VC funds are now chasing the seasoned veterans in the intelligent driving industry.

If you take a close look at the current primary market of embodied intelligence, you will find a strong sense of mismatch.

On the one hand, several leading companies that have been exposed with listing plans are falling into the panic of "tightening IPO policies for humanoid robots"; on the other hand, a group of new startups are advancing with overwhelming momentum, whose financing pace, financing amount and investor background are all astonishing.

The most typical example is Kunlunxing Robotics: it has completed 3 rounds of financing in total in less than 90 days since its establishment, with its valuation directly exceeding 1 billion US dollars, successfully ranking among the unicorns. According to media reports, the first round of financing of this company was snapped up by top institutions less than 3 days after the enterprise was registered.

Behind the rapid rise of Kunlunxing Robotics, the hard-core background of the founding team obviously plays a decisive role. Ren Geng, the CEO, has successively served in Huawei, Alibaba and ENN Group, with his resume covering hardware engineering, AI innovation, hundred-billion-level commercial operation and global operation; the other co-founder Lang Xianpeng is the "No. 1 employee" of the intelligent driving department of Li Auto, who built the intelligent driving R&D system of Li Auto from scratch, and led the team to deliver the end-to-end + VLM solution. Before leaving Li Auto, he was also in charge of the hardware body of Li Auto's humanoid robot project.

In the words of GL Ventures, the investor, this is a truly scarce team in the current embodied intelligence market.

When you spread out the list of startups that have received large-scale financing recently, you will find that teams with intelligent driving background like Kunlunxing Robotics have almost become the standard configuration — Huang Qingqiu, Yang Ruigang, Liu Nianqiu, Zhang Ying... all of them are experienced veterans with rich technical experience in the field of intelligent driving in the past.

It is worth mentioning that along with this wave of entrepreneurship, the "preference" of capital has also changed. In the past, the primary market was willing to pay for the final imagination of "general-purpose robots" — the grander the story, the fancier the technology, the higher the valuation; but in the narrative of this new batch of companies, the most frequent keywords have been replaced by commercialization, data flywheel and vertical scenarios.

Whether it is the scramble for "veterans" in the intelligent driving industry or the emphasis on commercialization and data flywheel, to some extent, it reflects the attitude change of the current capital market towards embodied intelligence — compared with the grand final story, they now pay more attention to "certainty" and "sense of security".

Veterans in intelligent driving are collectively turning to embodied intelligence

Let's start with the latest news.

Last month, Feimo Yi Technology, an embodied intelligence startup, announced the completion of its first round of financing of nearly 100 million US dollars, led by IDG Capital, with participation from Yunqi Capital, Didi Chuxing, Fosun RZ Capital, Yoerso Capital and Hesai Technology. The company was established earlier this year, and its founder Liu Nianqiu used to be the technical partner of Motovis, participating in promoting the implementation of Motovis' map-free intelligent driving solution, and the evolution of the technical system to end-to-end models and VLA large models.

On the same day, media reported that Zhang Ying, former head of Horizon HSD, had left his job to start a business, and his new company had completed multiple rounds of financing just two months after its establishment. Although the report said it was positioned as a Physical AI company, according to our understanding, this project is also located in the field of embodied intelligence.

Looking further back. At the beginning of July, Moqi Intelligence, founded by Huang Qingqiu, former head of Huawei's intelligent driving AI department at the end of 2025, announced the completion of a series of angel round financing of more than 1 billion yuan, with both Alibaba and Tencent participating, and more than ten institutions including BlueRun Ventures, Legend Star, Gaorong Capital and Source Code Capital also appearing on the shareholder list.

It is worth mentioning that when the financing was officially announced, the company's first product had not even been officially unveiled, but its valuation had exceeded 7 billion yuan.

In addition, Nuwa Robotics, founded by Yang Ruigang, former director of Baidu's Autonomous Driving and Robotics Laboratory and former CTO of Inceptio Technology, in February this year, also successively announced two rounds of financing within 3 months, with investors including Plug and Play China Fund, Blue Lake Capital, Butong Capital, Gongqingcheng Puyi Investment and others.

The projects listed above are only newly established in the past six months. Earlier projects, such as Tishi Intelligent Navigation founded by Chen Yilun, former CTO of Autonomous Driving of Huawei's Intelligent Automotive Solution BU, and Zhijian Power founded by Jia Peng, former head of intelligent driving R&D of Li Auto, have also completed multiple large-scale financings one after another this year.

More projects are still under the radar. Hazel, an embodied intelligence practitioner, told me that not long ago she went to Beijing on a business trip and organized an industry gathering. During the dinner, she heard from the side that there are quite a lot of people leaving the intelligent driving industry to start embodied intelligence businesses recently, many of whom are from leading intelligent driving companies such as Huawei, Horizon, and "NIO, XPeng, Li Auto". "The industry is getting more and more competitive," Hazel sighed.

The more samples there are, the more it shows that this is not a aimless chase for the trend, but a collective transition with a highly consistent direction. So, why now?

In the view of Zhang Jiaqi, Executive General Manager of Chentao Capital, this is to some extent a "timely turn" after trade-off.

He further explained that currently the technical route of the entire intelligent driving industry is converging rapidly, and leading players in urban NOA have successively entered the engineering implementation stage of mass production and volume release. For a core talent who has mass-produced end-to-end and VLA solutions, staying in the original system to go through another version will not bring high marginal professional benefits, and may even face potential professional uncertainty caused by possible technical route switching. "At this time, while the primary market still maintains enthusiasm for embodied intelligence, starting a business outside can get good attention and valuation," said Zhang Jiaqi.

On the other hand, for intelligent driving talents, their capabilities can almost be transferred without loss. Cao Jishan, founding partner of YunDao Capital, said that in the past ten years, autonomous driving may be the largest and most demanding industrial training in the field of physical intelligence. It operates on open roads, facing heavy rain, hail, sudden pedestrians darting out from obstacles, retrograde electric vehicles, and almost endless long-tail events; at the same time, it needs to meet the requirements of low latency, high reliability and functional safety. More importantly, it has achieved large-scale terminal deployment and continuously generates data in the real environment.

"Today, many industrial robots are still stuck in small-scale pilots in a single factory; products for lawn mowing, snow removal, pool cleaning and other scenarios are far from the terminal scale and environmental complexity of automobiles. Therefore, what autonomous driving leaves for this round of entrepreneurship is not just several sets of algorithms, but a complete set of industrial capabilities: multi-modal perception, real data closed loop, model training, end-side deployment, software and hardware coordination, safety redundancy, and experience of long-term delivery to large customers. Transferring these capabilities to industrial inspection, unmanned mining or logistics distribution, in many cases, is indeed like standing on the shoulders of giants."

In addition, in the article "Exclusive | Liu Nianqiu, the veteran of intelligent driving, starts a business and raises nearly 100 million US dollars in the first round" previously released by ChinaVenture, Liu Nianqiu, founder of Feimo Yi Technology, also mentioned a key point.

In the previous wave of embodied entrepreneurship two years ago, due to the incomplete supply of the supply chain, if you wanted to develop embodied intelligence at that node, you needed to invest a lot of energy to start from the basic components. But now the situation is completely different — there are a large number of motors and reducers on the market that are specially optimized for robots, and the difficulty of body manufacturing has been greatly reduced. At the same time, the supply of AI infrastructure is far more abundant than it was two years ago.

"This means that we no longer need to spend one or two years 'reinventing the wheel'. Instead, from the first day, we can focus on thinking about the product itself, and building a closed loop for continuous iteration of products, data and models," Liu Nianqiu said.

When the three elements of timing, capability and infrastructure are all ready at the same time, the window for transition opens.

From "general-purpose myth" to "data flywheel", capital starts to verify actual value

It needs to be clarified that talents from the intelligent driving industry have always been the main force of embodied entrepreneurship, which is not a phenomenon unique to this round. For example, Gao Jiyang, the founder of Xinghaitu, an earlier batch of startup, also came from the intelligent driving system. What has really changed is the stories told by this new batch of teams since the end of 2025, and the definition of a "good company" by the capital behind it.

In the past, the fanciest narrative of embodied entrepreneurship was the final imagination of general-purpose humanoid robots. At that time, as long as the team background was strong enough and the story was flexible, route adjustments within one or two years could be tolerated; but this round of funded startups have changed their focus — they are more willing to clarify which type of real scenario they will enter first, for whom, what problem they will solve, how to calculate the single transaction revenue, and take "commercialization", "real data" and "data flywheel" as the key emphasized parts.

For example, an embodied intelligence company I wrote about in July — Zhengqi Future, whose founder Wu Shulin once served as Vice President of the Intelligent Driving Business Group at Baidu before starting his own business. This company takes short-distance travel as its entry scenario, collects "real data with human intention", and forms a set of data flywheel through the QUORRA DoorMind world model and the QUORRA FLAT motion execution platform. The company completed 3 rounds of financing within 8 months, with investors including CDH VGC, Linear Capital, SAIC Hengxu, Bolear Capital, Zhengxuan Investment, Fosun Chow Tai Fook Strategic Investment and others.

Even companies like Kunlunxing that write "general-purpose humanoid robot" into their positioning are no exception. What the founder repeatedly talks about in public interviews is to "make mass production and self-sustainability run through first".

In Cao Jishan's view, this is not entirely investors or startups "coining new terms", but the industrial stage has changed. In the early stage, "embodied intelligence" was more of an academic perspective. Everyone talked about papers, humanoid robots, and a slightly sci-fi final vision. At that time, the story could be grand, the landing could be far away, and the capital market was willing to give time. But when the concept has evolved from embodied intelligence to today's Physical AI, what corresponds to it is not a copywriting upgrade, but a downward shift of focus: from telling stories to delivering tangible results.

"At the time point of 2026, when evaluating whether a company is worth investing in, the measurement standard must be shorter and stricter. When the upper-layer large models and world models are already crowded with giants taking sides, entering the stage of integration and capitalization, the places where ordinary early institutions can find high odds returns are actually in the more 'hard' industrial gaps," Cao Jishan said.

In this regard, Zhang Jiaqi has a similar view. He recalled that from the end of last year to the first half of this year, the market clearly preferred general-purpose model companies — investing in general-purpose models could get dozens of times of return in less than a year, while investing in solutions might only get two or three times of return. Therefore, a large number of scenario solution companies did not have smooth financing at that time; by 2026, the general-purpose track has been "crowded into a mass", and while competition raises risks, the commercialization pace of companies is generally facing greater tests. "In this case, some funds have turned to invest in solution companies, and we are also very willing to invest in such companies, because commercialization path and PMF (Product-Market Fit) are the core competitiveness at the company level."

In addition, Zhang Jiaqi also provided a reference coordinate, which is RIVR, a company acquired by Amazon in March this year.

This is a wheel-legged delivery robot company headquartered in Zurich. Its early investors include Linear Capital and HSG, and later Bezos Expeditions of Jeff Bezos and Amazon have successively invested in it. The company is telling a story of "real world + data flywheel": every door-to-door delivery generates real data, which is then fed back to the model through large-scale reinforcement learning and self-supervised training, leading to the evolution towards general-purpose physical artificial intelligence.

In his view, such targets are taken over by giants with real money, which to some extent also drives domestic teams and capital to pay attention to and bet on this track.

Capital is getting less, and the screening standard is getting stricter

Although many cases of large-scale financing obtained recently are mentioned above, after overall communication, everyone has a consensus that entering the second half of the year, the primary market's investment in embodied intelligence has begun to cool down.

A large reason behind this is the increasing uncertainty of exit channels. For a period of time in the past, the expiration of the special listing channel on the Hong Kong Stock Exchange and the rumor of "tightening IPO policies for humanoid robots" have undoubtedly disturbed the sentiment of investors; the regulatory statement of "no barbaric growth, and return to value creation" has also made many institutions tighten their investment pace.

Once the "engine" driving the trend stalls due to the fading listing expectation, the early-stage companies that have just started are often the first to be affected. When capital becomes less, the screening of "people" begins to become strict.

Cao Jishan said that when an embodied intelligence (Physical AI) company comes for financing, he mainly looks at four indicators:

At the early stage, the first thing to look at is the team. Whether the founder has actually worked on a truly complex project, whether the core members have cross capabilities of technology and commercialization, and whether the team has a high enough upper limit.

The second is the scenario. Whether the market is large enough, whether the demand is real, and whether the vertical scenario entered today can be expanded to more customers or adjacent industries.

The third is the capital structure. Embodied intelligence often requires long-term R&D, hardware iteration and customer verification, which is a business that requires continuous financing. The team's execution capability, shareholder structure and subsequent financing capability will all affect whether it can achieve commercialization.

The last is the technical ceiling. Scenarios will change, models will be upgraded, and today's solutions may soon become obsolete. The team cannot only solve the current delivery, but also have the ability to keep up with the changes of the underlying paradigm.

Specifically for those intelligent driving talents who have transformed to the embodied intelligence industry, Zhang Jiaqi also has a set of talent screening logic. In his view, a high-potential leading intelligent driving team for embodied intelligence needs to have four capabilities at the same time — 1) Have practical experience in implementation and mass production, which must be the first priority; 2) Have trained end-to-end models with sufficiently large parameters; 3) Have processed massive driving data; 4) Have used computing power clusters at the level of nearly 10,000 cards.

"After screening according to this standard, there are only about 20 people in the whole industry. If we also consider the level of enterprise management and operation, this number may be even smaller," Zhang Jiaqi said.

During the communication process, many investors also mentioned a major variable that may change the pattern of the embodied intelligence industry in the future, that is, the intelligent driving giants. The technical advantages of the intelligent driving industry transforming to embodied intelligence have been discussed above. Compared with startups, once giants such as Huawei, Horizon, NIO, XPeng and Li Auto enter the market, in addition to the so-called technical advantages, they also hold a large amount of cash, computing power and scenarios, whose impact is self-evident.

However, they also emphasized that the current technical route of embodied intelligence has not converged, and there is no certain large-scale revenue yet, so giants should not rashly enter the market in the short term.

This means that startups still have a window of opportunity at present. What they need to do is to build higher data and time barriers before giants enter the market.

This article is from WeChat official account "ChinaVenture", author: Wang Manhua, authorized for release by 36Kr.