Amiao Robotics' team shrinks to just over 70 people, as the hot-and-cold divergence in the embodied intelligence industry intensifies
Amiao Robotics' former CTO, COO, and product head have all departed, and the team has shrunk from a peak of about 130 people to just over 70, with a new round of personnel optimization still underway. This industrial embodied intelligence company, founded by Liu Fang, former head of mass-production smart driving at Xiaomi Auto, and led by Anker Innovations, has raised nearly 200 million yuan across three rounds, but lead investor Anker Innovations did not add further investment in the first half of 2026. The article also lists a number of robotics companies that have entered bankruptcy liquidation or disbanded, and points out that a gap exists between the industry's signed contract amounts and delivery and acceptance.
Why read this
AI score 58
Starting from Amiao Robotics' contraction, this article sorts through the layered reality in which financing heat and a wave of bankruptcies coexist in embodied intelligence, and points out the key problem of the gap between order figures and actual delivery.
Topics
Body · AI translation
Recently, media reports said that Amiao Robotics' former CTO, COO, and head of product have all departed, with the team shrinking from a peak of about 130 people to just over 70, and a new round of personnel optimization still underway. This industrial embodied intelligence company, founded by Liu Fang, former head of mass production for smart driving at Xiaomi Auto, and led by Anker Innovations, was once seen as the new face most likely to break out in 2026. Its contraction reflects the reality of the robotics industry, where some are thriving while others are freezing.
Since the start of this year, Longhui Medical, Suzhou Houzhi, Chuangze Intelligent, Dongguan Lesheng, Hongda Xingye Robotics, and Yinghe Robotics have successively been accepted by courts for bankruptcy liquidation; Wujing Dynamics has suspended operations over unpaid wages; and Yixing Robotics has dissolved. From surgical robots to warehousing and logistics, from established unicorns to companies founded just half a year ago, exits have occurred one after another. The problem is no longer with any single company; an industry propped up by financing narratives is being forced to settle its accounts.
The crisis arrives
Jiemian News reported that Anker Innovations added an equity investment of 35.5275 million yuan in Amiao Robotics in 2025. The real signal is that in the first half of 2026, it did not put in more money. Once industrial capital stops adding, the path to the next financing round is half narrowed.
A person familiar with the matter put it bluntly: the commercialization cycle for embodied intelligence is too long, and Amiao Robotics has run into cash flow problems. The response from one participating institution was just one sentence: the company is still operating normally. In venture capital circles, that sentence explains limited things; it usually only means the company has not yet closed its doors. Whether the people are still there, whether the money is still enough, and whether orders can be renewed are all outside that sentence.
Amiao Robotics' crisis did not begin with the departures; it began with the next financing round "hitting obstacles."
From SU7 to Amiao Robotics
Founder Liu Fang's resume is Amiao Robotics' most valuable asset.
With a bachelor's and master's degree in computer science from Shanghai Jiao Tong University, he joined Google China in 2006 and was among its earliest engineers. In 2012, Liu Fang entered Xiaomi with these credentials, working on MIUI Life Yellow Pages, and later managing Xiao Ai speakers, translators, and car-mounted rearview mirrors. After Xiaomi Auto launched, he moved to autonomous driving, serving as product technology lead and mass production lead, building the team from scratch and pushing it all the way to mass production, with the results landing on the SU7.
In the second half of 2024, Liu Fang went to Anker Innovations to lead its newly formed embodied robotics team, reporting directly to Anker vice president Zhu Fanghao. A few months later, Liu Fang left and founded Amiao Robotics in September of the same year. Anker turned around and became the lead investor, serving as both incubator and funder.
In March 2025, Amiao Robotics, founded only half a year earlier, completed a seed round of about 100 million yuan, led by Anker Innovations, with Z Fund (the Zhipu Ecosystem Fund) and K2VC participating. Soon after, Sunwoda added an investment of several tens of millions of yuan, and CICC Capital and Junshan Capital participated in an angel round of several tens of millions of yuan, bringing the three rounds of cumulative financing to nearly 200 million yuan.
The financing pace was fast, yet equity was not diluted much. Liu Fang directly holds 50.5519%, and indirectly holds 14.8285% through the holding platform Amiga, totaling about 65.38%. Qichacha shows that Amiga was established in March 2025, and its executive partner is Liu Fang himself. On the institutional side, Anker Innovations holds 8.4253%, Xinglian Dingsen 7.9326%, CICC Guanbo 5.1709%, Qianhai Hongsheng 4.2127%, K2VC 1.6851%, and the two Huahan funds together hold 5.17%; none exceeds 10%.
For an early-stage company that has just completed three financing rounds, this structure is unusual. Financial investors typically take a higher proportion in the seed and angel stages, but here most of the space was left to the founder.
It is worth noting that, according to the annual industrial and commercial report, Amiao Robotics had only 23 people covered by social insurance in 2025.
Orders, but no numerical verification
Amiao Robotics does not lack order stories. According to its official website, it completed the first batch of robot deployments in factories in October 2025, and in December 2025 the company signed its first commercial order and its first overseas commercial order. Among the customer list, Sunwoda is the most noteworthy. It indirectly holds shares through Qianhai Hongsheng, making it both shareholder and customer. This dual identity of industrial party plus customer is very common in the embodied intelligence industry; the advantage is that the scenario is readily available, and the disadvantage is that orders are easily seen by outsiders as related-party support.
The overseas customer is Canada's CAMT Automotive, and another partner, Dibike Energy, has deployed Amiao Robotics' solution in factories in Southeast Asia and the Americas. Dibike mentioned one detail: in the past, production line retrofits required a two-week shutdown, but after using Amiao Robotics' solution, changeover time was compressed to 3 hours. This detail explains the problem better than the order amount; at least it proves that the robots have actually run on the production line.
In November 2025, Liu Fang told the media that the company had three KA customers, with robots running alongside production on customer lines; a single unit was priced at about 200,000 yuan, could replace 2 to 2.5 workers, and had a payback period of one to two years. But not a single item was disclosed: order amount, number of units delivered, acceptance status, repurchase rate, or revenue scale.
At the Shanghai Industrial Expo in October 2026, Inovance Technology exhibited a flexible assembly solution for sheet metal parts on an automotive welding demonstration line. Inovance provided the robot body, drive, force control, and production line integration, while Amiao Robotics handled the perception and decision-making layer. In its official statements, Inovance listed Amiao as an "AI partner" and mentioned thanks for its support of the smart automotive welding island sample site. This is technical collaboration at the sample-site level; neither side disclosed procurement amounts, the number of units delivered, or the production lines where it was deployed.
Based on public information, Amiao Robotics' order situation can be roughly discerned: the customers are real—Sunwoda, Canada's CAMT, and Dibike can all be verified. The scenarios are also real; sheet metal assembly and 3C loading and unloading are workstations that genuinely exist on production lines. There are two versions of the factory entry dates, August and October, and the difference is not significant; the Inovance sample line is technical collaboration and can prove capability. What is missing is transaction-level numbers: the amount per order, the number of units delivered, and whether any customer has placed a second order—none of these has been made public to date.
Such disclosure is not surprising for an early-stage hardware company: customers do not want to go public, and the company does not want to be price-compared by peers. But Amiao Robotics has already reached the point of contraction, and the same absence changes in nature. What investors care about next is not how long the customer list is, but whether that list can turn into cash collection within a year.
The stratification of the industry
Amiao Robotics is not an isolated case; it is just one notch on the thermometer.
According to IT Juzi statistics, in the first half of 2026, 322 financing deals occurred in China's embodied intelligence sector, totaling about 93.47 billion yuan, exceeding the 45 billion yuan for all of 2025, with about 20 new unicorns valued at over 10 billion yuan added in half a year. IDC data shows that in the first half of 2026, global humanoid robot shipments were nearly 25,000 units, with China accounting for more than 19,000 units, or 77.9% of the global total.
On August 19, Unitree Technology listed on the STAR Market with an issue market value of about 61 billion yuan, and in November its market value briefly stood at 320 billion yuan. Mech-Mind listed on the Hong Kong Stock Exchange on September 1, and Benmo, Xinghuan, Huanchuang, and Youdi successively...
Taken together with public information, Amio Robotics' order situation is broadly clear: the customers are real, and Sunwoda, Canada's CAMT, and Dibico can all be verified. The scenarios are real too—sheet metal assembly and 3C loading/unloading are workstations that genuinely exist on production lines; there are two versions of the entry-into-factory timeline, August and October, and the difference is not significant; the Inovance sample line is a matter of technical collaboration and can demonstrate capability. What is missing is transaction-level figures: per-order amounts, number of units delivered, and whether any customer has placed a second order—none of these has been disclosed to date.
It is not unusual for early-stage hardware companies to disclose in this way: customers do not want to go public, and the company does not want to be price-compared by peers. But Amio Robotics has already reached the point of contraction, and the same gaps take on a different character. What investors care about next is not how long the customer list is, but whether that list can turn into cash collection within a year.
Industry stratification
Amio Robotics is not an isolated case; it is just one notch on the thermometer.
According to IT Juzi statistics, in the first half of 2026, 322 financing deals took place in China's embodied intelligence sector, totaling about 93.47 billion yuan, exceeding the 45 billion yuan for all of 2025, with about 20 new unicorns valued at over 10 billion yuan added in half a year. IDC data shows that in the first half of 2026, global humanoid robot shipments were nearly 25,000 units, with China accounting for more than 19,000 units, or 77.9% of the global total.
On August 19, Unitree Technology listed on the STAR Market with an issue market value of about 61 billion yuan, and in November its market value briefly stood at 320 billion yuan. Mech-Mind listed on the Hong Kong Stock Exchange on September 1, and Benmo, Xinghuan, Huanchuang, and Youdi successively went public, with more than ten robotics companies sprinting toward IPOs over the year.
But an IPO is no longer a safety cushion either: of the 14 new listings in Hong Kong since September, 8 broke issue price on their first day, and Maccura fell 42.93% on its first day. AgiBot launched its Hong Kong IPO in July with a target valuation of HK$40 billion to HK$50 billion, announced a postponement in October, and pivoted to launching a new funding round at a valuation of 50 billion yuan. Even the leaders have to watch the window.
On the other side, more companies are falling: according to media statistics, more than 20 robotics companies have collapsed in the past 19 months. CloudMinds Robotics raised about 5.4 billion yuan cumulatively, with a peak valuation of over 20 billion yuan, and its Beijing company was accepted for bankruptcy liquidation by the Fangshan District Court in October 2025. DeepBlue Robotics was accepted by the Pudong Court in May 2026, with claims from 104 employees exceeding 23 million yuan, while its parent DeepBlue Technology was once valued at 14 billion yuan. Yinghe Robotics was accepted for bankruptcy liquidation by the Shenzhen Intermediate People's Court this summer; ByteDance and Meituan had both invested.
The problem with order reporting standards is more worth examining than the number of orders. What the industry counts as big orders often includes framework agreements, intended procurement, and channel stocking. The signed amount can be very large, while acceptance, payment, and repeat purchases follow a different process, with delivery capability in between. UBTech's publicly verifiable contract amount this year is about 1.27 billion yuan, but delivery guidance for the U1 series was cut from a guaranteed 10,000 units to 1,500–2,000 units. A contract is signed, but whether it can be delivered as agreed is another matter; this gap says more about the industry's real pace than the contract amount itself. Wujing Dynamics pushed the gap to an extreme: eight months after founding, it claimed 1 billion yuan in orders and 110 million yuan in financing, while actual deliveries were zero; in October it stopped work over unpaid wages, and the actual controller became unreachable.
What Amio Robotics has left
Back to Amio Robotics: its technical direction is not wrong. Flexible assembly is the hardest and most valuable part of industrial scenarios, and first-person video plus real-machine reinforcement learning fits the current mainstream VLA approach. Liu Fang's mass-production experience is also real, not a PowerPoint résumé.
The problem is that these advantages have not yet turned into verifiable commercial figures. The company says its technology system is modular and that product iteration will not be interrupted by personnel changes. But the simultaneous departure of the CTO, COO, and product head takes away experience honed through working with customers.
Anker Innovations confirmed an investment loss of 1.539 million yuan, which is not large in absolute terms. Roughly back-calculating from its 8.4253% stake in Amio Robotics, if only the net loss shared under the equity method is counted, this loss corresponds to the investee's loss of about 18 million yuan in the first half of 2026. For a hardware company that puts most of its cash into R&D and delivery, this pace is not out of control. A loss of 18 million yuan in half a year is not fatal in itself; what is fatal is that the losses continue while the refinancing channel is narrowing.
The one that stopped was a lead investor, not an ordinary financial investor. Anker Innovations both participated in incubation and is the largest external shareholder, so it knows best how the company is progressing. Its decision to stop putting in money is equivalent to closing half the door on the next round.
The remaining paths are just a few: find new industrial parties or state-owned capital to inject funds; being acquired is also a possibility; and further down is continued contraction, cutting personnel and expenses to a level that existing orders can cover, and surviving first.
There are still a few cards left to count. Sunwoda holds shares through Qianhai Hongsheng and is also a customer; this relationship at least guarantees a stable validation scenario. Dibico has already deployed the solution to factories in Southeast Asia and the Americas, and the retooling demand for overseas capacity is real. At a unit price of 200,000 yuan, compared with the comprehensive labor cost of 200,000 to 300,000 yuan per workstation per year in manufacturing, replacing 2 to 2.5 workers means customers can recoup their investment in one to two years.
What truly determines whether these cards can be played is data. In a November 2025 interview, Liu Fang gave a timetable: that year the company planned to collect thousands of hours of data, and by the end of 2026 reach tens of thousands of hours. By his own judgment, that magnitude is only the foundation for training robots to do rigid assembly; to handle complex work such as flexible assembly, tens of thousands of hours more are needed. Even if the end-of-2026 milestone is met on schedule, it will only mean the first leg of the journey is complete.
Another goal Liu Fang set at the same time was to compress the deployment cycle for new workstations from several months to within a week. These two things are actually two sides of the same coin: once enough data is accumulated, retooling may be completed by on-site workers themselves, no longer relying on an algorithm team stationed at the site. If the goals are fulfilled, the business model will be much more solid than it is now. It just needs time, and a company with tight cash flow is precisely the one that lacks time most.
This article is from the WeChat public account "Rongzhong Finance" (ID: thecapital), author: Dai Xianchao, editor: Wuren, published by 36Kr with authorization.