A second act, from cars to bodies
On 29 July, reports said Li Auto was planning two-wheeled and bipedal robots, and its shares rose 10 percent that day. XPeng’s own share rallies have likewise tracked robot news. The pattern shows a shift in the underlying story: the three newcomers, NIO, XPeng and Li Auto (collectively “Wei-Xiao-Li”), are starting a second act, repositioning from electric-car makers to embodied-AI players.
As Li Xiang puts it, a car is not just transport but a robot that runs in the physical world with continuous perception and real-time decisions. That reframing changes the valuation logic. In a maturing, slowing EV market, newcomers without scale or profit can no longer lean on the new-energy story. Embodied AI drops them into a far higher-growth lane. By 2050, the humanoid robot market alone could reach 7.5 trillion US dollars.
The new valuation math
The new logic is already in research notes. Citi moved Li Auto and XPeng from a single price-to-sales multiple to a sum-of-the-parts (SOTP) model, giving XPeng’s car business 1.5x sales and its robot business 8x sales. But institutions adopting SOTP look more like they are betting on high odds than buying certainty. The newcomers that lost on EVs may repeat the mistake on embodied AI.
The reason: large models decide the contest, and the newcomers’ model capability trails both the leading AI startups and the tech giants. The gap comes from money and freedom. With their core business in a price war and no self-funded cash flow, the market tolerates long, risky exploration poorly, eroding their edge in talent and capital.
From building cars to building brains
The operating picture is uneven. Among the three, the most profitable, Li Auto, posted losses in two of the last three quarters, while NIO and XPeng have never turned an annual profit after more than a decade. Yet spending keeps widening across chips, autonomous-driving models and robots. Li Auto’s 2025 R&D hit ten times its profit; XPeng’s R&D rose 47 percent.
The pivot is explicit. Li Xiang calls the company’s future an embodied-AI firm; He Xiaopeng says an AI car company will become a robot company. He Xiaopeng has stacked plans for a second-generation VLA, Robotaxi, the IRON humanoid and flying cars. Li Auto rebuilt its R&D around “organ” and “brain” systems for chips, datasets and operating systems.
Why the same river, twice
The EV math is brutal. In the first half of 2026, BYD’s NEV sales were three times the three combined; Geely also exceeded them. With larger scale, Geely’s NEV sales still grew 10 percent while Li Auto and XPeng fell 5.1 and 15.8 percent. Having lost the car race, embodied AI becomes the new valuation anchor.
But institutions price the robot units without a premium. Huatai and Citi value XPeng’s robot business at 9.6 to 26.6 billion yuan, the highest among newcomers, yet Unitree is worth 42 billion and Galbot, Star Era and Qianxun all top 20 billion. The gap shows no clear edge. In the first half of 2026, some 44 billion yuan flowed into embodied AI, over half to “brain” camps, because when hardware is no longer scarce, the side that owns the model takes the premium.
A generation’s limits
Wei-Xiao-Li trailed in EVs and faces walls in embodied AI, but the root cause lands on the three founders. Li Bin (1974), He Xiaopeng (1977) and Li Xiang (1981) grew up as the internet moved from PC to mobile, winners of the golden twenty years. Their core strengths are traffic, product definition, user insight and financing.
That made them sharp at catching visible conflicts and building hits during the EV blue ocean. But in a red ocean, their limits show: they integrate supply chains well yet lack BYD’s depth in core components. Wang Chuanfu and Li Shufu, first-generation industrialists, believed in self-reliance and scale from the start, and pulled away on cost. In embodied AI, leadership is shifting to a younger cohort: Unitree’s Wang Xingxing, AgiBot’s Peng Zhihua and Star Era’s Chen Jianyu are all post-90s engineers. When there is no mature supply chain and little room for error, the technically fluent founder predicts the trend better. Each generation has its own window; yesterday’s winners rarely steer the next one.
Editor’s note: This article is based on reporting by OFweek Robot. Read the original in Chinese here: https://robot.ofweek.com/2026-08/ART-898890-12003-30697259.html.