Want to switch to the robotics industry for a quick buck? | 36Kr

Source: Still from the film Dunkirk

By Xiao Man

Edited by Yang Xuan, Li Qin

"If you don't jump now, it might be too late." "Staying here will make you less and less valuable."

Sun Xinyang has been hearing these remarks more and more frequently from headhunters and resigned colleagues. This sense of urgency also prompted him to seriously consider leaving XPeng this spring.

In the three or four months before his resignation, he got a new direct supervisor every month, and the workstations around him filled up and then emptied again. Xu Zhou also works at XPeng, and his team faced the same situation: in half a year, he saw 9 colleagues leave, and had to say "wish you all the best in your future endeavors" in the work group every month. The destinations of these leavers were highly consistent — robotics companies.

This is a robotics talent recruitment wave that started in 2024 and has lasted for nearly 2 years, with the automotive intelligent driving sector being the largest source of outgoing talent. This migration wave seems to be coming to the end of its most frantic talent acquisition phase.

There used to be widely circulated stories in the industry: when the demand for talent was at its peak, robotics companies could poach automotive intelligent driving engineers with triple salaries; the top leader of one robotics company personally approved a condition of covering the mortgage for a candidate to attract him to join as a co-founder.

But today, these most extreme stories have become rare, and the more common job-hopping condition is a 50% salary increase — there is still a premium, but it has become more reasonable.

The criteria for selecting talents have also evolved from being broad at the beginning to being clear. Initially, anyone who had experience in autonomous driving could get an offer from a robotics company. As long as you had experience in perception or planning, with a period of R&D experience in the automotive industry, you could get a doubled salary and a stock option grant.

A year later, the criteria have narrowed. Robotics companies now only want technical talents who have experience in VLA (Vision-Language-Action model technical route), or even those who have actually operated real robots.

Capital and resources move in cycles, and so do career opportunities. As Unitree Robotics' market value once exceeded 400 billion yuan on its IPO debut, and the list of leading robotics companies queuing for listing has been confirmed, capital in the robotics track is concentrating on top players. Chen Wei, former Chief AI Scientist of Li Auto and founder of Leap Forward Intelligence this year, told 36Kr that he believes the opportunity window will close in 2028.

"We have passed the stage where only ideas and models matter. Now everyone is focusing on products and implementation," Chen Wei told 36Kr. The embodied intelligence industry is now "all ready except for the final model". Even if the development of physical models is slightly slower than that of large language models, their capabilities will mature in only two or three years. As robotics companies evolve, the talent profiles they need are also changing accordingly.

Should you change jobs? Is moving to the robotics industry only a quick way to make a buck? Will you end up in a mess after switching? Many people are still waiting and seeing. The automotive industry is grounded, stable and realistic, while the robotics industry has not yet converged on its technology path, and is still at the stage where capital "pays for dreams". Most of the hundreds of embodied intelligence companies will go through a knockout round in the future. To this end, 36Kr interviewed many "veteran automotive practitioners" who have already switched to robotics companies, and invited them to share their real experiences after stepping into this new field.

This is not just a story about automotive talent flowing into the robotics industry. Faced with any new industry and new opportunity, these stories may repeat themselves over and over again.

Those Who Left

Lu Yuan joined Li Auto right after graduating in 2023. That year, Li Auto sold 370,000 vehicles. Li Xiang just started leading executives to learn from Huawei, and would talk publicly on social media about his vision for the industry, organizational structure and products. In the automotive industry at that time, people still believed that the pattern of the industry would be led by product managers who focused on design, rather than procurement managers who squeezed suppliers.

It was also a time when the automotive industry was still heavily investing in technology R&D, and regarded technology as the key to competition. Lang Xianpeng, Jia Peng, Xia Zhongpu were all still at Li Auto, and Lu Yuan worked with them through the 100-day end-to-end sprint and the R&D of Li Auto VLA.

Later, the team that brought Li Auto to the intelligent driving table dissolved, along with the automotive industry's faith in technology. Executives including Xia Zhongpu, Jia Peng and Wang Jiajia left one after another to start their own businesses. Lu Yuan's colleagues also left one by one: some followed Jia Peng to Join Dynamics, others were poached by other embodied intelligence companies.

Every once in a while, a familiar name would disappear from the work group.

Later, the private discussions between Lu Yuan and his colleagues were no longer related to intelligent driving, but about "whether robotics companies need VLA talents", "whether embodied intelligence companies are still recruiting intelligent driving practitioners", "whether it is worth joining a robotics company"...

Photo of XPeng Robotics IRON

In fact, over the past two years, almost all leading automotive companies have been conducting pre-research on robotics. Last year, XPeng Robotics IRON only walked on stage for two minutes, which drove XPeng's stock price up 20% in three days; Li Auto also officially established its humanoid robot business at the beginning of this year; BYD and Changan also have internal pre-research projects.

But Lu Yuan, Xu Zhou, Sun Xinyang and others hardly considered staying at the original automakers.

"Intelligent driving is still the core business, and the company will not allow too many people to transfer to the robotics sector," Lu Yuan said. All of Li Auto's resources including talent, computing cards, computing power and data are allocated to the automotive business. Of course, there are also many intelligent driving engineers in the robotics team, but according to Lu Yuan, most of them are people who did not develop smoothly in the intelligent driving system.

Xu Zhou has a similar experience. When he first joined XPeng, he knew the company had always had a robotics team. In 2025, the company internally demonstrated demos of robots moving goods and screwing screws in factories. But when it came to the official launch event, most of the demos could not be presented live.

"Many people in the robotics team were transferred from the intelligent driving team, but most of them were the ones who lost the internal competition," Xu Zhou said.

Automakers do not want too much talent loss. XPeng has a 3-month "quiet period" requirement for resigned employees. "Robotics companies are in urgent need of talent, and a 3-month delay will easily make the offer invalid," Xu Zhou said with indignation.

"It's impossible to stop it. The general trend is irreversible," Xu Zhou said.

In their conversations, Xu Zhou, Lu Yuan and Lin Yejun all mentioned a point of view: "We have already seen the end of the (intelligent driving) path, and there is not much meaning to continue working on it."

"Two years ago, intelligent driving was in the stage of moving from 60 points to 80 points. Now it has reached the stage of moving from 90 points to 95 points. If you keep working on it, the marginal effect will decrease. But embodied intelligence has just started, and it is in the stage of moving from 20 points to 60 points. For people who want to achieve great things, this is more promising," Xu Zhou said.

In the three years he spent at Li Auto, although Lu Yuan participated in many key projects, he was still only an IC (individual contributor). If he stayed, he would most likely move up the promotion system slowly. But after joining a robotics company, Lu Yuan has become the leader of a 10-person team, in charge of the full-link R&D of embodied intelligence technologies.

Lu Yuan's undergraduate major was robotics, and he learned manipulator control back then, using rules to control robots. But current machine intelligence is entirely a system problem centered on AI, and only by touching the full link can you access the most critical technical issues.

This is also the first time he feels that he has finally returned to what he originally wanted to do, and believes he can make a difference in this field.

"In a steadily developing industry, it is difficult for you to sit at the same table with those geniuses. But in a chaotic new industry, everyone gets a chance to reshuffle the cards," Lu Yuan said.

Embodied intelligence gives more people the opportunity to achieve career advancement. Another algorithm engineer working at an automaker told us that his former colleague, who was only a P7-level regular employee at NIO, now leads a pre-research team of more than ten people at Agibot, responsible for basic model research and has published several papers in a row.

"Working at an automaker is tiring, and working in embodied intelligence is equally tiring, but the pay is higher, with stock options and a brighter future. You don't even need to think hard to figure that out," Lin Weichang, HR of BYD's intelligent driving business, told 36Kr. "Joining the embodied intelligence industry is a big bet. Even if you lose the bet, your salary won't be lower, and robotics companies offer very high cash compensation."

This kind of salary re-pricing quickly spread across the entire industry. A headhunter who has long been in charge of intelligent driving recruitment told us that among the more than 60 candidates he successfully placed in the first half of this year, 95% came from intelligent driving teams, and 90% of them were looking for embodied intelligence positions.

From Sun Xinyang's perspective, the jobs recommended by headhunters to him are also changing. Two years ago when he changed jobs, 3/4 of the positions recommended by headhunters were from intelligent driving companies, and 1/4 were from embodied intelligence enterprises. But this year, the situation is completely the opposite — 3/4 of the positions are from embodied intelligence enterprises, and 1/4 are from automotive-related companies.

If Sun Xinyang stays in the intelligent driving field, he will most likely only get a 30% salary increase, while switching to embodied intelligence, a 50% salary increase is the baseline.

"The total annual compensation for junior practitioners in cutting-edge directions ranges from 600,000 to 800,000 yuan, and senior engineers may get around 1 million yuan," the above headhunter said.

When an emerging industry rises, the salary of talents is sometimes determined by the financing intensity of the industry and high-density recruitment games. The last time engineers' salaries skyrocketed in non-internet fields was in the semiconductor industry in 2020. At that time, many GPU startups emerged in China, and some engineers changed jobs once every three months, with their salaries more than doubling in a year. Five years later, when the valuations of these GPU companies soared after listing, the engineers who made the choice back then had already earned income that other industry engineers would take decades to accumulate.

The headhunter gave an example: an algorithm engineer who worked at Xiaomi for only one year switched to an embodied intelligence company, his total annual compensation rose from 700,000 yuan to 1.1 million yuan, plus stock options. Even if the options become worthless in the end, it is a cost-effective deal for him.

According to data from IT Juzi, in the first half of 2026, there were 288 financing events in China's embodied intelligence and robotics sector, with the total financing amount exceeding 460 billion yuan.

According to incomplete statistics, there are more than 20 companies in the embodied intelligence track with a valuation of over 20 billion yuan, and this number is still growing. "Many institutions have dedicated people keeping an eye on Shanghai Jiao Tong University, Harbin Institute of Technology, and the Hong Kong University of Science and Technology, and some are also long-term tracking talents from DJI and automakers," a primary market investor said. "As soon as someone leaves to start a business, we will contact them immediately."

Another source close to CATL's industrial investment told us that there are two types of internal projects that are most likely to be approved: "genius startups, and startups founded by veterans from the automotive industry."

Even if you are not the core business leader, you can still get a considerable amount of financing in this wave. Industry insiders revealed that the core HSD algorithm leader from Horizon Robotics got swarmed by top investment institutions right after he left to start his own business, and the company's valuation immediately exceeded 10 million US dollars.

A migration has begun.

For many intelligent driving engineers, what they are leaving is not so much the automotive industry, but "an industry whose end is already known". What they are heading for is a place where no clear answer has been found yet.

Moving from a High-rise Building to a Thatched Cottage

"After I came here, I found that there is nothing here." Less than a month after joining the new company, Xu Zhou realized that he had not entered another intelligent automotive industry.

Another former automaker algorithm engineer said: "I used to live in a high-rise building, and now I have moved into a thatched cottage."

The "data platforms, training frameworks, and computing power platforms" that are taken for granted at automakers need to be built bit by bit at robotics startups.

The company where Xu Zhou currently works leases computing power from Alibaba, Baidu, ByteDance, Tencent and other companies. The computing power platforms are not interconnected, which makes data transfer extremely troublesome when he trains a shared model. "This kind of thing usually takes a whole day for one person to handle," Xu Zhou said.

Back at XPeng, the training cards we used were provided by Alibaba, and the storage was also Alibaba Cloud services. The computing power cards could directly access the training data. "We always thought this was the basic configuration," Xu Zhou said.

Startups are small in scale, so they can only buy computing power in small batches, and cloud vendors do not give them much resource preference, so they can only purchase in batches. The founder has already assigned a dedicated person to solve this problem. "There is no way, this is how startups work," Xu Zhou said.

Not just data, many things in startups need to be built from scratch. After Lu Yuan became the team leader, he found that what he spent most of his time on was not training models, but recruiting, sourcing robot bodies, and setting up workflows.

Many things that were already handled by dedicated people at Li Auto, have to be done all over again from zero here. "You have seen the efficient combat mode at Li Auto, where everyone is responsible for their own breakthrough direction. Only after you leave that environment do you realize how rare that state is. Now there is nothing, and you have to manage everything by yourself," Lu Yuan said.

Some people can't adapt. Many colleagues in Xu Zhou's team came from Alibaba, Baidu and ByteDance, and they soon became anxious. For them, the startup is "too small and unstructured". Some people started looking for their next job just two months after joining, and are considering switching to more "established" startups like Agibot and Extant, which have more complete systems, so they can go back to being a cog in a machine.

Xu Zhou is not that flustered. He said that maybe it's because he has experience in mass production, and he is more "down-to-earth". In the early days, intelligent driving was also built bit by bit, collecting data, cleaning data, running models... a lot of dirty and tiring work had to be done by themselves.

Xu Zhou is still in the honeymoon period after just joining the company, while Lu Yuan has been in the embodied intelligence industry for half a year. His most direct feeling is that "although intelligent driving and robotics are both part of embodied intelligence, they are not exactly the same."

"Embodied intelligence is a completely new industry, and there are no ready-made job references," Lu Yuan said. The degrees of freedom and motion algorithms of robots are different from those of automobiles.

Another algorithm engineer Li Liyang also experienced the same gap. Before joining, he thought that anyone who has worked on intelligent driving can do this job, but after he actually started training models, he found that it was a job he had never done