XPeng (3Q25 Minutes): Launching seven 'dual-energy' models by 2026

The following is a summary of the XPeng (XPEV.US) FY25 Q3 earnings call minutes organized by Dolphin Research. For an interpretation of the financial report, please refer to "XPeng: The 'Eastern Tesla' Story Unfolds, Is Selling Cars Still Important?"

1. Review of Core Financial Data

Performance Overview:

Deliveries and Revenue: Total vehicle deliveries reached 116,007 units, a year-on-year increase of 149%. Total revenue reached RMB 20.38 billion, up 101.8% year-on-year.

Improved Profitability: Gross margin exceeded 20% for the first time, reaching 20.1%. Net loss significantly narrowed to RMB 380 million.

Cash Position: Cash and equivalents remained ample, reaching RMB 48.33 billion by the end of the quarter.

Expenditure: To support future growth, R&D expenses increased by 48.7% year-on-year to RMB 2.43 billion; selling and administrative expenses rose by 52.6% year-on-year to RMB 2.49 billion, mainly driven by increased commissions and marketing expenses due to higher sales.

Segmented Revenue: Automotive sales revenue was RMB 18.05 billion, up 105% year-on-year. Service and other revenue was RMB 2.33 billion, up 78%, mainly due to increased service revenue from the Volkswagen Group.

2. Detailed Information from the Earnings Call

2.1 Key Information from Executive Statements

Strategy and Outlook:

The company aims to achieve breakeven in the fourth quarter. Over the next decade, the goal is to establish XPeng as a leading global entity in the field of physical intelligence. The company focuses on applications of physical AI, building a complete system encompassing technology, products, and supporting business ecosystems. As AI models deeply integrate with real-world data, machines are gradually acquiring the ability to interact, communicate, transform, and create in physical environments, reshaping future mobility and daily life.

Product Progress and Outlook:

Vehicle Products: The newly launched XPeng P7 quickly ranked among the top three in sales of pure electric sedans in the RMB 200,000 to 300,000 range, with monthly deliveries exceeding 40,000 units since September. The XPeng X9 Super Long Range version began pre-sale on November 6, with a comprehensive range of 1,602 kilometers, and pre-orders have nearly tripled those of the previous generation X9; the model will be officially released on November 20, with deliveries starting immediately, expected to set a new delivery record in December. Three super range-extended products will be launched in the first quarter of 2026, focusing on extending pure electric range and enhancing 5C supercharging speed to address range extender user pain points. In 2026, four new dual-energy models will be introduced, including the first products in several key segments. The seven dual-energy models to be launched next year will significantly expand the TAM and create substantial sales growth opportunities.

Robotaxi:

Three ride-hailing models are planned for launch in 2026. The autonomous driving technology stack does not rely on HD maps or LiDAR, effectively addressing industry pain points such as high costs, scene limitations, and insufficient generalization capabilities. The company plans to pilot XPeng Robotaxi operations in China in 2026, while continuously optimizing Robotaxi software and hardware and building an operational ecosystem. Volkswagen Maps will be the first ecosystem partner for XPeng Robotaxi.

Humanoid Robots:

The new generation Iron robot was unveiled at the recent XPeng Tech Day. The goal is to start mass production of advanced humanoid robots by the end of 2026. Once mass-produced, Iron will first be deployed in commercial scenarios, providing services such as guidance, retail assistance, and patrolling. By the end of next year, Iron is expected to work as a new member in XPeng stores, campuses, and factories. XPeng Robotics will open the STK platform to global developers, inviting cross-industry partners to participate in secondary development. By 2030, XPeng Robotics aims to achieve annual sales of over one million units.

Technological Development:

VLA 2.0 Model: The upcoming VLA 2.0 model will have ten times the parameters of its predecessor, significantly enhancing the safety and user experience of intelligent driving. A co-creation plan with early users will begin in late December, with plans to fully deploy the VLA 2.0 model across the Ultra series in the first quarter of 2026. The mass production of VLA 2.0 marks a major breakthrough in the field of physical AI models. Volkswagen will be the first customer of the VLA 2.0 model.

Turing AI Chip:

XPeng's Turing AI chip has received official procurement certification from Volkswagen, and the jointly developed models are expected to begin production early next year. Revenue from external technology licensing will continue to be invested in R&D, primarily for iterative upgrades of the Turing chip and BLA model.

Global Business:

Maintaining strong sales growth through localization strategies. In September 2025, overseas monthly deliveries exceeded 5,000 units for the first time, a year-on-year increase of 79%. In the third quarter, 56 new overseas stores were added, expanding the sales and service network to 52 countries and regions worldwide. The first European localized production base officially started production in Graz, Austria, with the first batch of XPeng G6 and G9 models rolling off the line. The Munich R&D center in Germany has officially started operations, helping to better understand overseas user needs. Three new overseas models are planned for launch in 2026, including popular small and medium-sized SUVs catering to diverse global consumer preferences.

Financial Guidance:

With the launch of the dual-energy product cycle, total deliveries in the fourth quarter are expected to reach 125,000 to 132,000 units, a year-on-year increase of 36.6% to 44.3%. Fourth-quarter revenue is expected to reach RMB 21.5 billion to 23 billion, a year-on-year increase of 33.5% to 42.8%. XPeng's AI smart car business is in the early stages of rapid expansion in scale and market share, while the Robotaxi and humanoid robot projects are accelerating towards mass production.

2.2 Q&A Session

Q: What are XPeng's long-term competitive advantages in physical AI, and how does the company plan to continuously enhance its strengths in these areas?

A: Traditional car companies and the new physical AI model have different business model DNA. Traditional car companies focus on positioning, user segmentation, and integrating tier-one supplier supply chains. The physical AI model relies on defining future technologies, involving full-stack technical capabilities and customized integration, such as the release of the Iron robot. Future cars will be a new form of robotics, becoming a reality in the next 5 to 10 years.

The integration of traditional supply chains and the current cross-domain integration of physical AI technology (involving software, hardware, and infrastructure upgrades) are completely different. In traditional car development, software accounts for a small proportion, whereas now software occupies a large part of new product development. In future car development, the proportion of physical AI components will soon exceed 50%.

Q: When will revenue related to the cooperation with Volkswagen, especially the Turing chip, start to be generated? What is the outlook for revenue contribution in the fourth quarter of 2025 and the entire year of 2026?

A: In the third quarter, we completed several key R&D milestones on schedule, and technical cooperation revenue increased significantly quarter-on-quarter. We expect fourth-quarter technical cooperation revenue to be flat with the third quarter. Our Turing SoC has been selected by Volkswagen for two jointly developed B-class cars, and we have begun supplying chips to pre-production validation vehicles for partners. Therefore, Turing SoC revenue is expected to start being recognized on a small scale from the fourth quarter. As the jointly developed vehicles begin production early next year, Turing SoC revenue will grow with the sales of these two cars. For the entire year of 2026, as long as the key milestones for the year are completed as planned, we expect the overall scale of technical cooperation revenue to be comparable to that recognized in 2025. We will reinvest revenue from technology licensing and cooperation into R&D.

Q: Can you discuss the technical roadmap for humanoid robots and compare it with peers? Where are our medium- to long-term competitive advantages?

A: XPeng Robotics' technical roadmap and product development are progressing according to its own plan. Our product philosophy is highly anthropomorphic, aiming to develop highly human-like robots. Our robots integrate muscles and bionic skin, attracting people to embrace them, which is different from traditional robots.

Many robots currently on the market are in the 3rd to 5th generation, mainly driven by joints and hardware operations, with similar walking and movement methods, making it difficult for these robots to be commercialized. The robots demonstrated by XPeng are based on our full-stack R&D capabilities and cross-domain integration, resulting from the integration of hardware and software driven by integrated AI.

XPeng has several advantages in robot development: our physical AI resources and AI automotive business can create synergies; we consider providing humanoid robots with potentially higher performance than automotive-grade; our knowledge and expertise in the electric vehicle industry drive the commercialization and mass production of robots; in terms of future sales network layout and globalization, synergies can be created with the existing automotive business. In the future development of robots, some players will come from the automotive manufacturing industry, and XPeng, with its data, SoC chips, and capabilities, will undoubtedly have a first-mover advantage.

Q: What are the key milestones for the commercialization of humanoid robots? What are the capacity and operational scale expectations by the end of 2026, and what are the long-term application scenarios under the goal of one million units by 2030?

A: The mass production of the Iron robot is the most challenging project I have experienced. Our ultimate goal is to enable it to be easily trained through human language, which requires a large amount of capability integration, such as achieving walking or running in various safe postures, involving joint embedded management and harness coupling; achieving more general dexterous hand movements, which requires hand-based VLA capabilities, expected to be integrated early next year; and having the ability to communicate with humans based on language, derived from VLM and VLT model capabilities.

We will enter the 1.0 phase of the new generation mass production model next month and believe that we can steadily advance the mass production process within the next 10 months. Increasing robot capacity is much simpler than cars, but commercialization is extremely challenging. We hope to first deploy in several commercial scenarios such as guidance and retail assistance. In 2026, we expect to see robots conducting on-site tests in XPeng stores and campuses. At the same time, we will open the SDK to partners so they can purchase robots and conduct commercial training.

Regarding the potential future application scenarios, such as commercial robots being able to change "skins" to enter industrial production scenarios. Robots entering the home environment may have the opportunity to be realized in 5 years. By opening the SDK, we hope to work with partners to tackle diverse, long-tail application scenarios and build a more complete ecosystem.

Q: Why choose to launch the Robotaxi service in 2026? What is the technical inflection point? How fast is the cost declining? Compared to other domestic companies, what are XPeng's technical path, business model, and advantages?

A: Our R&D strategy is based on full-stack self-research and cross-domain integration. We expect a series of inflection points to appear within our own development system in 2026. For example, we will be able to configure existing models into the Robotaxi fleet, and the inflection point will arrive. At the same time, our VRL model will continue to provide new capabilities for future vehicles, making them more like robots. In addition, our second-generation VLA model can train our intelligent driving Ultra models and may also be used to train Volkswagen models with similar large models in the future.

Based on the cross-domain capabilities obtained from robot development, we can effectively solve many current limitations of Robotaxi, such as high production costs and restrictions on driving destinations. Current Robotaxi cannot handle very complex road conditions and a large number of unpredictable scenarios in residential areas well, and many solutions rely on LiDAR for perception.

In 2026, we hope to achieve the dual-track development of unmanned L4 models and assisted driving L4 models by commercializing fully unmanned L4 capabilities in Robotaxi. With the launch of these two methods or roadmaps, XPeng's superior business logic compared to other Robotaxi companies will soon be verified, which will bring us a huge competitive advantage.

Q: How does management view the commercialization of the Robotaxi business? What are the future planned milestones, such as fleet size, launch plans in different cities or overseas markets? What are the details of the cooperation with Volkswagen Maps, and are there plans to cooperate with other mobility platforms in the future?

A: XPeng will launch three Robotaxi models at different price points next year to meet different travel needs. In the next stage of development, with regulatory approval, our priority is to ensure the smooth operation of the entire technology, operation, and business model. Therefore, we hope to cooperate with more commercial partners in the ecosystem. For example, Volkswagen Maps will be an important partner, providing us with more development support in areas such as traffic, payment, operation, and services, which distinguishes us from many autonomous driving companies.

In the future, for different countries, regions, and different stages of development, we will establish more partnerships with service providers in different links. For XPeng, what we need to do is to build our toolbox and open interface capabilities so that we can cooperate with more ecosystem partners in more countries and cities in the future. Once we successfully achieve commercial operations in different environments, we can quickly build our ecosystem.

Q: Do we have any rough revenue estimates or breakdowns for new businesses such as Robotaxi and humanoid robots in the next 1-3 years?

A: For these future development areas, we currently do not provide any numerical guidance. However, all three areas are expected to achieve scaled production and operation within the next 12 months. For example, our flying car company's land carrier target will be delivered to end customers by the end of next year, reaching a scale of thousands of units. As for humanoid robots and autonomous driving Robotaxi, a large number of operational tests and capacity enhancements will be conducted next year to prepare for mass production and use.

Therefore, the contribution of these businesses next year may be limited. But once the product models and stability are verified on the consumer and application sides, production is expected to increase rapidly. Achieving annual sales of one million humanoid robots by 2030 is our long-term goal, and we are confident in this. The potential of these future fields is enormous, but it is currently impossible to provide exact breakdowns or precise cost estimates, as these are still evolving.

Q: Can you provide more details about the new models, including their segments and price ranges, and the sales targets for 2026?

A: We believe that both single-power models and dual-energy models present highly attractive opportunities and are strategic initiatives to expand the sales of each model. On November 20, we will announce the price of the X9 Super Electric version, which will be our first super electric product. In addition, we have completed the MIIT declaration for three super electric models, which are planned to be launched in early 2026. We will also launch four new models next year, all of which will offer both pure electric and range-extended power options and be positioned in different price ranges, enhancing our product portfolio in each target price range.

For next year's growth, we believe these seven dual-energy models will significantly drive growth. Another growth driver is the international market, with overseas monthly deliveries reaching 5,000 units for two consecutive months in September and October. Among the seven new cars to be launched next year, at least three will enter the international market, so we believe the international market will continue to be a very important growth driver in 2026.

Q: How do you view the growth potential of the new range-extended models in 2026? And the economics of humanoid robots, especially after adopting technologies such as solid-state batteries, can their prices reach a family-affordable RMB 200,000 or lower?

A: Regarding range-extended models, from the sales data of the X9, we found that the actual user groups of pure electric and range-extended versions are very different. We expect that after the delivery of the new version of the X9, it will achieve several times the month-on-month growth, and different customer groups will also have different usage scenarios. Specifically, the user choices for different size models are also different: the proportion of range-extended is higher in large cars, while the proportion of pure electric is higher in A-class, especially small passenger cars. More specific answers need to wait for data from the fourth quarter of this year and the first quarter of next year.

Regarding the affordability of robots, the pricing logic of robots is completely different from that of cars. The cost of our sixth and seventh-generation robots was still high last year, but when preparing for true mass production in the first half of this year, we have every reason to believe that the future retail price of robots can be similar to car prices. Traditional car pricing is based on weight, involving the elements and components such as steel and lithium; while from day one, software accounts for more than 50% of robots, while cars only account for 10% to 20%.

On the other hand, a lot of costs need to be invested in training software and models, and strong integration capabilities and domain controller capabilities are required, such as integrating four SoCs into a super domain controller to achieve as light and cheap as possible. This is still very difficult for many industry players. We hope to handle a limited number of SKU integrations in the future, unlike car manufacturing, which is so complex, and strive to make robot prices as affordable as possible, hoping to truly help and empower thousands of households in the future.

Q: Please introduce the latest progress in overseas localized production next year, and how do we plan to use intelligent driving capabilities to drive sales growth in the international market?

A: Regarding overseas plans, we have established our first factory in Indonesia in the second half of this year and cooperated with Magna to set up another production plant in Austria, with capacity gradually increasing. We expect the output of these two factories to continue to grow next year to support overall overseas growth. In Europe, we expect to achieve tens of thousands of localized production; in Indonesia, the target output is slightly lower, but the scale can still reach a high level of thousands of units. In addition to these two factories, we continue to seek opportunities to establish localized production capabilities in other markets and build local supply chain capabilities. We will increase local content, local store materials, and implement further localization strategies.

Looking ahead to global product sales next year, we expect international market growth to outpace the domestic market and seek higher economic contributions from these markets. Therefore, in the next one to two years, we expect international business to achieve faster growth and higher profit contributions.

Regarding intelligent driving, we will begin deploying the VLA 2.0 model to global users in the first quarter of 2026, with Volkswagen as the first customer of the model. The revenue obtained from technology licensing will be reinvested in R&D, forming a virtuous cycle. We will invite more car manufacturers and tier-one suppliers to jointly promote the R&D of Turing chips and VLA 2.0, working together to promote the widespread application of advanced intelligent technology in China and global markets.