查看完整报告 特别申明： 本订阅号中所涉及的证券研究信息由光大证券研究所编写，仅面向光大证券专业投资者客户，用作新媒体形势下研究 信息和研究观点的沟通交流。非光大证券专业投资者客户，请勿订阅、接收或使用本订阅号中的任何信息。本订阅号 难以设置访问权限，若给您造成不便，敬请谅解。光大证券研究所不会因关注、收到或阅读本订阅号推送内容而视相 关人员为光大证券的客户。 点击注册小程序 1）3Q25小鹏汽车业务收入同比+105.3%/环比+6.9%至180.5亿元（ASP同环比-17.7%/-4.9%至15.6万元， 交付量同环比+149.3%/+12.4%至11.6万辆），汽车业务毛利率同比+4.5pcts/环比-1.2pcts至13.1%。2） 3Q25 R&D费用率同比-4.2pcts/环比-0.2pcts至11.9%，SG&A费用率同环比-3.9pcts/+0.4pcts至12.2%。3） 截至3Q25，在手现金等价物合计约483.3亿元。我们判断，1）3Q25小鹏汽车业务毛利率环比回落主要由于 Mona M03的销量占比偏高+旧车型清库等因素影响。2）我们预计随着X9超级增程版上市，公司月度销量 有望进一步爬 ...

每经记者｜孙磊 每经编辑｜程鹏 余婷婷 11月17日，小鹏汽车披露2025年第三季度财报。受益于汽车销量的显著提升，小鹏汽车的业绩同比与环比均实现增长。 具体而言，小鹏汽车第三季度共交付11.6万辆汽车，同比增长149.3%，环比增长12.4%；实现营业收入203.8亿元，同比增长101.8%，环比增长11.5%；净 亏损收窄至3.8亿元，上年同期净亏损18.1亿元，今年第二季度净亏损4.8亿元。截至2025年9月30日，现金及现金等价物、受限制现金、短期投资及定期存 款为483.3亿元。 "公司经营的持续改善，让我们更加笃定地投入物理AI（人工智能）研发，加速第二代VLA模型、Robotaxi和人形机器人在2026年实现规模量产。"小鹏汽 车董事长何小鹏表示。 欲在四季度实现盈亏平衡 尽管小鹏汽车在营收方面实现了增长，但仍不及市场预期。"本季度（三季度）总收入为203.8亿元，低于市场预期的204.7亿元，低于预期主要是被卖车业 务所拖累。"海豚投研表示，"但服务和其他业务收入本季度达到了23.3亿元，大幅高于市场预期的16.6亿元。" 此外，海豚投研还表示："本季度卖车收入为180.5亿元，大幅低于市场预 ...

上个月，小鹏汽车的市值短暂超过了理想汽车，成为国内市值最高的造车新势力公司。然而「汽车」这 个词，是两家公司都想快速甩掉的。 作者 |新皮层NewNewThing 吴一凡 本文字数：4836，阅读时长大约8分钟 封图 |何小鹏 小鹏想要甩掉它的速度比理想还要快。 2025.11.13 11月5日，小鹏汽车在广州举办科技日，slogan从去年的「未来出行探索者，面向全球的AI汽车公 司」，升级为「物理AI世界的出行探索者，面向全球的具身智能公司」——仅仅是汽车的AI化已不足 以满足这家公司对自身战略和市值的渴望。科技日中，小鹏汽车发布了第二代VLA模型，以及包括 Robotaxi、第二代IRON人形机器人和飞行汽车在内的三大具身智能产品。 何小鹏认为，继互联网、移动互联网、数字AI时代后，数字世界和物理世界的融合将催生「物理 AI」，机器将逐步拥有理解、交互和改变世界的能力。在这个概念下，小鹏汽车的业务涵盖物理AI时 代的技术底座，从模型到芯片，从基础设施到上层终端，比如汽车、Robotaxi、人形机器人和飞行汽 车。 证明了其二代IRON人形机器人里没有真人之后，小鹏美股股价在11月6日盘中一度涨超14%，市 ...

Core Viewpoint - Xiaopeng Motors aims to redefine its identity beyond just an automotive company, focusing on becoming a leader in "physical AI" technology, which integrates digital and physical worlds [5][7]. Group 1: Company Strategy and Market Position - Xiaopeng Motors briefly surpassed Li Auto in market capitalization, becoming the highest-valued new energy vehicle company in China, with a market cap of $21.4 billion as of November 7, 2025 [5][7]. - The company has shifted its slogan to emphasize its focus on "embodied intelligence," indicating a broader technological ambition beyond just electric vehicles [7]. - Xiaopeng plans to launch three Robotaxi models in 2026, positioning itself as the first new energy vehicle company in China to enter the Robotaxi market [21][22]. Group 2: Technological Developments - The second-generation VLA model was developed to enhance the vehicle's understanding and interaction with the environment, utilizing a significant amount of computational resources and training data [11][20]. - The second-generation VLA is expected to improve performance in complex driving scenarios, with a 13-fold increase in average takeover mileage on complicated roads [20]. - Xiaopeng's Robotaxi will utilize a Level 4 autonomous driving system, designed without reliance on high-precision maps or LiDAR, which is considered an aggressive approach in the industry [22][24]. Group 3: Robotics and AI Integration - The second-generation IRON humanoid robot showcased at the technology day has sparked discussions about its capabilities, focusing on locomotion rather than manipulation, which is a more challenging area in robotics [27][28]. - Xiaopeng has shifted its strategy regarding the application of humanoid robots, moving away from industrial tasks to focus on service-oriented roles such as guiding and shopping assistance [30][31]. - The company acknowledges the significant technical challenges in developing dexterous hands for humanoid robots, which are crucial for broader applications in the future [32].

Core Insights - The article emphasizes the shift of AI from the virtual world to the physical world, marking the emergence of "Physical AI" as the next competitive frontier in technology [3][4][5] - Companies are adopting different strategies to capture the opportunities presented by Physical AI, with some opting for vertical integration while others focus on ecosystem empowerment [5][6][7] Group 1: Trends in AI Development - The consensus in the tech industry is that Physical AI is the next battleground, with leaders like Huang Renxun and Alibaba's CEO highlighting its significance [4][5] - Companies like Tesla and XPeng are pursuing vertical integration, creating a closed-loop system that encompasses chip design, sensors, software, and vehicle manufacturing [5][6] - Other players, such as NVIDIA and Huawei, are acting as ecosystem enablers, focusing on providing the necessary computational power and infrastructure for Physical AI [5][6] Group 2: XPeng's Strategy - XPeng aims to become a "Physical AI world explorer" by integrating multiple transportation scenarios, facing challenges in chip development, algorithm innovation, and commercialization [9][18] - The core of XPeng's technology is the second-generation VLA model, which allows for direct action based on visual signals, enhancing response speed and reducing information loss [9][10] - XPeng has invested heavily in computational power, achieving 2250 TOPS on its self-developed "Turing" AI chip, significantly surpassing industry standards [10][11] Group 3: Product Development and Market Positioning - XPeng plans to launch its Robotaxi in 2026, designed specifically for autonomous driving, which will reduce costs and improve reliability by not relying on expensive lidar and high-definition maps [11][15] - The company is also developing humanoid robots and flying cars, with the "Land Carrier" already receiving 7,000 orders, indicating a move towards scalable commercial applications [13][15] - These products are interconnected, forming a "capability ladder" that enhances XPeng's Physical AI capabilities through real-world applications [15][16] Group 4: Data Acquisition and Ecosystem Challenges - The success of Physical AI hinges on acquiring high-quality data, with XPeng emphasizing the importance of capturing valuable "long-tail and anomaly data" for model training [19][21] - The complexity of the physical world necessitates collaboration across the industry, leading to a paradox where companies must balance openness with maintaining core competencies [21][22] - XPeng's strategy involves a combination of core self-research and open collaboration, ensuring that it retains control over critical technologies while fostering partnerships [22][23]

Core Viewpoint - The article discusses the recent unveiling of Xiaopeng's humanoid robot "IRON" during the "Technology Day" event, highlighting its advanced capabilities and the strategic shift in Xiaopeng's robotics direction, moving away from industrial applications to more interactive roles like guiding and shopping assistance [4][18][20]. Group 1: Robot Presentation and Public Reaction - Xiaopeng's "Technology Day" featured the release of the humanoid robot "IRON," which showcased basic walking capabilities, sparking public interest and skepticism regarding its authenticity [4][6]. - The event generated significant media attention, with Xiaopeng leveraging social media to enhance visibility, including a video of the CEO interacting with the robot [6][7]. - The public's reaction included doubts about the robot's capabilities, with some speculating it was merely a human in disguise due to its lifelike appearance [6][9]. Group 2: Technical Aspects of IRON - The "IRON" robot demonstrated advanced gait capabilities, attributed to its complex mechanical structure and generative algorithms that allow for real-time adjustments in movement [10][14]. - The robot's design features significant upgrades in its joints and movement flexibility compared to earlier versions, indicating a substantial leap in hardware complexity [14][22]. - The integration of generative models into the robot's control systems enables it to perform actions based on high-level commands rather than pre-programmed sequences, mimicking human-like decision-making [15][30]. Group 3: Strategic Shift in Robotics - Xiaopeng has shifted its focus from industrial applications to more consumer-oriented roles for the IRON robot, such as guiding and shopping assistance, reflecting a change in market strategy [18][20]. - The initial applications for the robot have been revised from factory automation to more interactive scenarios, indicating ongoing uncertainty in the robotics business [23][24]. - The timeline for mass production of advanced humanoid robots has been set for the end of 2026, but the specifics of the robot's capabilities and applications remain unclear [24][25]. Group 4: AI and Autonomous Driving Developments - Xiaopeng's advancements in AI and autonomous driving technology are crucial for its core automotive business, with the second-generation VLA model being introduced to enhance driving capabilities [29][30]. - The new VLA model aims to improve data processing efficiency and reduce reliance on labeled data, indicating a significant evolution in the company's AI strategy [30][31]. - Despite the progress, there are concerns about the competitive landscape in the autonomous driving sector, with Xiaopeng's performance reportedly declining over the past year [29][34].

Core Insights - The article highlights the advancements of Xiaopeng Motors in the field of AI and robotics, showcasing their transition from an AI automotive company to a comprehensive "physical AI" ecosystem developer [3][7][9] - Xiaopeng's innovations, including the second-generation VLA model and humanoid robots, are positioned to redefine the automotive and robotics industries, emphasizing a shift from traditional manufacturing to intelligent, data-driven production [10][22][24] Group 1: Technological Innovations - Xiaopeng's second-generation VLA model allows for end-to-end generation from "vision to action," significantly improving the vehicle's capabilities, such as a 13-fold increase in average takeover mileage [11] - The new humanoid robot, IRON, features advanced bionic structures and is designed for practical applications in industries like inspection and service, with plans for deployment in collaboration with Baosteel [17] - The "陆地航母" flying car is set for mass production in 2026, with 7,000 orders already secured, indicating strong market interest and potential for low-altitude tourism [21] Group 2: Strategic Direction - Xiaopeng's shift towards a "physical AI" approach reflects a broader trend in China's manufacturing strategy, moving from mere production to intelligent, self-evolving systems [10][22] - The company maintains a research and development investment of over 20%, utilizing advanced cloud computing resources to support rapid iteration of their AI models [22] - Xiaopeng's initiatives align with national strategies to enhance AI integration in manufacturing, aiming for over 70% application rate of new intelligent terminals by 2027 [16][24] Group 3: Market Positioning - The interaction between Xiaopeng and global tech leaders, such as Elon Musk, signifies a growing recognition of China's role in the global innovation landscape [3][9] - Xiaopeng's comprehensive ecosystem, which includes vehicles, humanoid robots, and flying cars, positions the company as a leader in the emerging "physical AI" market [14][24] - The company's focus on practical applications and commercial viability of its technologies reflects a strategic move to establish a competitive edge in the rapidly evolving tech landscape [19][21]

Core Insights - The core question raised during the Xiaopeng Technology Day is how AI can safely integrate into the real world as physical partners, moving beyond mere algorithms on screens [1] Group 1: Robotaxi Development - Xiaopeng announced plans to trial L4-level Robotaxi by 2026, showcasing a robust technological framework that includes dual redundancy in six core systems, ensuring safety without a physical safety driver [2] - The second-generation VLA model allows AI to operate without reliance on high-precision maps, enabling it to adapt to various traffic rules and environments, thus enhancing its decision-making capabilities [2] - Xiaopeng aims to accelerate commercialization by opening its SDK to partners like Amap, which expands the safety responsibility within a more complex collaborative ecosystem [2] Group 2: IRON Robot Features - The IRON robot, standing at 178 cm, represents AI's entry into personal living spaces, emphasizing its ability to navigate safely rather than just its mobility [4] - Equipped with an indoor AEB (Automatic Emergency Braking) system, the IRON robot can detect and avoid obstacles in home environments, enhancing its practical utility [7] Group 3: Physical World Model - Both the Robotaxi and IRON robot are powered by a physical world model, which fundamentally differs from internet models like ChatGPT, as it must adhere to strict physical laws [6] - Xiaopeng's model, with 720 billion parameters, enables AI to understand complex physical interactions, such as recognizing reflective surfaces or assessing water depth, marking a significant advancement in physical AI capabilities [9] Group 4: Safety as a Priority - Xiaopeng's approach emphasizes "edge-side reinforcement," where high levels of intelligence are paired with robust safety redundancies, highlighting the critical nature of safety in physical AI applications [10] - The industry is exploring alternative safety frameworks, such as the "system-level collaboration" exemplified by Mogu Car Union, which aims to enhance overall traffic safety through interconnected AI systems [10] Group 5: Future Challenges - The future of physical AI will hinge on its ability to interpret human social intentions and the legal implications of accidents, which will determine the extent of its integration into society [12] - The industry has made significant strides by prioritizing safety over mere technological display, marking the beginning of a crucial phase in the development of physical AI [12]