The end of XPeng's physical AI is Elon Musk's cash flow.
The market trend of XPeng has come to a temporary halt.
On November 12, XPeng's US-listed stocks fell by 2.81%. In the previous four trading days, affected by the new product information released at XPeng's Technology Day, XPeng's stock price had risen by about 29% in total, and its market value exceeded that of Geely Automobile.
This correction has temporarily cooled the recent market's technological frenzy for XPeng.
At XPeng's Technology Day on November 5, XPeng's IRON humanoid robot sparked heated discussions. To prove that the product was not a person in disguise, XPeng cut open the robot's "clothes" to expose its mechanical structure. He Xiaopeng even choked with emotion on the spot: "There are always people who don't believe that the world is changing so fast... I hope this is the last time we need to prove that the robot is truly itself."
At the event, XPeng announced four important innovative initiatives - the second-generation VLA architecture, XPeng Robotaxi, the new-generation IRON robot, as well as the land aircraft carrier and flying car. He Xiaopeng publicly stated that by 2026, XPeng's high-order humanoid robots will enter the mass production stage, and XPeng will be upgraded to a physical AI explorer and become a global embodied intelligence company.
This press conference is pushing XPeng's positioning from a new energy vehicle manufacturer to a Tesla-style physical AI enterprise. A person close to the industry told "Market Insights": "The positioning of XPeng in the capital market is increasingly looking like a mini-Tesla."
On the other hand, the third-quarter position information of JPMorgan Chase recently showed that in the third quarter, JPMorgan Chase reduced its holdings of 1.8239 million XPeng US-listed stocks, leaving only 6,204 shares, a quarter-on-quarter decline of 99.7%. During the same period, NIO and Li Auto also suffered significant reductions.
From the perspective of price-to-sales ratio, as of November 13, Tesla's ratio was around 15 times, while XPeng's was 2.54 times. The gap between the two is still huge. Besides the fact that Tesla achieved annual profitability as early as 2020 and XPeng is still struggling to achieve break-even, there is also a large gap in their scales. In the first ten months of 2025, XPeng delivered a total of 355,209 smart electric vehicles, a 190% increase from the same period last year. However, Tesla's financial report showed that in the third quarter alone, it produced 447,000 pure electric vehicles globally and delivered up to 497,000 vehicles. The two are competing on different levels.
This may be the weakness of XPeng's transformation into a physical AI enterprise - what supports a giant is not only technology but also the basic automotive business and cash flow that support the technology. The market cares about how many good stories you have, but it also cares more about how much resources you can pour into these stories.
01 It's Not Easy to Be a Mini-Tesla
How long does it take to fill a pool when water is being injected on one side and drained on the other?
This is a classic type of math problem. If the layout of embodied intelligence and full-stack self-research is the drainpipe for an automaker, then the automotive business is the water inlet pipe. Despite the grand goals of new businesses, the automotive business remains the core indicator for the market to measure XPeng at present.
XPeng's voluntary announcement showed that in October this year, XPeng delivered a total of 42,013 smart vehicles, a 76% year-on-year increase and a 1% month-on-month increase. In that month, XPeng set a new monthly delivery record. If we look at the first ten months of this year, XPeng has delivered more than 355,000 new vehicles in total, a 190% increase from the same period last year.
The hidden danger lies in the structure. Public data showed that the main sales driver for XPeng is the XPeng MONA M03, which has a starting price of 119,800 yuan and competes with BYD Qin L in the same price range. Since June this year, the monthly sales of the XPeng MONA M03 have remained above 14,000 units, accounting for about 40% of XPeng's monthly sales. XPeng's announcement showed that the XPeng MONA M03 achieved cumulative deliveries of more than 200,000 units in just 14 months.
During the same period, the sales of some of XPeng's mid-sized cars declined. The monthly sales of the XPeng P7+, priced at around 200,000 yuan, dropped from over 10,000 units in December last year to 5,568 units in October this year; the sales of the XPeng G7 dropped from nearly 7,000 units in August this year to 3,463 units; and the sales of the G9 dropped from over 3,500 units in May this year to less than 1,000 units in October.
The X9 is XPeng's main model targeting the high-end market, but its sales have been declining continuously this year. In May this year, the XPeng X9 sold 2,767 units, while in October, the figure was only 835 units. However, XPeng still insists on targeting the high-end market. Recently, XPeng held a press conference for the X9 Kunpeng Super Extended-Range Technology, announcing that it will use the next-generation extended-range solution, Kunpeng Super Extended Range, to solve the range anxiety problem. This technology will be first launched on the super extended-range version of the XPeng X9, with a starting price of 350,000 yuan. The pre-orders on the first day exceeded 5,000 units.
During the same period, Tesla's sales also declined. According to the data from the Passenger Car Association, Tesla China's wholesale sales in October were 61,497 units, a 32.28% month-on-month decline; and the retail sales were 26,006 units, a staggering 63.64% month-on-month decline.
The reason lies in the intense competition in the price range of 200,000 to 300,000 yuan, where models such as the BYD Seal, XPeng P7, Xiaomi SU7, and Tesla Model 3 are all competing.
However, there are still obvious differences in their financial data.
Tesla achieved profitability as early as 2020. In the first half of this year, Tesla's revenue was $41.83 billion, and its net profit was $1.581 billion. In contrast, XPeng's total revenue in the first half of the year was 34.09 billion yuan, and it had a loss of 1.14 billion yuan.
Facing the pressure of losses, Li Bin of NIO next door has repeatedly issued orders, demanding that all employees work hard to ensure profitability. He Xiaopeng has also made a commitment. As early as at the 2024 earnings conference in March this year, He Xiaopeng said that XPeng is expected to achieve profitability in the fourth quarter of 2025.
On the other hand, XPeng's physical AI business still needs further investment.
At the beginning of this year, He Xiaopeng said in an interview with the Securities Times that XPeng may need to work on its humanoid robot for another 20 years, and the investment scale may reach 50 billion yuan or even over 100 billion yuan: "At present, we are still in the early stage, and the investment will be relatively conservative. It is very difficult for humanoid robots to leap from Level 2 to Level 3."
XPeng's R & D expenses have also increased sharply. In 2024, this figure was about 6.5 billion yuan, and He Xiaopeng revealed that this year, XPeng's R & D expenses are about 10 billion yuan: "A lot of R & D expenses are spent on software, not just hardware." To achieve profitability in the fourth quarter, XPeng must do better in areas such as increasing the share of high-end models and controlling costs.
02 Full-Stack Self-Research Isn't Cheap
In XPeng's physical AI landscape, the self-developed Turing chip is the core computing power base. He Xiaopeng once publicly declared: "One self-developed Turing chip can be as powerful as three traditional chips."
This chip was successfully taped out in August last year and began mass production in the second quarter of this year. It is claimed to be able to achieve 100% utilization of computing power and can be applied not only in the field of smart cars but also in robots and flying cars.
On the positive side, self-developed chips are a common choice among NIO, XPeng, and Li Auto. It can not only solve the risk of supply disruptions and relieve geopolitical pressure but also promote the narrative of full-stack self-research and raise market expectations.
On the negative side, self-developed chips are an expensive investment, and the claim that self-developed chips can save costs is a story that needs a long time to prove.
A person close to the industry told "Market Insights" that they do not have high hopes for XPeng's self-developed chips in the short term because it is extremely difficult to recoup the investment. Not only does a single tape-out require tens of millions of US dollars, but to reach the 5-nanometer process, XPeng also needs to queue up with TSMC. Moreover, a large amount of inventory needs to be stocked, which requires a large-scale mass production of products to spread the cost.
The high-computing-power AI chips self-developed by automakers are also difficult to become a common solution in the industry. On one hand, according to public reports, the price of NVIDIA's Orin chip is around $500, and the delivery cycle has been reduced to four weeks. It has formed a monopoly-like advantage in the intelligent driving market and has become the standard for automakers to move towards high-order intelligent driving.
On the other hand, from the perspective of external sales, it is also difficult for automakers' self-developed chips to compete with established suppliers. Elon Musk once mentioned in Tesla's earnings conference that Tesla's chips are only for Tesla's own use. Because they are designed for a single customer, Tesla can have a huge advantage in chip optimization and eliminate unnecessary problems. The premise is that Tesla has a larger mass production scale, better market prospects, and more confidence in spreading the cost.
XPeng's solution is to try to spread the R & D cost through its physical AI business and external sales.
Recently, Gu Hongdi, the vice chairman of XPeng, said in an interview that compared with Tesla's model, XPeng's ecosystem is more open and inclusive. In the future, it will package its intelligent driving, chip, and Robotaxi capabilities to attract external customers. At XPeng's Technology Day on November 5, He Xiaopeng announced that the second-generation VLA intelligent driving system and the Turing AI chip have been selected by Volkswagen, and Volkswagen has become the first strategic partner. In addition, according to South Korean media reports, Hyundai Motor is in talks with XPeng about cooperation in autonomous driving technology, focusing on the VLA model and the Robotaxi solution.
Obviously, XPeng is quite lucky. Among NIO, XPeng, and Li Auto, this is the first time an automaker has announced that its self-developed chips have won an external sales contract. However, Volkswagen is not an outsider. As a shareholder of XPeng, Volkswagen has long been seeking technological localization through XPeng. The two have jointly produced various models. In the fierce competition of the domestic electric vehicle market, XPeng's technology can help Volkswagen accelerate its electrification and intelligent transformation.
However, XPeng faces resistance in obtaining more contract partnerships. Currently, Huawei is providing brand output to traditional automakers through the smart selection model, and Momenta has the ability to provide intelligent driving solutions with fast delivery. They are regarded by the market as the Apple and Android of intelligent driving solutions respectively.
XPeng's Turing chip + second-generation VLA model solution will face delivery challenges in the future. The performance of the first batch of Volkswagen models equipped with XPeng's solution will determine how far XPeng's model can go.
03 Entering the L4 Market
Another major move by XPeng is to enter the Robotaxi market. As early as at the second-quarter earnings conference this year, He Xiaopeng announced that XPeng will mass-produce Robotaxi models in 2026 and launch operational services in some areas. On November 5, XPeng announced again that in 2026, it will launch three dedicated electric vehicles for Robotaxi and plans to start trial operations within the year.
Compared with other Robotaxi companies, XPeng is obviously more cautious in its actions.
Firstly, it invites partners to join. According to XPeng's plan, in the initial testing stage, XPeng will be responsible for the operation. If the test results are not satisfactory, it will promote commercialization together with its partners.
In this experiment, Amap will be an important partner for XPeng, responsible for providing user access and traffic data to help XPeng solve the problem of user acquisition.
Secondly, it opens up the development toolkit. This means that XPeng is not mainly focused on operating a heavy-asset self-owned fleet. Instead, it positions itself as a technology provider. Automakers can install XPeng's Robotaxi solution as a component and quickly launch their own Robotaxi services and build their own Robotaxi brands. XPeng can then collect technology licensing fees and service fees or drive the overall vehicle sales revenue through the sales of L4 models.
In other words, XPeng's goal is not to directly earn operating fees but to try to be the Android in the Robotaxi field and make money by selling the operating system. In an interview, He Xiaopeng said that XPeng does not pursue self-operation but is like a toolbox. It has cars, software, and SDK interfaces. With this "toolbox", one can operate a local Robotaxi service:
"You can contract hundreds of Robotaxis in your hometown. I think this will be a business opportunity."
After all, if XPeng were to operate on its own, the probability of success would be extremely low.
The Robotaxi market is known for its long R & D cycle and difficult profit returns. As early as in 2019, Elon Musk claimed that he would build Robotaxi vehicles within a year and deploy one million of them. However, until now, Tesla's Robotaxi fleet is still relatively small, and its operation is mainly concentrated in the Austin area of the United States. At the third-quarter earnings conference, Musk was evasive about the fleet size, and the ambition of a million-vehicle fleet seems to have disappeared.
From the experience of the market, the competition in the Robotaxi industry is currently fierce. Robin Li has publicly stated that in the future, taking a driverless car will be half as cheap as taking a taxi. Elon Musk said that the