Li Auto - Earnings Call - Q3 2025

Transcript

Speaker 6

Hello, ladies and gentlemen. Thank you for standing by for Li Auto's third quarter 2025 earnings conference call. At this time, all participants are in listen-only mode. Today's conference call is being recorded. I will now turn the call over to your host, Ms. Janet Zhang, Investor Relations Director of Li Auto. Please go ahead, Janet.

Speaker 0

Thank you, operator. Good evening and good morning, everyone. Welcome to Li Auto's third quarter 2025 earnings conference call. The company's financial and operating results were published in a press release earlier today, and they were posted on the company's IR website. On today's call, we will have our Chairman and CEO, Mr. Xiang Li, and our CFO, Mr. Johnny Tie Li, to begin with prepared remarks. Our President, Mr. Donghui Ma, and CTO, Mr. Yan Xie, will join for the Q&A discussion. Before we continue, please be reminded that today's discussion will contain forward-looking statements made under the safe harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995. Forward-looking statements involve inherent risks and uncertainties. As such, the company's actual results may be materially different from the views expressed today. Further information regarding risks and uncertainties is included in certain company filings with the U.S.

Securities and Exchange Commission and the Stock Exchange of Hong Kong Limited. Any obligation to update any forward-looking statements except as required under applicable law. Please also note that Li Auto's earnings press release and this conference call include discussions of unaudited GAAP financial information, as well as unaudited non-GAAP financial measures. Please refer to Li Auto's disclosure documents on the IR section of our website, which contain a reconciliation of the unaudited non-GAAP measures to comparable GAAP measures. Our CEO will start his remarks in Chinese. There will be English translation after he finishes all his remarks. With that, I will now turn the call over to our CEO, Mr. Xiang Li. Please go ahead.

Speaker 3

大家好，我是李响。2025年的第三季度，也是理想汽车面向第二个十年的第一个季度，我们经历了产品周期、公关舆情、供应链爬产、政策变化等问题带来的各种挑战。这些因素对我们的交付和经营也产生了影响。但是今天更想和大家聊一聊我们长期的思考，理想汽车面向第二个十年，三个最重要的关键选择应该怎么做，分别是组织、产品、技术。第一个关键选择是组织。我们所面临的选择是创业公司的管理模式，还是职业经理人的管理模式。理想汽车在过去的十年里，最初的七年是创业公司的管理模式。随着规模的扩大，大到了我之前创业所没有经历过的营收规模。在2022年前后，很多人都会建议我们走向职业经理人的管理模式。因为在历史上，奔驰宝马这样的百年汽车企业，以及微软、苹果这样的科技巨头，都在这一模式下取得了非常大的成功。过去三年，我们非常努力地让自己变成职业经理人的治理体系。在真实的体验和落地后，我们认识到创业公司和职业经理人是两种完全不同的治理体系。它与流程、组织结构无关，而是管理理念和要素的差异，以及各自适用于不同阶段和行业环境。职业经理人的管理方式可以非常成功，但是需要三个要素：第一，行业和技术周期相对稳定；第二，企业的行业地位领先且稳固；第三，创始人和创业团队要么没有动力干了，要么已经不在了。如果满足这些条件，职业经理人的管理模式是非常好的选择。苹果和微软也都在职业经理人接班后，从千亿美金成长为万亿美金的企业。而创业公司的条件恰恰相反：第一，行业和技术周期在发生巨大的变化；第二，行业格局不确定，企业还不是领先者；第三，创始人和创业团队每天还在投入地努力工作，充满动力。在人工智能重塑各个行业的今天，我们所处的环境，我们自身的特点，更符合创业公司的条件。创业公司的管理的核心是四件事：第一，更多的对话，深度的对话，而非更多的汇报。在高度变化中，深度的对话是提升认知和大胆决策的关键要素；第二，聚焦用户价值，而非只是交付。对用户真正有价值的东西才值得交付，而不仅仅是完成各种交付的任务；第三，持续提升效率，而非占有更多的资源。去年花十块钱做的事情，今年就要花八块钱去做，从而才有更多的资源去做短期不创造收入的长期投资和能力建设；第四，要识别关键问题，而不是创造信息不对称。只有价值增加了，效率提升了，关键问题解决了，才可以在高度竞争、高度变化的环境中持续满足用户的需求。过去三年，我和创业团队努力学习职业经理人的管理体系，逼迫自己接受各种变化，但是我们却变得变成了越来越差的自己。而英伟达和特斯拉今天仍以创业公司的方式管理。如果全世界最强的那些公司都可以使用创业公司的管理方式，我们有什么理由放弃自己最擅长的方式？从1998年开始，我有27年的创业公司的管理经验，而且从来没有在任何大企业中以职业经理人的方式工作过。当下面对的又是一个行业高度竞争、技术高度变化的大环境。我自己热爱汽车，热爱产品，热爱人工智能，工作就是自己最大的爱好。为什么不用自己最擅长的能力和方式来管理理想汽车？这是理想汽车面向第二个十年第一重要的选择。因此，从今年四季度开始，我们坚定地回到创业公司的管理模式，以此来面对新时代、新技术的挑战。组织的选择是一切的基础。进入新的十年，更关键的问题是我们真正要为用户解决什么问题，产品是什么，技术往哪里走。这是永恒不变的本质。首先，关于产品，我们同样要做一个选择。面向未来十年，我们的产品到底是什么？是电动车？是智能终端？还是具身机器人？当产品停留在电动车，竞争的逻辑就会变成参数大战。续航多20公里还是少20公里？车长多两厘米还是少两厘米？如果只是电动车的竞争，就是更大的空间、更多的续航、更便宜的价格，顺便copy一下已经被验证成功的那些设计，比如理想L9。所有除此之外的研发投入都是研发成本的浪费。更强的传感器、更强的模型、更强的算力、更强的主动悬架，都是整车成本的浪费。甚至性能强大的算力和主动悬架等产生的能耗还会给续航带来负面效应。如果我们选择做智能终端，就会更加关注屏幕里的那些事，本能地变成智能手机功能的重复建设。过去几年，行业在智能终端上大部分创新，本质上是把智能手机的功能搬到车上，把手机的APP放到车机里，把车在车里上更大的语言模型，甚至希望能够在车里做coding，在车里做deep research。用户买车，到底是为了工作还是为了更好的生活？明明在手机和电脑上有更好的体验、更自然的应用，为什么非要放在车上？这些投入对于用户的价值提升非常有限，甚至是企业的自娱自乐。我们还可以让车变成物理世界的具身智能产品，说白了就是机器人。《变形金刚》的动画片和电影明确地告诉我们，机器人最大的分类就是两种形式，一种像人一样，一种像车一样。皮里欧霞和赛车总动员也清晰地告诉我们，汽车是机器人的核心形态之一。如何把车变成机器人？给它眼睛和耳朵的感知能力，给它大脑和神经的模型能力，给它心脏等器官一样的计算能力，改造硬件形成更强的本体能力。让汽车机器人具备顶级司机的能力，不仅可以开车，还可以每天迎接你，帮你停车，帮你充电，给你开关车门，无微不至地为你提供便捷、体验和关怀。它还可以扮演父母、助理、乘务员的角色，在车内给你提供最便捷的生活服务和空间照顾，就像飞机的总等舱服务，就像小时候妈妈一直在我们身边的照顾和关怀。如何定义一个好的具身机器人？从被动的机器成长为自动的机器，再进化为主动的机器人。未来十年，具身智能最有价值的产品一定是具备自动和主动能力的汽车。产品的竞争也是自动和主动能力达到什么程度，而这些价值会融入高频的生活和体验中，有了就再也回不去了。电动车的选择不是不好，而是不够。智能终端的选择也不是不好，而是不够。只有选择具身智能这个最难的题，我们才可以真正改变用户的生活，提供自动和主动服务的具身智能机器人产品。恰好它就是《变形金刚》的汽车形态机器人，是赛车总动员皮里欧霞的汽车形态机器人。新时代给予汽车企业和创业者们最好的机遇和最高难度的挑战。接下来是技术的选择。更具体地讲，是完整AI系统的技术选择。是面向数字世界的语言智能，还是物理世界的具身智能？这意味着要构建完全不同的系统能力。要想做好具身智能，最重要的是构建一套不同于语言智能的AI系统。它包含具身智能的感知，相当于眼睛和耳朵；具身智能的模型，相当于大脑；具身智能的操作系统，相当于神经；具身智能的算力，相当于心脏；具身智能的本体，相当于人类的身体。当下没有任何第三方供应商可以完整地提供这套系统，甚至没有任何一个是可以供应的。语言智能最大的特点是关注好模型和计算。更大规模的模型，更大规模的计算，就会带来更强的能力。而具身智能必须增加对物理世界的感知，模型也是必须基于对物理世界的理解。精准是首要任务，其次才是泛化。操作系统需要确保软硬的最优融合，提高更高的帧率，整体系统要快和精准。为具身智能的感知、模型、操作系统等提供的算力，也需要在端侧供应，而不是云端。最后，还要改造硬件成为具身智能的本体，比如三维神经控制的线控和主动AI悬架系统，提高本体执行的效率和准确性。如果以具身智能理解整个AI系统，你发现可以改变以及急切需要改变的实在太多了。首先是感知。基于现有的感知模型和端侧的算力，现在的3D DEV、OCC占用网络、2D VIP，有效的感知距离，我讲的是有效，而非理论上最大，只有一百多米，远远不如人类的眼睛。如果升级成人眼工作原理相似的3D VIP，有效距离可以扩大两至三倍。今天我们辅助驾驶可能50%以上的常见问题自然就解决了。3D VIP不仅限于自动驾驶的领域，在车外和车主的主动交互与服务，在车内与家人的主动交互和服务，也就变成了可能。这需要感知模型的研究和研发的重大突破，还必须有M100这样的为具身智能定制设计的芯片和强大的编译团队高效配合才可以实现。其次是模型。有了3D VIP，才有了真正对物理世界的理解。VLA模型中的VL才可以更好地对物理世界的感知和理解。人类的数据才可以更高效地被用于训练。世界模型生成的数据才可以更好地去完善训练。现有最好的计算平台，一个4B的MOE的模型运行帧率只有10赫兹，而执行系统是60赫兹。如果模型运行的帧率可以快两至三倍，现在辅助驾驶的一些舒适性的问题、反应迟钝的问题都可以有效地解决。这同样需要对传统的GPU架构和算力进行深度的改造和定制，以及专有的操作系统。M100芯片就是为解决具身智能这些本质的问题而研发的。最后是本体。人类的刹车转向的最快响应速度在450毫秒左右。目前自动驾驶的感知到执行的完整链路在550毫秒左右。坐在驾驶员的位置上，本能的反应就是慢很多，像老年人在开车。电动体系可以把整个链路的响应速度提升到350毫秒。别小看这200毫秒的差距，这大概可以把事故率下降50%以上，而且让人感觉开得比自己还要好，还要安全。既是理性的真安全，又是感性的安全感。基于这样的需求，整个控制的方式都会变得完全不同。如果只是关注语言智能那样的模型规模的提升，模型规模扩大一倍，算力提升带来的性能提升只有5%-10%。如果从具身智能整体来看，每个环节全系统地去解决最关键的问题，那接下来一轮自动驾驶的性能提升应该有5-10倍。具身智能为用户提供的自动和主动的快速交互和真实服务的能力，更是有何无的质变。过去三年，我们对于具身智能完整系统的技术储备，让我们下一代的产品充满信心。具身智能机器人的元年从汽车机器人正式开启，千亿收入只是起步。以上三个关键的选择决定了我们第二个十年发展的基础。它比上一个十年更难，也更具挑战性。我们深知未来的竞争不在一时一刻，而在方向的长期选择和持续投入的定力。依托坚实的财务基础，我们会保持专注，用我们最热爱的创业公司的管理模式，打造领先的具身智能产品，确保理想汽车能够穿越周期，引领技术，成为一家长期为用户和社会创造价值的企业。我也希望以后更多地以这样的方式与投资人们进行沟通和交流，而不是每个季度以固定格式的汇报。感谢我们的投资人在我们最难的时候仍然支持和信任我们。我们一定尽最大的努力，在未来的三到五年内，让理想汽车成为具身智能领域表现最好的企业，用户价值最高的企业。谢谢。 Last translation for Mr. Li. The third quarter of 2025 was also the first quarter in the second decade of Li Auto. We went through many challenges, including supply chain, product lifecycle, PR challenges, as well as changing policies. All these factors have had a negative impact on our operations and deliveries. However, today I want to take this opportunity to talk about our long-term thinking in the next decade and three most important choices that we need to make: organization, product, and technology. The first choice we need to make is organization. The challenge we're facing is whether to choose an entrepreneurial model or a professional management model.

In the last ten years of Li Auto, the first seven of those years we operated as an entrepreneurial model company, but as we scaled over time to a scale that we've never seen before, especially in terms of revenue, around the time of 2022, many people suggested to us to shift to a professional management model. Because historically, whether it's Mercedes-Benz or BMW or any of these 100-year-old car enterprises, as well as Microsoft and Apple, which are the tech giants, have all operated under this model and had great success. In the last three years, we tried very hard to make ourselves used to this professional management model. After implementation, we realized, came to the realization that the entrepreneurial model and the professional management model are fundamentally different, and it is irrelevant to processes and organization structures.

The difference really lies in management principles and key operating principles, and also they are tailored to different stages of growth and industry environment. The professional management model can be very successful, but it relies on three factors. The first one is that the industry and technology cycle has to be relatively stable, the second is that the enterprise is already in a leading position, and the position is relatively stable. The third one is that the founder and the founding team are either lost their motivation or are not actively involved in the company. If all these three criteria are satisfied, a professional management model could be a very ideal choice. Whether it is Apple or Microsoft, both have flourished after professional management took over and grew from hundreds of billion dollars in revenue to trillion-dollar companies. However, the entrepreneurial model is catered to an entirely different environment.

First of all, the industry and technology cycles are going through fundamental changes. Second, the industry is very unstable, and the company enterprise itself is not yet a leader. Thirdly, the founder and the founding team are still devoted to everyday work with their full passion and fully motivated. As AI is shaping many industries today, the environment that we live in and considering the traits of this company, we think that we fit into the entrepreneurial model way better. The entrepreneurial model really is about four things. First of all, there needs to be more conversations as opposed to reports. In a rapidly changing environment, deep conversations are really key to increasing our knowledge and judgment of the world, as well as to making bold decisions. Secondly, it's focusing on user value as opposed to just short-term deliveries.

Only those things that create value for the users are worthy to be delivered, as opposed to only focusing on how many tasks that we've delivered on. The third one is keep increasing efficiency as opposed to occupying more resources. For example, if we spent $10 on doing something last year, and this year we need to do it with $8, that's how we have resources to really spend on projects and investments that do not generate short-term revenue but really benefit us in the long term. The fourth, the key is to recognizing the key issues as opposed to just creating information asymmetry. Only as we create more value and increase efficiency and solve the key issues can we really thrive in a highly competitive and rapidly changing environment and consistently meet customer demands.

In the last three years, me and my team have tried very hard to adapt to the professional management model, and we have forced ourselves to embrace all kinds of changes. However, we all realized that we became diminished versions of ourselves. NVIDIA and Tesla are still operating as an entrepreneurial company, and if the largest and strongest companies are all operating in the entrepreneurial model, there is no reason for us not to utilize our strengths and what we're most used to. Since 1998, I have 27 years of running entrepreneurial companies, and I have never worked in any large corporation as a professional manager. Now we're facing a highly competitive and rapidly changing environment with rapidly changing technologies. I personally am passionate about products, about automobiles, and about AI, and work is my largest passion.

Why don't I focus on what I'm most used to and what I'm best at to manage Li Auto? That's the most important first choice as we look into our second decade. As a result, starting from Q4 this year, I and my founding team will firmly revert back to the entrepreneurial model to embrace the new era and new technological challenges. The choice of organizational model is the foundation of everything. Looking into the next decade, the next key question is how we really solve issues for our customers. First of all, what products do we build, and where is technology headed? That's always the essence of everything. First of all, on products, we also need to make an important choice. What kind of products should we really build for our users? Is it electric vehicles? Is it smart devices? Or is it embodied robots?

If we only focus on electric vehicles, competition is really all about an arms race in spreadsheet. Do you have 20 km of range more? Do you have a car that's 2 cm longer in dimensions? If it's only focused on electric vehicles, it's all about larger space, more range, and cheaper prices, and maybe copy some proven designs, just like how Li Auto has been copied. Other than that, all R&D investments are a waste. Stronger sensors, bigger models, more computing power, better active suspension are always a waste of cost. Even stronger computation power and active suspension may even have a negative impact on range. Secondly, if we choose to focus on smart devices, then we'd automatically be focused more on what happens in the screen. Features that used to belong to smartphones and smart tablets will be migrated to the car environment.

In fact, most of the innovations in smart devices are really about moving what's already available in smartphones into vehicles and moving mobile apps into head units, deploying larger language models in head units, and even decoding in cars and conducting deep research. We ask ourselves the question, when our users buy our cars, do we really buy it for their work or to live a better life? If certain experiences are already better in mobile phones and tablets or computers and more natural, why should we even bother putting them in cars? All these investments create very little incremental value for users. Thirdly, the third route is for us to make our cars into an embodied AI in the physical world, or in layman's terms, robots. The movie Transformer told us that there are broadly two types of robots.

The first type looks like human beings, and the second type looks like cars. Knight Rider and Cars, these TV shows or movies have clearly shown us car-shaped robots are going to be a mainstream form factor of robots going forward. How do we transform our cars into robots? We need to give it ears and eyes for perception. We need to give it brains and nerves, which is modeling capability. We need to give it hearts, which is computing power, and we need to reshape the hardware to make it a stronger physical presence. Our robots need to parallel the top drivers. It can not only drive but also pick you up, park for you, help you charge the car up, help you close the door, open the door, and meticulously make your life more convenient and safer.

It can also play the role of parents, assistants, or even flight attendants to provide you the convenience and take care of you within the sphere of the car, just like first-class cabin and the services on planes. It is also like when we're little that our mother takes care of us and makes us happy. How do we define a good embodied robot? How do they make them change from passive machines into an automated machine and then further into proactive machines? In the next decade, the most valuable embodied AI product is going to be vehicles that are automated as well as proactive. Competition is really how automated and proactive can we make these products, and how can we fuse them into high-frequency life experiences, something that once we get used to, we can never go back.

Whether it's electric vehicles or smart devices, these are not necessarily bad choices, but we think they're not sufficient. Only if we choose embodied AI, which is the hardest of all these three problems, can we really change the life of our users and really provide automated and proactive services that only embodied AI products can provide. It's really like what you see in Transformers movies, they're car-shaped robots, or what we see in cars or Knight Riders, they're robots that are shaped in cars. I believe that this is the biggest challenge and opportunity that we entrepreneurs see in this new era. The next choice is about technology, or more specifically, our full-stack AI system. What do we choose? What kind of technology do we choose to power this full-stack AI system?

Is this something that's language-based, that's based towards the digital world, or is it something based towards the physical world? These two options require completely different system capabilities. If we want to build a good embodied AI, we need to build an AI system that's completely different from language-based AI models, including perception like eyes and ears, including the model itself like brain, including the operating system like nerves, and including the computation power, which is like hearts, and also, of course, the physical body itself, just like human body. At this moment, there's no third-party supplier that can provide this full-stack system. In fact, not any company can provide even part of this system. The focus of large language models is really focusing on the model itself and computation. Larger models and more computation power are always going to generate stronger capabilities.

However, for embodied AI, we need to better understand the physical world, and the model is also built on our understanding of the physical world. Accuracy is the first priority, and generalization only comes next. Operating system needs to make sure the optimal integration is made between the hardware and the software and also provide a higher frequency. The system needs to be fast and precise. The computation power that powers the perception, the model, and the operating system needs to reside on the device side as opposed to the cloud side. Lastly, we'll also need to modify the hardware itself to become a really embodied hardware. For example, our active suspension is just like a 3D nerve system, nerve control, and it can increase the efficiency and position of execution in the physical world.

If we look at this entire AI system through the lens of embodied AI, you will see that there's so many changes that need to happen and desperately need to happen. The first change comes in perception. Based on the current model and the computation power that can be deployed on the device, the current 3D BEV or occupancy network or 2D vision transformer, the effective range of perception, I'm talking about effective as opposed to theoretical maximum, is only just about over 100 meters, which is way less than human eyes. However, if we upgrade it to 3D vision transformer, which is just similar to how human eyes work, this range can be improved by two-three times, and it can solve more than 50% of the common issues we see in autonomous driving.

3D vision transformer is not only limited to autonomous driving, but it can also benefit interactions with the car inside and outside of the car. These can also all become possible. That requires fundamental breakthroughs in perception models, both in research and also development. It also requires tailored chips for embodied AI, just like M100, which we have developed, and also requires a very strong compiler team and high-efficiency cooperation. The next area of improvement is in models. It's only with 3D vision transformer can we really understand the world. The VL in the VLA can really understand and perceive the world better, and human data can be more effectively used for training, and world model can also be used more effectively for training.

For example, in the status quo computation platform, a 4 billion parameter MOE model can only run at 10 hertz, but the execution frequency is 60 hertz. If we can increase the frequency of the model by two, three times, it can also automatically solve many issues, including comfort and speed of reaction in autonomous driving. It also requires us to fundamentally modify and customize the traditional GPU architecture and to have a dedicated operating system. M100, again, is really designed for solving these embodied AI problems. Lastly is the embodied hardware itself. A human being can typically react to braking and steering in about 450 milliseconds. For a typical autonomous driving system, from perception to execution, the entire closed loop takes about 550 milliseconds. For a typical driver today, it is obvious to them that autonomous driving is much slower.

It's like an elderly driving car. The drive-by-wire system can reduce the response time to about 350 milliseconds. The difference of 200 milliseconds is not to be underestimated. It can roughly reduce the accident rate by over 50%. It also feels better even than driving by themselves, and it's also safer. It's safer both in the subjective as well as the objective sense. Based on these needs, the entire control mechanism will be different. If we only focus on increasing the scale of model just like we did in language models, for example, if we increase the size of model two times and with corresponding increase in computation power, the really performance increase is only going to be 5-10%.

If we look at this from an embodied AI perspective and to solve the key issues in every stack, on every level of stack, the next generation autonomous driving can really increase the performance by five- to tenfold. That is what can power embodied AI to perform fast and accurate and valuable services. That is the difference between zero to one. In the past three years, we have made a lot of progress in technology and systems for embodied AI. That makes us very confident about the next generation of products. The start of embodied AI robots starts with car robots, and it is starting this year, I believe. Hundreds of billions of revenue is only a starting point. The above three key strategic choices really lay the foundation for the next decade of our development. It is more challenging than the last decade.

We are deeply aware that real competition is not really about short-term wins. It is about staying on the right path over the long term and having the dedication to keep investing in it. Backed by a strong financial foundation, we will stay focused, embrace our beloved entrepreneurial management style, and build leading embodied intelligence products. Li Auto can navigate market cycles, lead technological transformation, and become a company that creates unique, lasting value for users and society in the long run. Finally, I also look forward to engaging with you guys in this manner moving forward rather than presenting a quarterly report in a fixed format. I want to express my gratitude to all of you for your support and trust, especially during our most challenging times.

We're fully committed to making Li Auto the best-performing company in embodied intelligence and the greatest creator of user value within the next three to five years. Thank you.

Speaker 6

Thank you, John. Hello, everyone. I will now walk you through some of our third-quarter financials. Given time constraints, my remarks today will be limited to the financial highlights. All figures will be called in RMB unless otherwise stated. For further details, I'll be encouraged you to refer to our earnings press release. Total revenues in the third quarter were RMB 27.4 billion, decreased 36.2% year over year, and 9.5% quarter over quarter. This included RMB 25.9 billion from vehicle sales, decreased 37.4% year over year, and 10.4% quarter over quarter. Mainly due to lower vehicle deliveries, the sequential decline was partially offset by a higher average selling price due to the different product mix. Cost of sales in the third quarter was RMB 22.9 billion, down 22.9% year over year and 5.3% quarter over quarter.

Gross profit in the third quarter was RMB 4.5 billion, down 51.6% year over year and 26.3% quarter over quarter. Vehicle margin in the third quarter was 15.5% versus 20.9% in the same period last year and 19.4% in the prior quarter. The year-over-year decrease was mainly due to estimated Li Mega recall cost and the higher per unit manufacturing cost from a lower production volume. The sequential decline was mainly due to the same recall-related cost. Excluding such recall costs, vehicle margin would have been 19.8% in the third quarter. Gross margin in the third quarter was 16.3% versus 21.5% in the same period last year and 20.1% in the prior quarter. Excluding the above-mentioned Li Mega recall cost, gross margin would have been 20.4% in the third quarter.

Operating expenses in the third quarter were RMB 5.6 billion, down 2.5% year over year and up 7.8% quarter over quarter. R&D expenses in the third quarter were RMB 3 billion, up 15% year over year and 5.8% quarter over quarter. The year-over-year increase was mainly due to the impact of the pace of new vehicle programs and the increased investment in expanding our product portfolio and technology, along with expenses from the product configuration adjustment. The sequential increase was mainly due to those same product configuration adjustment expenses. SG&A expenses in the third quarter were RMB 2.8 billion, down 17.6% year over year and up 1.9% quarter over quarter. The year-over-year decrease was mainly due to the recognition of share-based compensation expenses regarding the CEO performance-based awards in the third quarter of last year.

Loss from operations in the third quarter was RMB 1.2 billion versus RMB 3.4 billion income from operations in the same period last year, and RMB 827 million income from operations in the prior quarter. Operating margin in the third quarter was negative 4.3% versus 8% in the same period last year and 2.7% in the prior quarter. Net loss in the third quarter was RMB 624.4 million versus RMB 2.8 billion net income in the same period last year, and RMB 1.1 billion net income in the prior quarter. Diluted net loss per ADS attributable to our ordinary shareholders was RMB 0.62 in the third quarter versus diluted net earnings of RMB 2.66 in the same period last year and RMB 1.03 in the prior quarter. Turning to our balance sheets and cash flow, our cash position remains strong with a quarter-ended balance of RMB 98.9 billion.

Net cash used in operating activities in the third quarter was RMB 7.4 billion versus RMB 11 billion provided in the same period last year, and RMB 3 billion used in the prior quarter. Free cash flow was negative RMB 8.9 billion in the third quarter versus RMB 9.1 billion in the same period last year and negative RMB 3.8 billion in the prior quarter. Now for our business outlook. For the fourth quarter of 2025, the company expects the delivery to be between 100,000 and 110,000 vehicles and total revenue to be between RMB 26.5 billion and RMB 29.2 billion. This business outlook reflects the company's current and preliminary view on its business situation and market conditions, which is subject to change. That concludes our prepared remarks. I will now turn the call over to the operator and start our Q&A session. Thank you.

Speaker 3

Thank you. If you do wish to ask a question, please press star then one on your telephone and wait for your name to be announced. If you wish to cancel your request, please press star two. If you are on a speakerphone, please pick up the handset to ask your question. For the benefit of all participants on today's call, please limit yourself to two questions. If you have additional questions, you may re-enter the queue. If you are a Mandarin speaker, please ask your questions in Chinese first, then follow with English translations. Your first question comes from Yingbo Zhu at Citigroup. Please go ahead.

好的，感谢管理层。我是中兴证券的许英博。我有两个问题。第一个呢，我们很高兴地听到李想总关于过去的一个波折的反思，也更高兴地听到管理层计划重新回归创业公司的这么一个状态。也十年做了一个产品规划和技术展望。我们也同意就是说AI时代的产品大概率不是一个简单的升级，应该是基于AI时代的用户体验的全新产品设计。那车和具身智能也都是物理AI的这个重要的载体，也有同源的技术和相似的底层逻辑。但是我们其实更想关注的就是说任何技术和产品的代际的研发和切换，它都是需要时间的，它是很难立竿见影的。所以我的第一个问题就是，如果我们只展望未来一年，2026年有什么我们能够期待的技术和产品，以及基于未来十年的考量，我们投资人大概在多远的未来真正能够看到公司产品和技术的一个跳升，感受到真正的不同。我的第二个问题呢，是关于三电的，就是公司产品从增程向纯电切换，不可避免会面对很多挑战。我们也已经面对了这样的挑战。那么围绕着电动的这个技术和产业链，在未来纯电车的三电领域，我们做了哪些有效的技术储备和供应链的储备，能不能给投资人更多的信息和更多的信心。 I have two questions. The first question is about, we are very glad to hear, the company's return to entrepreneurship and next decade's plan, but any R&D and development needs time. My first question is that if we just say next year, 2026, what kind of technology or product progress can we expect? Also from the investor's perspective, how long can we really see a technology or product jump in the future? How long? The second question related to BEV, the company's transition from EREV to BEV, these challenges. Can you please give us more information or confidence in this BEV part, how we prepare for the effective technology reserve and the supply chain preparation? Thank you.

哎，英博，我来，我是李想，我来回答第一个问题。那，呃，我们认为在2026年的时候，也就是在明年，然后搭载我们以M100芯片，然后为核心的，然后AI系统，然后开始交付在产品面上的时候，然后我觉得这个真正的价值和体验就会发生改变。那我觉得具体的改变还是我之前讲的，对，会让汽车这样的产品，然后从被动的使用，然后变成真正的，然后自动化和主动化来为用户提供服务。而且这种服务和体验，然后并不是说我们要搞一个Deep Research，或者我们去做一个Coding，或者我们生成一个视频，然后而是它会体现在我们日常的生活和高频体验中，每天都会感受到，每天都能够体验到这样的价值带来的根本性的变化。对，我觉得这是讲的第一点。我第二点说，如果未来的十年，然后到底有什么样的，然后明确的功能啊，或者价值啊，然后可以呈现。那我觉得基于我们对，然后人工智能和具身智能的理解，对，它没有办法像传统的编程或者硬件功能，对，给出明确的功能这样的一个特性。而是说，当我们去构建整个人工智能的系统化能力的时候，只要一些关最关键的核心的问题被解决，或者一些关键的性能进行突破，它会带来一系列的，然后体验在我们具体的生活中，或者产品价值上的，然后一系列的，然后改变，而且这些改变甚至会远超我们的预期。我觉得这是我们正确的，我们自己理解的人工智能和具身智能，然后能够给我们的物理世界，给产品，给我们日常的生活，然后带来的转变以及背后的驱动力，而非传统的编程啊和功能模式的这样的一个呈现方式。

On your first question about 2026, next year we will be launching our AI system based on our internally developed M100 chips. Once this system gets in the car, that's where we will start to see real value and change of user experience. As I mentioned earlier, our products would go from a passive, a machine that passively takes orders to a more automated machine and even a proactive machine that can provide services for the users. Unlike large language models, which can conduct deep research or video generation, this embodied AI products can really benefit our users in their everyday use at a very high frequency. On the second part about the next 10 years, unlike programming or traditional rule-based programming, we do not have a feature list or a list of functions.

Instead, AI really, for a complex AI system, if we can solve key issues in some important areas and improve performances in some bottleneck points, we will start to see a series of changes that are unimagined before. That is our understanding of embodied AI and AI system. This is really the room for imagination for the next 10 years. Thank you.

哎，那个英博你好，我是理想汽车马东辉。我来回答第二个问题，关于三电领域的研发和供应的布局。其实理想汽车一直啊，聚焦在三电核心技术上进行的话自研，包含电驱动啊、电池和电控技术的一些突破。然后我们在电驱系统方面的话，我们主要是聚焦对于用户有价值的，像能效啊，以及到化体验。那我们自己的话是自研代工了碳化硅功率芯片，然后并且我们自研自制了功率模块以及电机控制器的生产和研发。那同步呢，理想汽车也自己建成了专属驱动电机的工厂。那这样的话就构建了从碳化硅功率芯片，然后再到功率模块，再到电机的全链条自研能力。然后并且我们的电驱技术啊，啊，不仅是用在纯电车型上面，也包含征程的全系车型，可以保证在驾驶过程当中很安静很平顺，然后并且也有效的话是优化了能耗的水平，提升的话车辆的续航里程。那关于电池技术的话，我们是聚焦在充电速度和安全，然后我们围绕的话5C超充的电池，然后也构建了全站的自研能力，包括电芯的化学体系，然后电池管理系统BMS控制模型和算法，然后以及电池Pipe的布局和结构设计，可以实现超快充啊、长续航、长寿命三重的核心优势。然后在电池的共用布局方面的话呢，呃，我们规划的话是采用外购，然后加自研合资的双模式。那明年我们也将量产理想自主品牌5C的电池，然后通过自研技术产业化的落地，进一步强化电池的安全性能，然后升级啊，用户的本能体验。那最后一点的话是在电控，我们通过软硬件的全自研，然后打造极致的驾驶质感。那在软件端，我们可以实现对于动力的控制啊、能量的管理以及发动机的标定等全站的话自研，然后在硬件端的话，像一些核心的预控制器啊、PCB layout呀，以及底层软件的自主研发啊，都是我们全站自研。通过与自研底盘技术的协同，然后可以实现对于车辆驾驶性和平顺性的精准优化，然后打造好开一上手的驾驶特质。那总体来说，通过三电技术的自研，我们是希望能够以技术硬实力，然后为用户带来充电快啊、续航长、好开又安全的确定性体验。

On the key in-house dev-related technologies, we focus on three areas: electric drive, battery systems, and electronic control. First of all, on the electric drive system, our focus is on efficiency and user experience. We have an in-house developed and outsourced our manufacturing of silicon carbide power chips and in-house developed and self-manufacturing of power modules and motor controllers, but also established our own dedicated drive motor factory. We have built a full chain in-house development capability, spanning from silicon carbide power chips, power modules to electric motor assemblies. Our electric drive technology covers all BEV and EREV models, ensuring quiet and smooth driving experience while also optimizing for energy consumption and vehicle driving range. Secondly, on the battery system, our focus is on ultra-fast charging and safety.

We have built a full stack in-house capability around 5C ultra-fast charging batteries with full control over self-chemistry BMS control modules and algorithms, as well as battery pack layouts and structural design, achieving three core advantages across ultra-fast charging, long driving range, and long service life. On the supply front, we also have a combined strategy of external procurement and in-house development. Li Auto's own 5C batteries will enter mass production next year. This industrialization of in-house developed technology will further strengthen our battery safety and also improve user experience. Thirdly, on the electronic control system, our goal is to provide the best driving experience also through in-house developed hardware and software. On the software side, we have full stack in-house development capability of powertrain control, power management, and engine calibration.

On the hardware side, our core domain controller PCB layout are all developed in-house, as well as the underlying software. Together with our in-house chassis technology, we were able to enhance driving smoothness and comfort and make the drive experience easy and intuitive to our users. Through a combination of three electrical technologies, including battery electric control and electric drive, we provide our users with a special fast charging, long range, and smooth and safe driving experience. Thank you.

Speaker 6

Thank you. Your next question comes from Tim Hsiao from Morgan Stanley. Please go ahead.

Speaker 1

Q3 的经营活动现金流的流出进一步的扩大，这也导致于可能现金储备有所减少。主要的原因是什么，我们怎么看待后续现金流的一个改善的一个情况。 I have two quick questions focusing on the near-term operation. The first one about the product, the ICUs. Could the management team share the latest updates on orders and deliveries of Li I8 and I6? In the meantime, how and when could you staff the current supply bottleneck of the Li I6 and I8? How should you think about the normalized sales volume of the two ICUs models in the following months? The second question is about cash flow. Li Auto actually registered increasing operating cash outflow of about CNY 7.4 billion or free cash outflow of CNY 8.9 billion during this quarter. This caused quite a significant drop in company's cash reserve. I want to know why, why is that, and how should you think about the cash flow in the following quarters? That's my second question. Thank you.

Speaker 8

呃Tim你好，我是理想汽车马东辉。我来回答第一个问题，然后呃今年的话我们通过I6和I8，然后构建了BEV的产品矩阵，然后分别去覆盖主流和高端的家庭纯电市场，也为理想汽车纯电业务长期的稳定增长奠定了坚实的基础。那对于我们来说的话，我们也完成了征程和纯电拴能源形式的产品的一个布局，实现的话有效的互补，满足用户不同的需求。那对于纯电车型的话核心亮点，那I系列产品实现了区域渗透率的大幅突破，特别是像对于像上海啊、北京、江浙沪等纯电的话核心市场，那9月份上述地区的订单大幅的话增加。那关于供应的话，目前呃理想I8和I6两款产品呃正经历产能爬坡啊交付提速以及市场渗透发展的路径。那从11月份起，为了解决产能爬坡，然后I6电池的供应将正式启用双供应商的模式，但是我们会保障两家供应商电池的性能跟质量标准是一致的。然后预计的话，呃明年年初I6的月产能将稳步提升到2万台。那在此呢，呃我们也想向理想I6的订单车主啊致以诚挚的歉意，受核心零部件供应的规划和产能爬坡的影响啊，您的提车时间有所延迟。感谢您的信任和选择，也恳请哈给予我们更多的支持和时间啊，我们将全力加班加点提升产能，加速交付的进程。谢谢。

Speaker 5

This year we established our map portfolio with I8 and I6 models, and respectively they cover the mainstream and premium segments for the family BEV market. These two cars create a solid foundation for the long-term stable growth of our BEV business. We also deployed our products to support the dual energy strategy, namely EREV and BEV, which effectively complement each other and meet the diverse needs of our users. A key highlight that's worth mentioning is that we have made breakthroughs in key regional markets. The I series has successfully entered core BEV markets such as Beijing, Shanghai, Jiangsu, and Zhejiang, with orders in these areas starting to increase significantly from September. Li I8 and I6 are steadily going through the path of production ramp-up, delivery acceleration, and market penetration.

Starting in November, to address production ramp-up challenges, we will officially start to begin a dual supplier strategy for our batteries on Li I6. We'll ensure consistent performance and quality standards between these two suppliers. We expect monthly Li I6 production capacity to steadily increase to about 20,000 units starting from early next year. We sincerely apologize to customers who placed orders on I6 and are still waiting for their cars to be delivered. Due to constraints in the supply chain planning of key components and the pace of production ramp-up, your vehicle delivery schedule has been affected. We deeply appreciate your trust and choice of Li Auto, and we kindly ask for your continued understanding and patience. Our team is working around the clock to accelerate production and expedite the delivery process.

Speaker 2

For the second question, Tim, this is John. I think for the operating cash flow, it's about two reasons. First, as we guided in the last earnings release, the third quarter we faced great pressure on the deliveries, and the delivery decrease will make the revenue decrease, which will finally impact the operating cash flow. Also, the impact of shorten of the payment cycle to suppliers. This is, as you may know, due to the government's authority starting from June in the national line. Actually, we value our partnership with our supply chain partners and actively respond to their requirements. Currently, the settlement period for all our accounts payable is 60 days, and the payment is either the wire transfer or bank notes without any business notes or some kind of certificates from the yen, just the normal bank note. Thank you.

Speaker 6

Thank you. Your next question comes from Ming Hsun Li from Bank of America. Please go ahead.

Speaker 1

谢谢关理财给我的提问机会，我是美银证券的 Ming，那我这边就是有两个问题，一并请教。第一个问题，公司怎么看明年置换补贴政策的变化，以及购置税提高至5%的影响，如果置换补贴推播的话，公司相对应的销售策略会是什么。 My first question is that because next year the trade-in subsidy policy will change and also the EV purchase tax will increase from 0% to 5%, if the subsidy declines next year, what will be your sales strategy for 2026? 我第二个问题是因为我们明年2026年的 L 系列也会有重大的改款，请问有哪一些亮点可以期待。 My next second question is that in 2026, your EREV L series will have a new generation, so what can we expect the most, the new features, specs, and what will be the new advantages for your new models? 谢谢。 你好，我是马东辉，我来回答这两个问题。第一个问题是关于置换补贴以及购置税的问题。我们认为，这也是汽车行业从原有的政策驱动转向市场驱动的一个过程。这也是企业核心竞争力的一个凸显期。购置税政策的退坡，短期市场一定会产生波动，因为用户会在2025年年底集中去锁定优惠，然后会出现类似的跳尾效应，并且在2026年Q1的销量自然会出现环比的下调。但长期来看，我们还是维持乐观。NEV的渗透率也会稳步增加。2026年国内新能源渗透率我们预估可能会达到55%-60%。对于高端市场的NEV渗透率会突破60%。对于理想汽车的应对策略，首先在过渡阶段还是要保障用户的利益，并且我们2020年新开发的车型要适配新的法规规范。在过渡期，我们也推出了安心购车的方案，比如承担 I6 年底前所单用户跨年交付的购置税差额，并且我们二流款所有产品油耗电耗全部达成新政策的要求，可以享受2026年新政策的优惠。长期还是要强化用户的价值，通过技术的升维来对冲影响，没有捷径。2026年我们全系采取800伏高压平台以及5C的超充电池，提高效率降低能耗，并且在2026年我们也将建成大约4,800座超充站，并且高速充电桩的占比预计也会超过35%。以及通过深化供应链的国产化、规模效应、稳定的价格。对于产品力我们只有加速迭代再加速迭代。总结来说，这个政策退坡肯定是行业高质量发展的一个分水岭。理想2026年我们是有信心实现交付量历史的突破，然后以产品力和用户价值去穿越周期，巩固我们在高端市场的领先地位。谢谢。

Speaker 8

We believe this change marks the auto industry's transformation from policy-driven adoption to organic market-driven adoption, and it is precisely during this phase that the value of stronger players can really stand out. As the purchase tax policy phases out, there will be fluctuations in the first short term, we believe. We expect to see a pull forward effect, namely as customers rush to lock in their incentives at the end of 2025, that will naturally lead to a substantial dip in the deliveries in Q1 2026. Looking into the longer term, we are optimistic about the penetration rate of NEVs. In 2026, the NEV penetration rate in the domestic Chinese market will probably reach between 55-60%, with the rate in the premium segment exceeding 60%.

At Li Auto, our response strategy is to guarantee user benefits and adapt to new standards with our new vehicles rolling out during the transition period. For the transition period, we have a peace of mind purchase program, covering the purchase tax difference for I6 customers who locked in their orders in 2025 but take deliveries in 2026. All of our 2026 models meet the new standards for gas and energy consumption, so they will qualify for 2026 incentives. In the longer term, we will continue to be dedicated to user value and offset policy impacts through technology advancements. For example, we will be fully adopting 500-800 volt high voltage platform, and 5C ultra-fast batteries to enhance efficiency and reduce energy consumption in 2026. We aim to operate about 4,800 supercharging stations by 2026, 30%-35% of which will be on highway service stations.

We'll continue to deepen our supply chain localization and leverage economies of scale to stabilize pricing while accelerating product iteration to keep all of 2026 models at the forefront of product competitiveness. At the product strength, we must accelerate model iteration and accelerate further. In summary, this policy phase out marks a watershed moment for the industry's shift towards high quality development. Li Auto is poised to achieve a historic breakthrough in deliveries in 2026, and we will navigate this cycle through superior product strength and user value, thereby consolidating our leadership in the premium market.

Speaker 5

delivery rhythm, and certain user value to calmly respond to market uncertainty. Thank you. Usually on product release dates and more details we need to choose an appropriate time to release publicly, but today I still want to take the opportunity to give a glimpse on our products rollout for next year. The next year for L series is going to be a major generational upgrade, and the changes are based on deep research of users and their feedback, as well as our accumulation of technology over the years, and we want to build a very strong product that's also fundamentally different from the current generation. All this is to support our goal of reclaiming leadership in the EREV market in 2026. In terms of model configuration, we'll be going back to the simplified SKU approach, which balances market coverage as well as supply chain efficiency.

Even the base model will not compromise in terms of user experience. That will have all features as standard. In terms of design upgrades, while retaining our iconic design DNA, we will be upgrading the premium feel and craftsmanship. We will strike a balance between strong brand identity and fresh user appeal, and refine our products to better serve the needs of family users. On the core technology front, 5C standard supercharging will be standard on all models and seamlessly integrate with our existing charging network to efficiently address range anxiety. At the same time, we will reinforce our position as the EREV leader, building on our first mover advantage and deep expertise in EREVs. The 2026 L series refresh is about responding to market uncertainty with certainties on technological upgrades, delivery cadence, and user value.

We'll announce the specific launch timing and further details at the appropriate time. Please stay tuned. Thank you.

Speaker 6

Thank you. Your next question comes from Paul Gong at UBS. Please go ahead.

Speaker 2

and the future R&D targets and iteration progress? I have two questions. The first one is regarding the recall of the Mega. I noticed this was announced in Q4. Why are you booking it in Q3 and how did you determine the amount and the sharing between yourself and the supply chain? What is the impact for the Q4 gross profit margin? If possible, please also update us about the latest situation of the recall as well as the latest order of Mega. My second question is regarding the AI. Can you please update us on the latest development of VLA, world model, and the user feedback? If possible, please also give more color for the future targets and upgrades process. Thank you. Paul, this is Johnny. I think I will, due to time constraint, respond to your question very shortly. First, we recognize this in Q3 as we regard this event as a subsequent event, so it will be accrued in the most recent quarter we can recall. It is basically a calling center. For the recall, I think we have an announcement and I do not want to repeat most of the details covered. Currently, we just make all the battery pack to fulfill the recall requirement. That is the demand, which means we lower the delivery of our 2025 Mega deliveries. Which means all the battery pack, we shift most of them to replace the 2024 recall. I think that best serves the customer's benefits and that is the company's value proposition for us.

Speaker 6

M100 的上车。后续我们也会有更多创新的功能陆续落地，推出行业首创防御型的 ES 功能，升级安全防护的能力。我们也在探索任意车位到任意车位全场景的泊车，并且结合我们自建的超充站去实现智能找车的功能，真正完善智能出行的生态。谢谢。

Speaker 5

We rolled out our VLA driver to all of our AD Max vehicles in September, and with the strong migration capability of our model across all releases, all of our AD Max users have access to this new model, including the new I series users as well as the L9 users who bought the car back in 2022, and they're able to experience its core capabilities across the board. User feedback and data analysis have very clearly shown the effectiveness and the level of experience improvement. We can see that the L series and I series owners have a strong will, the I series owners have a stronger willingness to use VLA driver with both DAU and MPI showing improvements.

In the meantime, users generally report the VLA to be smoother, especially in longitudinal control, and more proactive and decisive in detours, and more accurate in route selection at complex intersections. With ongoing iterations, the functions of VLA will continue to achieve further breakthroughs. For OTA 8.0, which is our first full-scale rollout, the priority is mostly focused on safety, and in early December we will release OTA 8.1, which further enhances VLA perception capabilities for more precise and responsive behavior. By year-end, we will deploy an architecture upgrade to strengthen language behavior interaction and streamline the decision-making process, which will be compatible with our upcoming in-house developed M100 chip.

Beyond core system improvements, we're rolling out a series of innovations, including the industry's first defensive driving AES feature to enhance safety capability and any point-to-any point full scenario automated parking and a smart finder to find charging stations and park automatically. All this completes the smart mobility ecosystem. Thank you.

Speaker 6

Thank you. Your next question comes from Tina Hu from Goldman Sachs. Please go ahead.

Speaker 0

好，感谢管理层接受我的提问。我是高盛的 Tina 啊，由于时间关系，我就问最后一个问题啊。我想请问一下，就是刚才公司也提到咱们自研的芯片啊，所以想问一下这个自研芯片以及自研操作系统啊，咱们开源的一个进展和后续的规划。 Thanks management for taking my question. I just have one question, what is the progress in terms of our in-house developed SOC as well as the operating system, and then the progress in terms of open source and then future development? Thank you.

Speaker 5

Hi Tina, let me answer your question. We believe that a system for a system is a core foundation for intelligent vehicles. To achieve this efficiency, the system must be designed as integrated architecture, not as separate parts. Our in-house design controller hardware and the operating system have enabled us to reduce development time from industry average of 15 months to 9 months, while lowering cost by 20%. Many modules in the inference stack still come from suppliers. To innovate faster together, we open sourced Halo OS, enabling collaborative development with our partners and ecosystem. In September, we established the Halo OS technical steering committee, and the system companies across intelligent vehicle value chain signed the community charter, including OEMs, chip makers, software and hardware service providers, and component suppliers. At the same time, we are undergoing our own vehicle foundation model for physical AI.

Our focus is to improve perception, understanding, and response, so the model can see further, understand better, and react faster. The AI inference chip is the computing engine of this system. Our controller built with our in-house design chip M100 is now undergoing large-scale system testing. We expect commercial development to take place the next year. Co-designed with our foundation model, compiler, and software system, we expected the M100 within our next generation VLA-based autonomous driving system to achieve at least three times the performance-to-cost ratio of today's high-end chips. On the basis of highly efficient AI inference and execution systems, our next priority will be faster iteration, continuous performance improvement, and lower cost. Development of our next generation platform and chip has already begun. Thank you.

Speaker 6

Thank you. As we are now reaching the end of our conference call today, I would like to turn the call back over to the company for closing remarks. Ms. Janet Zhang, please go ahead.

Speaker 2

Thank you once again for joining us today. If you have further questions, please feel free to contact Li Auto's investor relations team. That is all for today. Thank you.