The "Wei Xiaoli" trio (NIO, XPeng and Li Auto) that lost the new energy vehicle competition cannot secure a win in the embodied intelligence track either.
On July 29, media reports stated that Li Auto is planning to develop two-wheeled and bipedal robots, and the company's share price immediately rose by 10% in a single trading day. Coincidentally, several distinct rounds of share price increases for Xpeng recently are almost entirely tied to robotics-related developments.
This reflects a shift in the underlying narrative logic of Li Auto and Xpeng: they are embarking on a "second venture", transforming from electric vehicle manufacturers to embodied intelligence players.
In their strategic framework, automobiles have been integrated into the next stage of the entire AI industry. As Li Xiang put it, vehicles are not just means of transportation, but robots operating in the physical world with continuous perception and real-time decision-making capabilities.
This repositioning has altered the valuation logic for new Chinese EV startups.
In the mature automotive market with slowing growth, new EV startups that lack scale and profitability can no longer rely on the new energy vehicle narrative to support their valuations. Shifting to embodied intelligence grants them access to a high-growth track with explosive potential: by 2050, the global humanoid robot market alone will reach a size of 7.5 trillion US dollars.
The new valuation logic has already been reflected in research reports. Citi has adjusted the valuation models for Li Auto and Xpeng from the traditional single Price-to-Sales (P/S) ratio to the Sum of The Parts (SOTP) method, assigning a 1.5x P/S multiple to Xpeng's automotive business and an 8x P/S multiple to its robotics business.
However, the adoption of the new valuation logic by institutions is more of a bet on high odds than a purchase of certainty. NIO, Xpeng and Li Auto, which underperformed in the new energy vehicle race, may repeat their mistakes in the embodied intelligence track.
The reasoning is that large models are the decisive factor for success in embodied intelligence, but the large model capabilities of new EV startups lag behind both leading AI startups and tech giants.
This gap arises because technological innovation requires high-intensity capital investment and a relaxed R&D environment. The core businesses of NIO, Xpeng and Li Auto face fierce competition, lack sufficient self-sustaining profitability, and the capital market has low tolerance for their long-term high-risk exploration, which erodes their competitive advantages in capital, talent and resources.
More critically, the embodied intelligence sector does not have a mature industrial chain, requires massive technological investment, and offers very low fault tolerance for strategic choices. This means founders with solid technical expertise are better able to predict the medium and long-term trends of the industry and improve their success rate.
This also explains the core reason why NIO, Xpeng and Li Auto have stepped into the same trap twice: their founding teams face insurmountable generational limitations in the face of the technological evolution cycle.
/ 01 / Becoming a startup company again: from vehicle manufacturing to embodied intelligence
There is a clear contrast between the operations and investment arrangements of NIO, Xpeng and Li Auto.
On the operational side, none of the three brands have entered a completely safe zone. As internal competition in the new energy vehicle sector intensifies, Li Auto, the most profitable among the three, has recorded losses in two of the last three quarters; NIO and Xpeng, established over a decade ago, have still not achieved full-year profitability to date.
Yet on the investment side, the three companies do not behave like businesses fighting for survival at all. Their business layouts are still expanding, covering new energy vehicles, self-developed chips, autonomous driving large models and other fields. To this end, Li Auto's R&D expenditure in 2025 reached 10 times its profit, and Xpeng's R&D investment also rose by 47% year on year.
This contrast between operations and investment stems from the fact that new EV startups have launched their second venture, with their core positioning shifting from vehicle manufacturing to embodied intelligence. Li Xiang stated directly that the future of Li Auto is to become an embodied intelligence company, while Xpeng claims that "future AI automotive companies will eventually evolve into robotics companies".
To achieve this, He Xiaopeng has laid out a series of plans including the second-generation VLA, Robotaxi, humanoid robot IRON and flying vehicles, benchmarking against Tesla with the positioning of "I have everything you have, and you may not have what I have".
Li Auto has not only made AI-related layouts, but also adjusted its R&D structure, breaking the traditional division of software and hardware departments and reorganizing into a "visceral system" responsible for chips, datasets and operating systems, as well as a "brain system" covering perception, pre-training and reinforcement learning, among other divisions.
The transformation of new EV startups to embodied intelligence is not hard to understand. The electric vehicle sector has already shifted from a blue ocean to a red ocean, and NIO, Xpeng and Li Auto have largely lost their competitive edge against traditional automakers.
In the first half of this year, BYD's new energy vehicle sales reached 3 times the combined total of NIO, Xpeng and Li Auto, and Geely's new energy vehicle sales also exceeded the sum of the three brands. With a much larger scale base, traditional automakers still lead in growth: Geely's new energy segment sales rose 10% year on year, while Li Auto and Xpeng's sales in the same period fell by 5.1% and 15.8% year on year respectively.
In a mature market where the three brands lag in scale and lack stable profitability, the new energy narrative can no longer support their valuations, and their share prices have all dropped by 70% from their respective all-time highs.
Embodied intelligence has become the new valuation anchor that new EV startups are seeking. In their strategic statements, automobiles are classified as part of the next development stage of the AI industry. Li Xiang noted that vehicles are not just transportation tools, but robots operating in the physical world with continuous perception and real-time decision-making capabilities.
This repositioning has expanded the business boundaries of new EV startups.
If R&D remains limited to the "electric vehicle" dimension, competition will degenerate into a war of attrition over technical parameters, such as competing for a 20km increase in range or a 2cm adjustment in vehicle length. But if vehicles are regarded as the implementation carrier of physical AI, new EV startups will gain access to a market with extremely high growth potential: by 2050, the global humanoid robot market alone will reach a size of 7.5 trillion US dollars.
The new positioning has also changed the valuation model. Citi adopted the Sum of The Parts (SOTP) method for Xpeng, assigning a 1.5x 2026 forecast P/S multiple to its automotive business, and a far higher 8x forecast P/S multiple to its robotics business. Similarly, Citi adjusted its valuation approach for Li Auto from a single P/S ratio to SOTP to reflect the potential of its embodied intelligence business.
However, the SOTP valuation adopted by institutions for new EV startups is more of a bet on high odds than a confirmation of certainty.
/ 02 / Stepping into the same river for the second time
"The difficulty of starting a humanoid robot business is 20 to 100 times that of starting an automotive business, and Xpeng's own success rate is around 20%." This is He Xiaopeng's judgment on his company's success rate in the embodied intelligence track.
In horizontal comparison, the valuations institutions assign to the robotics businesses of new EV startups do not carry a premium. Institutions including Huatai and Citi value Xpeng's robotics business at 96 billion to 26 billion yuan, the highest among new EV startups, while Unitree is valued at 42 billion yuan, and Galaxy Universal, Star Navi and Qianxun Intelligence all have latest valuations exceeding 20 billion yuan.
The lack of obvious valuation premium is because new EV startups do not have clear competitive advantages in the embodied intelligence sector.
The core competitive focus of embodied intelligence can be observed from capital flows: nearly 440 billion yuan of capital poured into China's embodied intelligence track in the first half of 2026, more than half of which went to "brain faction" companies with strong model capabilities. After all, in the technology race, when hardware is no longer a scarce resource, parties that master model capabilities (thinking) are more likely to capture the premium. KPMG and Shanghua Capital have also expressed similar views that "a strong brain can even make up for mediocre hardware".
In terms of model capabilities, startup companies demonstrate stronger technical strength than NIO, Xpeng and Li Auto. For example, in February this year, UBtech open-sourced the lightweight embodied model Thinker, and official self-test results show that it ranks first in the 10B parameter group in 9 indoor task benchmark evaluations for robots.
In addition, according to LatePost reports, the ultimate question investment institutions raise for model startups is often how they compete with the oligarchic leading private model players such as Google, Alibaba, ByteDance and Anthropic.
In other words, in the eyes of institutions, the model capabilities of new EV startups lag behind those of large tech firms and specialized AI startups.
This disadvantage first stems from objective physical constraints. Model capabilities are highly dependent on data input, but the macro road data accumulated from autonomous driving has a natural "Domain Gap" with the desktop-level fine operation data required by humanoid robots, and cannot be simply reused.
On the technical level, innovative technologies require both massive investment and a favorable innovation environment. Both OpenAI and Physical Intelligence fit the typical frontier innovation pattern: a group of highly talented, visionary people gather and continuously invest in a direction with huge potential but long-term uncertainty.
The primary market environment also provides relatively good early incubation conditions: from its establishment in 2015 to the release of ChatGPT at the end of 2022, OpenAI actually did not make obvious commercial progress, but that did not stop it from attracting massive talent and resources. China's Deepseek currently also does not take profitability as its core goal.
In contrast, NIO, Xpeng and Li Auto, as listed companies trapped in the red ocean market with their core businesses mired in price wars and lacking stable positive cash flow, face much stricter financial constraints. As public companies, the capital market's tolerance for their long-term uncertain exploration may not last long. All these factors will indirectly put them at a disadvantage in attracting resources, talent and capital.
The lag in scenarios and technology is only an external manifestation. The core factor that makes it difficult for NIO, Xpeng and Li Auto to succeed in embodied intelligence is the insurmountable generational limitation of their founding teams in the face of the technological evolution cycle.
/ 03 / Each generation has its own limitations
NIO, Xpeng and Li Auto are gradually falling behind in the new energy vehicle race and facing multiple obstacles in the embodied intelligence track. There are many reasons behind this, but if the analysis does not eventually point to the three founders, all conclusions are no more than partial observations.
Li Bin was born in 1974, He Xiaopeng in 1977, and Li Xiang in 1981. They all grew up in the era when the PC Internet evolved into the mobile Internet. Li Bin founded Bitauto, He Xiaopeng developed UC Browser, and Li Xiang became famous through Autohome. All of them are winners of the two golden decades of the Internet industry.
However, when this group of entrepreneurs with deep-rooted Internet thinking habits cross over into the "physical world" sectors of new energy vehicles and embodied intelligence, they are all constrained by their inherent generational background and cognitive boundaries:
The core competencies of NIO, Xpeng and Li Auto, with their Internet origins, lie in traffic operation, product definition, user insight, capital financing and Internet marketing.
This allows them to excel at identifying obvious contradictions and quickly create hit products and attractive narratives. In the blue ocean period of the new energy vehicle sector, NIO, Xpeng and Li Auto became the catfish driving the new energy wave relying on product innovation and marketing. In the AI era, Li Auto and Xpeng have already obtained the embodied intelligence premium through narrative building even before launching mass-produced products.
But once the industry enters the red ocean stage, the limitations of NIO, Xpeng and Li Auto become prominent. During the vehicle manufacturing phase, they are good at integrating supply chains and defining complete vehicle products, but their in-house R&D depth for the three core electric systems (battery, motor, electronic control) is weaker than that of BYD.
In contrast, Wang Chuanfu and Li Shufu are the first generation of local industrial manufacturing entrepreneurs. They grew up in an era when China's industrial supply chain was extremely underdeveloped, which made them believe from the very beginning in the independent controllability of core components, and hold natural respect for scale effects and manufacturing efficiency. When the market turned into a red ocean, BYD and Geely completely pulled ahead of new EV startups in sales and profitability relying on deep in-house R&D of core components and extreme cost and scale advantages.
Entering the embodied intelligence era, the dominance of the track is shifting to a younger generation.
Wang Xingxing from Unitree, Peng Zhihui from Fourier Intelligence, and Chen Jianyu from Era Robotics are all post-90s tech entrepreneurs. They are the first generation of engineering academic natives who grew up in the period of rapid development of robotics disciplines in universities, the late stage of mobile Internet, and the outbreak of general AI. Their cognitive systems are built on robotics engineering and underlying AI algorithms.
Embodied intelligence does not have a mature industrial chain, requires massive technological investment, and offers very low fault tolerance for strategic choices. This industrial feature determines that founders with solid technical expertise are better able to predict the medium and long-term trends of the industry and improve the success rate of their bets.
Each generation has its own opportunities, and the beneficiaries of the previous generation's dividends can hardly become the navigators of the new era. This is not a failure in the strategic vision of the three founders, but an insurmountable fate under the iteration of industrial paradigms.
This article is from the WeChat official account "Read Finance", written by Yang Yang, and published with authorization from 36Kr.