Mass Exodus of Senior Executives: The Spectacle Behind the 93.5 Billion Yuan Embodied Intelligence Industry
Over the past six months, two types of "prosperity" have emerged in the field of embodied intelligence.
On one hand, capital is pouring in at an unprecedented speed. In the first half of 2026, the total domestic financing amount for embodied intelligence reached 935 billion yuan, 5 times that of the same period last year; there were 322 financing deals, a year-on-year increase of 137%. The number of decacorn companies has risen from 3 in 2025 to at least 22.
On the other hand, a group of core talents are leaving their posts. From founders and chief scientists to top product executives, CTOs, COOs and CFOs, almost every key position is seeing personnel changes.
Companies are getting increasingly valuable, but the most experienced insiders are stepping off the ship.
NewDing Capital judges that the industry will enter a reshuffling phase by 2026, and only 10 to 20 of the nearly 1,000 domestic humanoid robot enterprises will eventually survive.
The first phase of embodied intelligence has come to an end. Understanding the departure of every senior executive means understanding the future of embodied intelligence.
01
Who will buy the products after 10,000 units are manufactured
The departures of founder-level executives first drew public attention.
At the 2026 CCTV Spring Festival Gala, 6 humanoid robots from Magic Atoms performed on the same stage as Jackson Yee and Jordan Chan, while hundreds of "robot pandas" completed the world's first performance of 100 four-legged robots in clusters at the Yibin sub-venue. The company became famous overnight only two years after its establishment. But before the hype faded, Wu Changzheng, the founder, was exposed to have left his post to start his own business in March, after he stepped down from positions in several core subsidiaries under the company one after another. Citing people familiar with the matter, Lan Jing News stated that the core reason lies in the divergence of development concepts between Wu Changzheng and the shareholders; at that time, the company just released the signal of "IPO as early as 2026".
In June, Shi Xiaoxin, the top product executive of XPeng Robotics, resigned voluntarily and later founded the home robot company Aimo. According to media reports, he previously had disagreements with He Xiaopeng on the mass production pace: He Xiaopeng wanted to speed up mass production, while Shi Xiaoxin raised a very realistic question — "After manufacturing 10,000 robots, where do we place them?"
This sentence has become the watershed of the entire industry.
In previous years, the core task of robot companies was to prove that "they are capable of manufacturing the products". But in 2026, top product executives have to answer a more difficult question: Who will buy the products after they are manufactured?
Once the pace is misaligned, talents will become the cost of the divergence. In July, Mi Liangchuan, the head of core business of XPeng Robotics, also confirmed his departure, and He Xiaopeng later personally took the position of CEO of the robotics business. Zhang Li, co-founder and COO of ExDynamics, left his post in April. Top business and finance executives including Wang Yong, CFO of Dobot, and Mao Chongzhao, Vice President of Qiangchu Intelligence, also left their positions one after another in this half year.
It is worth noting that Shi Xiaoxin and his peers are moving in a highly consistent direction — the home scenario. Many members of his Aimo team are former XPeng employees. This is no coincidence: when the question of "who to sell the products to after manufacturing" in the factory scenario cannot be answered clearly for the time being, a group of people turn to the home market, which has a larger space but also faces more challenges.
Zhang Chi from NewDing Capital judged in an interview with *China Fund News* in August that both ontology enterprises that can reduce hardware costs and software companies that focus on robot brains have the chance to win.
02
The former employer invested in his next company
Another group of people who left are the technical souls of various companies. Their departure corresponds to deeper uncertainty in the industry: there is still no standard answer for the technical route so far.
Luo Jianlan, chief scientist of Agibot Robotics, disappeared from the partner list on the official website only 10 days after the company officially announced its plan to list in Hong Kong. He holds a doctorate from UC Berkeley, is a former research scientist at Google DeepMind, and is one of the authors of SERL, a real robot reinforcement learning system. After joining Agibot, he led the construction of the AgiBot World dataset with more than 1 million real robot data entries. Agibot responded that "it is a post adjustment and no one has left the company", but did not specify the new position, and his personal homepage has also deleted the relevant employment information of Agibot, so the real status remains a mystery.
In February, Xu Huazhe, co-founder and chief scientist of StarMap Robotics, left his post at the peak moment when the company's valuation was about 20 billion yuan and the cumulative financing was nearly 3 billion yuan. In September, Zhao Deli, head of the embodied intelligence laboratory of Alibaba DAMO Academy, was reported to have left her post.
Ding Yan, co-CTO of Luming Robotics, has changed three companies in 14 months: from Shanghai AI Lab to Yixing Robotics, which was dissolved 4 months later due to Geely's strategic adjustment; he brought more than ten core backbones to join Luming Robotics, and left again in less than 280 days. He advocates the UMI ontology-free data collection route, and once publicly judged that UMI and teleoperation "are not complementary, but replace each other".
These people are not ordinary engineers, but the ones who determine the "technical direction" of a company. Their frequent mobility reflects that the entire industry has not reached a consensus on "which route to bet on next": end-to-end VLA, world model, or the integration of the two? Should data be collected through teleoperation, ontology-free collection, or simulation generation?
Capital is voting with its feet. In the first half of the year, funds clearly shifted to "embodied brains". Moqi Intelligent completed an angel round of more than 1 billion yuan in 6 months, and its CTO Huang Qingqiu said bluntly, "If you don't work on the brain and pre-training, this wave of embodied entrepreneurship makes no sense". After four rounds of financing, the valuation of Zili Robotics exceeded 20 billion yuan, and Wang Qian, the founder, believes that the competition in the next stage is "essentially the capability of data closed loop and model evolution".
Once top technical executives, companies and shareholders place different bets on the route, separation is inevitable. After leaving, Xu Huazhe founded Crack Robotics in April, even completely abandoning the mainstream VLA route, focusing on the home scenario, and obtained tens of millions of US dollars in angel round within one month, and completed the 100 million US dollar level Pre-A round in August. He said bluntly that a large number of manufacturers send humanoid robots to factories to do the handling work that mechanical arms can complete, "essentially just using new humanoid robots to do the work of the old era".
Talking about Xu Huazhe's departure, Gao Jiyang, founder of StarMap Robotics, explained: the team needs to find a balance between creating customer value step by step and more forward-looking innovation. But he also emphasized that the two sides "do not have many disagreements", and StarMap Robotics is willing to invest in Xu Huazhe's first round of financing to support his exploration of home applications.
Xu Huazhe also has his own account of time. In an interview with Oasis Capital in May, he said "time cost is the most expensive", leaving one year earlier to do what he wants to do is worth giving up the wait for potential financial returns.
03
The truth of the resignation wave: there is abundant capital, but the rules have changed
Senior executives are leaving so intensively, is the industry cooling down?
On the contrary, companies are still snatching talents with high salaries.
In April this year, Ubtech recruited a chief scientist of embodied intelligence with an annual salary starting from 15 million yuan, up to 124 million yuan. The Research Office of the State Council disclosed that the ratio of demand to supply for technical talents in the robot industry has reached 5.2:1, with a talent gap of millions. Top talents usually get a pay rise when they change jobs. Unitree has implemented two option incentive plans covering 23 people, 9 of whom have left — but the fact that these people can find new jobs quickly itself indicates that the industry is still expanding.
Don't only focus on who left, but also see where they go after leaving. In this round, most people just left a certain company and soon reappeared in the industry — they left the company, not the track.
More critically, there are still organized new talents pouring in at the gate.
The first group is autonomous driving talents. Xiao Zhongyang, former head of intelligent driving algorithm of NIO, founded Corona Robotics, Liu Fang, former head of intelligent driving of Xiaomi Auto, founded Amio Robotics, and a group of senior executives from the Horizon Robotics system have entered the embodied intelligence field one after another. The perception, planning and control, and simulation training of autonomous driving are fundamentally connected with the underlying technology of robots, and the talents accumulated in the decade of intelligent vehicles are finding their second battlefield.
The second group comes from the automotive industry chain and large tech companies. SAIC, Chery, NIO and Li Auto are all making layouts. SAIC's "Nengzai No.1" has been on duty on the battery mass production line, and SAIC Capital has invested more than 800 million yuan in total; Tesla plans to achieve an annual production capacity of 1 million units for Optimus. Investors have ranked the discourse power in the order of "industrial capital first, state-owned capital second, financial VC third" — industrial capital is both a shareholder and a customer.
Old teams are split apart, and new teams are formed. This does not mean the card table is removed, but the seats are being rearranged.
However, few enterprises in the whole industry are profitable: Ubtech achieved revenue of 2.001 billion yuan in 2025 with a net loss of 790 million yuan; Unitree's non-recurring net profit in the first quarter of 2026 decreased by 52.55% year-on-year. Once an enterprise truly runs through the business model, the reshuffling will accelerate. HuaAn Capital judges that 2027 may become the first year of mass production delivery, and enterprises that cannot get orders will face survival crisis.
Now the moisture is being squeezed out, and the industry has entered the "focus on repurchase" stage.
In early September, Shao Tianlan, founder of Mech-Mind, publicly criticized some "gathering type enterprises" for making false revenue through data collection centers and related transactions, naming Galaxy General. Foreign media The Information reported the window guidance, and the National Development and Reform Commission emphasized preventing "rushing headlong into mass action". Customer lists and repurchase data began to appear in financing documents, and due diligence shifted from "reading papers and watching demos" to "inspecting production lines and order repurchase records".
This article is from the WeChat Official Account "Pencil Way" (ID: pencilnews), written by Xin Xiao, edited by Huang Xiaogui, and published with authorization by 36Kr.