell robots. And next, what it still has to prove is that the sold robots can in turn make it easier to sell the next robot.
The scales of shipments, accounts, and commercialization have always been there. The further time goes, the more accurate the scales become. And before that, AgiBot needs to be prepared to be weighed.
WeChat ID | TMTweb Official Account | Photon Star