Embodied AI company IPOs cool down: Zhiyuan may abandon listing this year, lawyers break down equity and order risks

“I’m still waiting to cash out through an IPO. The longer it drags on, the smaller I feel the embodied-AI valuation gets.”

The person saying this is an employee of an embodied robotics company that is rumored to be secretly filing through Hong Kong’s 18C specialist technology listing channel.

Holding substantial options and hoping the company goes public soon, some engineers with million-yuan salaries are now busy with R&D and testing while also finding time to track capital-market developments, afraid that the longer they wait, the less their options will be worth.

This anxiety is spreading across the embodied intelligence industry.

Almost everyone shares an unspoken perception: the IPO window for robotics companies has already passed.

After Unitree listed on August 19, the pace of robotics companies that had been racing toward the capital markets began to change.

△As of October 23, 2026, Unitree Technology’s daily K-line chart since listing

English text is being prepared.

Recently, multiple media outlets reported that AgiBot is advancing a new funding round of about 1 billion to 1.5 billion yuan, at a valuation of about 50 billion yuan.

AgiBot has not yet publicly confirmed this. If true, once this funding round closes, unless it obtains a special exemption from HKEX, AgiBot will need to wait at least 120 days after the financing is completed before it can list.

By a rough count, the Hong Kong IPO window for this year is basically closed. (doge)

On July 24 this year, AgiBot confirmed externally that it had begun preparations for a Hong Kong listing. At the time, investors expected a listing as early as August, with a target valuation range of HK$40 billion to HK$50 billion.

During that period, embodied-AI companies such as LimX Dynamics, EngineAI, X Square Robot, Galaxea, and Galbot all circulated news of confidential filings and similar moves.

English text is being prepared.

This is especially true for companies whose commercialization capability still needs to be verified. Their subsequent operating capability (such as R&D and sales) and the resulting performance data will become important observational bases during the listing process and for post-listing share-price assessment.

“(Regulators) will most likely focus on company quality. On top of routine listing issues, they will further focus on factors such as whether embodied intelligence companies have verifiable technological leadership and whether their products have commercial deployment potential. If these are difficult to prove through objective evidence, the listing process may slow.”

Huang Lichong, president of Huisheng International Capital, has served as a listing sponsor for HKEX Main Board and GEM listings.

He also said that when embodied intelligence companies list, the hardest thing to explain to the capital markets is how much of the value created by technology will ultimately remain on the company’s own books.

Better model capabilities and higher customer efficiency do not mean the company’s profits will also rise.

“Are customers willing to pay more? Does the company have pricing power? After deducting computing power, data processing, implementation, and manual review, how much is left? All of this has to be calculated.”

Going public is something that places great emphasis on “the right timing, the right place, and the right people.”

Companies may not be able to decide when the window opens. But it would be a pity if, when IPO enthusiasm picks up again, they are shut out by old problems such as equity, finance, and compliance.

What exactly does a robotics company need to accumulate in advance to seize the listing opportunity?

English text is being prepared.

When the IPO market cools, who is most anxious?

Once the pace of robotics company listings slows, the first to grow restless are often the investors behind them.

Behind the funds stand LPs (limited partners), and when the time comes, they have to consider how to exit.

Over the past two years, financing enthusiasm for embodied intelligence has continued to rise. Many robotics companies had already secured several funding rounds before finding stable application scenarios.

Valuations kept climbing, and shareholder lists grew longer and longer. Expectations for listing timelines accumulated accordingly.

But every robotics company has its own timetable for R&D and commercialization.

From model training and scenario validation to bulk customer procurement, it may not align with the funds’ exit cycles.

Li Ming, a securities lawyer with more than seven years of experience, often reminds founders to bring in investors cautiously and not to take just any money.

The company had originally been developing well, completed financing at a relatively high valuation, and then began preparing for an IPO.

Unexpectedly, external industry conditions changed, and the company’s performance fell short of expectations and began to decline, so the listing plan had to be shelved.

To make matters worse, because the investment agreement at the time made the listing timetable one of the valuation adjustment (bet-on) conditions, some investors immediately demanded repurchases.

But the financing funds had long been invested in R&D and business expansion, and the company could not come up with a large amount of cash in a short time.

Under the agreement, the founder could also face corresponding repurchase obligations.

A financing round obtained a few years earlier, when the listing progress slowed and the bet-on conditions were not met, instead became a heavy burden.

Moreover, the higher the valuation, the more trouble there is.

When one technology company arranged equity incentives for employees, it used a relatively high valuation. Later, when employees left and asked to exit their equity, according to the agreement, the founder had to take it over.

As a result, the higher the valuation, the greater the repurchase cost.

The equity incentives originally meant to attract talent instead became a financial pressure on the founder.

In addition, the fair value determination and tax treatment involved in equity incentives are also considerable work for IPO lawyers.

Similar cases, though they come from other technology companies, are not far off for embodied robotics companies currently undergoing high-valuation financing.

What's more, valuations in the primary market still have to be re-tested once they reach the secondary market.

Huang Lichong is not very much in favor of forcing the issue price to a level lacking operational support just to maintain the previous round's financing valuation.

A robotics company obtaining a post-money valuation of 50 billion yuan does not mean that, after listing, secondary-market investors will be willing to take over at an equivalent market value.

Especially for robotics companies still in the early stages of commercialization, they must explain clearly:

Is the current valuation paying for already-validated batch business, or is it counting in advance the sales, production capacity and service revenue of the next few years?

If revenue forecasts rely on pilots that have not yet converted, investors will naturally discount the timing of future revenue realization and the funding needs.

Therefore, embodied companies must also prepare in advance for valuations falling short of expectations, reduced offering size, or delayed listing.

Investors want to exit, founders worry about repurchases, and employees holding options are also hoping for a listing to cash out.

At Beijing's Yonghe Temple, Zhoushan's Putuo Island, and the Nanyue Grand Temple that bosses in central China love to visit, the thresholds are almost worn through.

Some bosses make a special trip to pray before important review milestones, seeking good luck.

Some even calculate in advance the date for submitting the prospectus.

Even more exaggerated, some take the sponsor's birth date and eight characters to calculate whether the two sides are compatible.

When talking about these unspoken understandings in the industry, Li Ming sighed: "Believing in fate really knows no age."

But choosing an auspicious day does not mean everything will go smoothly.

Some companies, only before filing, discover that early equity and funding transactions cannot be explained clearly; some submit applications but are unable to advance for a long time.

Others have already passed review but ultimately still fail to issue as originally planned.

Where do the problems lie?

A senior securities lawyer, drawing on years of IPO project experience, summarized several places most easily overlooked by embodied intelligence companies. Embodied intelligence companies preparing to list may wish to check against them in advance.

Trouble may start with the very first financing round.

Zhang Tao has dealt with a technology company that originally planned to sprint for an A-share listing but later turned to Hong Kong stocks.

The problem lay with the shareholders.

A shareholder introduced by the company in its early years involved an overseas trust arrangement. By the listing review stage, regulators paid special attention to the reasons for and reasonableness of the trust shareholding, and the company needed to conduct further verification based on feedback inquiries, which was quite complicated to handle.

According to the rules, trust shareholding itself is not a substantive obstacle to listing; the trouble is that, under a special regulatory background, it needs to be explained clearly to the review department and meet the corresponding review requirements.

When entrepreneurs raise funds, what they care about most is often the amount and the valuation. By the time of the IPO, lawyers still have to ask one more question: who exactly invested this money? Could the nature of the shareholder itself create unnecessary review obstacles for the listing?

In addition, for people like Li Ming, there are three other biggest headaches: nominee shareholding, capital contribution defects, and shareholder penetration.

For example, in the early days of a startup, a shareholder held shares on behalf of a friend. At the time, everyone was on good terms and may have thought it was no big deal.

A few years later, as the company prepares to list, lawyers have to re-verify who the shares actually belong to, whether an agreement was signed that year, and from which account the capital contribution was transferred.

Some shareholders' capital contributions are non-compliant, or the source of funds cannot be explained clearly.

Lawyers can only dig out bank statements from many years ago and check them one by one.

Li Ming has handled several projects where the company was originally operating well, but because of historical equity contribution and source-of-funds issues, it ultimately could not file, or had to withdraw midway.

"Once bank statements exist, they cannot be erased."

Now look at shareholder penetration.

Even if the investor on the surface is only one institution, lawyers may still need to continue verifying the actual beneficial owner behind it.

For example, whether there is shareholding by former personnel of the securities regulatory system, and whether the relevant personnel's work experience, timing and method of shareholding comply with the rules.

It is not that there is a problem simply because relevant personnel hold shares, but not a single matter that needs to be investigated can be omitted.

Sometimes, the biggest pitfall in preparing for a listing is buried when the company is just founded.

For example, in those years, in order to obtain financing, founders signed investment agreements in which investors often sought special arrangements such as repurchase rights and anti-dilution rights.

When the money arrived, a listing still seemed far away, and the two sides might not have agonized much over how these clauses would be handled later.

By the time of the IPO, the trouble gradually emerges.

Some special rights need to be cleaned up before filing, and some, even if agreed to terminate, come with restoration conditions. Once the listing fails, the investor's relevant rights may become effective again.

Those investors mentioned earlier who demanded repurchases because of the IPO delay are an example of how financing clauses can have subsequent effects.

Unitree Technology's prospectus disclosed that in the June 2025 financing agreement, investors enjoyed special rights such as repurchase, preemptive subscription and anti-dilution.

In September of the same year, the company signed a termination agreement with the relevant shareholders.

△Screenshot of Unitree Technology's listing legal opinion, public information

This agreement provides that the company-level repurchase arrangement is "completely terminated, void from the outset, and may not be restored."

In other words, even if the listing fails, the company will no longer be required to spend money repurchasing equity.

Essentially, this prevents the funds of the listing entity and future public shareholders from being used to cash out early investors' bets, protecting the interests of small and medium-sized investors in the secondary market.

There is another type of trouble that requires solving with employees' former employers.

Li Ming encountered a company that applied for an invention patent, and the inventor was an engineer who had just jumped ship from another company.

So the question arises: was this technology developed during the inventor's time at the previous employer, or was it independently developed after joining the new company?

If it involves a service invention, who should hold the rights?

To confirm ownership, Li Ming had no choice but to contact the engineer's former employers.

"Going one by one to his former employers to confirm. Think about the workload—isn't it painful?"

Embodied AI companies need to pay even more attention to similar situations.

Talent flows frequently in the robotics industry today, and many core R&D staff come from universities, research institutions, and other tech companies.

Algorithms, motion control, training data, and models may all involve R&D staff's long-term technical accumulation as well as collaboration with universities and research institutions.

When an engineer changes jobs, their R&D experience naturally does not disappear.

But exactly who owns the specific technical achievements still requires sufficiently clear evidence at the listing stage.

You can pick a date to ring the bell, but these old accounts cannot be avoided.

Huang Lichong also reminded robotics companies preparing to list in Hong Kong that they need to make commercial validation and business management more solid.

Orders, shipments, acceptance, revenue, after-sales obligations, and collections must match up; R&D staff, project expenditures, intellectual property, and data usage rights must have supporting documentation.

In addition, financing plans must follow business progress.

How much more money is needed in the next stage? If the listing is delayed, how long can the company's cash last? If the valuation falls short of expectations, can the founders and shareholders accept it?

These questions must all be thought through in advance.

"Especially for companies considering listing through Hong Kong's 18C regime, the relevant R&D investment requirements span multiple years, and historical records need to be accumulated continuously—they cannot be patched together at the last minute before filing," Huang Lichong told QuantumBit.

Plans may be meticulous, but unforeseen events are hard to guard against.

A common scenario is that, at the final moment, a company gets reported, or suddenly has to deal with supervisory inspections or on-site checks initiated by regulators.

Zhang Tao even described reports and on-site inspections as "a sword hanging over every listing applicant."

"This is a necessary measure to ensure filing quality; no one can guarantee they won't encounter it."

QuantumBit has learned that Unitree and Yuejiang both encountered reports involving patents, share ownership, and other issues during their IPO processes.

Li Ming once served a tech company that faced more than a dozen rounds of reports before listing, and he and the project team responded more than a dozen times as a result.

"A report itself does not mean the company has problems, nor does it necessarily affect the listing," Li Ming told us. What matters is whether the report's content is true and whether it involves major issues such as equity or core intellectual property.

For founders, what they can do is not harbor any wishful thinking, eliminate hidden risks in advance as much as possible, and do the filing work solidly.

Even when equity, financial, and compliance issues are all resolved, it does not mean a robotics company has gathered enough confidence for listing.

Having watched the secondary market for so many years, Huang Lichong knows clearly what investors there most want to see:

Customers have already paid, the product is already in use, and after using it they are willing to continue purchasing.

Because old accounts can be sorted out slowly, but orders cannot be conjured out of thin air.

Some embodied robotics company bosses constantly emphasize "storytelling" during IPO preparation.

"Frankly, I personally don't like that word." But Zhang Tao also understands that this track is too crowded—besides peer startups, large companies such as Xiaomi, XPeng, and Tesla are also entering.

Competitors are each more capable than the last, and everyone wants to prove they are somewhat different.

Of course, the pressure on embodied AI companies does not come only from peers.

According to Wang Peng, an algorithm engineer at an embodied robotics company, the difficulty embodied companies face in going public is also driven by the impact of large-model companies.

"Last year, when Gemini Robotics demonstrated a robot's ability to grasp and place different objects, it was still a much-watched technology demo in the embodied AI field. Now, you can achieve it by hooking up GPT-6 Astra."

Wang Xingxing, chairman of Unitree Robotics, has also felt this impact. After the listing, he publicly discussed the influence of GPT-6 Astra on embodied AI.

"Almost everyone in the world, including myself, still does not attach enough importance to and have enough faith in AI." On October 22, at the 2026 World Physical AI Conference, he put it even more strongly.

Wang Xingxing believes that physical AI has already shown signs of being broken through. And this node will arrive within 1-2 years at the fastest, or 2-3 years at the slowest.

On one side, general large models are becoming increasingly capable of generalization and beginning to penetrate the embodied AI field;

On the other side, traditional robotics companies that already have customers and scenarios are also starting to use large models to improve their intelligence levels and extend into the market of embodied AI companies.

"Those doing embodied pretraining these days probably all feel some kind of existential crisis to varying degrees." What leaves Wang Peng rather helpless is that domestic embodied AI has entered a technology plateau period, with no new paradigm-level breakthrough in sight in the short term.

However, before this new round of impact brought by general large models, some robotics companies had already delivered commercialization results.

Unitree's 2025 revenue was 1.699 billion yuan, with net profit attributable to the parent after non-recurring items of about 590 million yuan (278 million yuan before non-recurring items), and it sold 5,215 humanoid robots in one year.

CloudMinds is smaller in scale, but its revenue also grew from 50.1126 million yuan in 2023 to 337 million yuan in 2025, and it achieved profitability that year.

Geek+, which entered the warehousing and logistics scenario earlier, had cumulatively delivered about 56,000 AMRs (autonomous mobile robots) by the end of 2024, with revenue exceeding 2.4 billion yuan in 2024.

But for the capital market, how many units are sold is only the first step in judging commercialization capability.

"What customers buy is production capability, not a demonstration of a motion," Huang Lichong said.

Selling one robot is one thing, but is the buyer a university laboratory or a factory that genuinely needs robots on its production line?

The purchasing purpose, frequency of use, and ongoing demand behind it may be completely different.

When it comes to PoC (proof of concept), small-batch procurement, and batch replication, what Huang Lichong values more in embodied AI companies is repeat purchases after actual use.

Only when customers are willing to buy a second or third unit, or even extend robots from one production line to more factories, can demand be further proven sustainable.

But repeat purchases alone are not enough. Expanding from one pilot site to other sites requires that cost, reliability, and payment collection do not deteriorate noticeably.

Otherwise, the more orders there are, the harder the business becomes.

Right now, the industry controversy surrounding embodied intelligence plays out in exactly these areas.

A financial professional familiar with the embodied intelligence industry suggests that orders need to be broken down. Framework agreements, cancellable orders, purchases with acceptance conditions, and sales that have already been fulfilled carry different weight.

Accounting revenue must be judged based on contract performance and transfer of control; it cannot all be recognized as revenue simply because a contract was signed, a deposit was received, or equipment was shipped out.

Recently, multiple media outlets have reported that behind some robotics companies' orders, there are situations such as investors facilitating procurement and local industrial funds promoting cooperation.

Some companies form joint ventures with local capital to generate orders for robot equipment procurement and data services.

Some robotics companies even hope to create order flow by mutually purchasing hardware and data with data collection vendors.

According to Caijing, some data collection vendors have participated in such transactions, where both parties signed contracts but did not complete physical delivery.

As a result, there appears to be business on the books, but the actual delivery situation needs to be questioned.

Were the robots actually transported to the customer site? Was the data truly provided? And how do customers plan to use them after buying them?

These questions all relate to the authenticity of the transactions.

u1s1, industrial capital orders in the robotics industry cannot be lumped together either.

Some local industrial capital invests in robotics companies precisely because it hopes to bring in projects, cultivate the industrial chain, and promote local scenario applications.

It is also a common form of industrial synergy for investment institutions to help introduce customers to portfolio companies.

Investors facilitating procurement does not automatically equal a related-party transaction. Even if a related-party transaction exists, it does not mean the company will necessarily be unable to go public.

What truly needs to be explained clearly is whether the transaction has genuine commercial demand, whether pricing is fair, whether the equipment is delivered, and whether the relevant interest relationships are appropriately disclosed.

For embodied robotics companies, the more important answer is whether future orders can continue without investment relationships and local support.

Huang Lichong particularly cares about the revenue quality of technology companies.

He has found that many technical teams are very clear about what tasks robots can accomplish, but easily treat signing a contract, shipping equipment, passing project acceptance, and the customer paying as the same thing.

In reality, performance progress, transfer of control, payment terms, returns, or price concessions can all change judgments about revenue and cash collection.

For example, a project that appears to have high gross margin cannot be used to talk about growth based solely on the contract amount if it still requires long-term free maintenance, a large number of on-site personnel, or if payment collection depends on subsequent conditions.

Conversely, cash received must also be examined by its nature; customer prepayments cannot be treated as revenue already earned.

Investors can accept that you explain difficulties clearly, but it is hard for them to accept the same business appearing in three different ways in fundraising materials, financial statements, and bank statements.

Of course, for companies that have not yet achieved scale revenue but meet the corresponding listing conditions, one cannot simply judge them by revenue alone.

In 2023, HKEX introduced the Chapter 18C listing regime for specialist technology companies, providing a listing path for eligible pre-commercial technology companies.

The A-share STAR Market has also set different listing standards for eligible hard-tech companies.

The system provides companies with flexible choices. But no matter which path is taken, customers still have to be found one by one.

As for the commercialization ahead, Wang Peng is not that pessimistic.

"While there are no paradigm breakthroughs, domestic embodied AI partners should first focus on implementation, gather more scenario data, and under the impact of large models, be able to 'put up a stubborn resistance' for a while."

There is a bit of teasing in his words, but they are not without reason.

Wang Peng said bluntly that the Chinese market is large, and as long as robots entering the market are willing to "do the dirty work" (meaning in one scenario, frantically piling up data and some strange tricks until the success rate reaches 99.9999%), "supporting seven to eight embodied intelligence companies is no problem."

At a quadruped robot company Zhang Tao came into contact with, the founder is busy rushing around for product implementation.

The company is asking investment institutions to help connect with tech companies and research institutes on one hand, while actively reaching out to customers such as property management and hotels on the other, trying to sell robot dogs into security and patrol scenarios.

The company's sales team is also attending exhibitions and visiting potential customers everywhere. "Wherever customers go to exhibit, they follow."

One more thing

Regarding the shift in the wind for embodied intelligence IPOs, there are also voices saying the trigger may be related to one person:

Shao Tianlan, founder of Mech-Mind Robotics.

In September this year, he publicly questioned the related-party transactions and commercialization models of some companies in the embodied intelligence industry, bringing a topic previously kept quiet in the industry into the open.

Some industry insiders believe this controversy may also have made regulators pay more attention to the order structure and commercialization quality of embodied intelligence companies.

But this is only speculation in the market. On the path to listing, much can be planned in advance, and even more cannot be accurately predicted.

This article is from the WeChat public account "QuantumBit" (ID: QbitAI), author: Focus on Frontier Technology, published by 36Kr with authorization.