I. The Historical Context of the Automotive Industry
Looking back at the history of the automotive industry, it can be clearly divided into the era of traditional automobiles, the era of software-defined vehicles, and the emerging era of autonomous intelligence. In the nearly 100-year development of traditional automobiles, their R&D approaches and methodologies remained relatively stable, with slow changes. At that time, automobiles were considered reliable mechanical products, and the industry focused on controlling product quality, ensuring the stability and reliability of the vehicle's mechanical performance. From the perspective of automotive professionals, there was very little consideration given to product design and functional optimization from a human perspective.
With the advancement of technology, the application of software in the automotive field has gradually emerged, giving rise to the concept of software-defined vehicles, a phase that began approximately six or seven years ago. The integration of software has endowed vehicles with more functionality and flexibility, transforming them from simple combinations of mechanical parts into vehicles possessing certain intelligent characteristics. However, before people fully understood the implications of software-defined vehicles, the emergence of large-scale modeling technology brought a new wave of change to the automotive industry. With its powerful data analysis and processing capabilities, large-scale models have injected new vitality into the intelligent development of automobiles, opening the door to the era of autonomous intelligence.
In just ten years, the automotive industry's methodology and people's understanding of automobiles have undergone two major leaps. This rapid change has placed extremely high demands on industry practitioners, requiring them not only to continuously learn and update their knowledge, but also to deeply understand the core characteristics of each era and the intrinsic connections between them, because the problems and solutions faced by each era lay the foundation for the development of the next era.
Further examining the key milestones in the development of the automotive industry, 2015 and 2025 stand out as two significant turning points. In 2015, Elon Musk and his innovative vision had a profound impact on the automotive industry, driving a disruptive shift in automotive R&D thinking. Prior to this, the traditional automotive industry followed a relatively fixed R&D model; since then, the industry has accelerated its progress towards intelligent and electric vehicles.
2025 was originally predicted to be a key turning point for the second wave of transformation, and although the actual transformation may have occurred earlier, it is roughly within this period. The emergence of ChatGPT has attracted widespread attention globally and has brought new directions and technological approaches to the intelligent development of the automotive industry. ChatGPT's successful expansion has made people more aware of the enormous potential of artificial intelligence technology, which has also prompted the automotive industry to accelerate its exploration of how to apply similar technologies to areas such as autonomous driving, driving the development of automobiles towards a higher level of intelligence.
From the perspective of the subject-object relationship at different stages, the development of the automotive industry presents a clear evolutionary trajectory. In the traditional automotive era before 2015, it was a stage where the object revolved around the subject. Here, the object refers to the automotive product, and the subject is the engineers and all personnel involved in research and development. During this period, product design and development primarily revolved around the engineers' ideas and capabilities. The product gained market acceptance, and there was no core conflict between the engineers and the product. Because user market demand was relatively stable, and requirements for automotive functions and experiences had not yet reached today's levels, engineers could effectively control the product development process through established procedures, resulting in a relatively regular work rhythm. For example, in the automotive industry at that time, it was common to finish get off work at 4:30 or 5:00 PM.
In recent years, the automotive industry has entered a stage where the subject revolves around the object. With intensifying market competition and constantly changing user needs, industry competition has become increasingly fierce, and engineers have gradually become relegated to a passive position, revolving around the product itself. The characteristics of the product itself have become the dominant factor, determining the entire R&D and production process. During this period, whether it's the application of domain controllers, the development of centralized architectures, or the advancement of other related technologies, it all reflects the engineers' lack of control over the product development process. For example, to meet the ever-evolving functional requirements of products, engineers' working hours have been continuously delayed, with 9 or 10 pm becoming the norm, and the phenomenon of involution within the industry becoming increasingly apparent.
The automotive industry appears to be trending towards a disappearance of the human element. This "disappearance of the human element" doesn't mean engineers and other human agents no longer exist, but rather emphasizes the profound changes in the relationship between humans and products under the trend of highly intelligent development. As products become increasingly intelligent, they can autonomously complete many complex tasks, reducing reliance on human intervention to a certain extent. This trend reflects the automotive industry's shift from a human-led development model to a more intelligent and automated one, indicating that future cars will possess greater autonomy and adaptability.
Corresponding to the evolution of the subject-object relationship, the automotive industry has had different key themes at different stages: controllable quality, flexible iteration, and autonomous growth. In the traditional automotive era, controllable quality was the core focus. At that time, the automotive industry emphasized the mechanical quality and stability of products, placing time and user experience (based on smartphones) in a relatively secondary position. However, in the era of software-defined vehicles, with intensified market competition and diversified user needs, flexible iteration became crucial. Companies needed to continuously update the software functions of their products based on market feedback and technological developments to meet the ever-growing demands of users. Entering the era of autonomous intelligence, autonomous growth has become the pursued goal. Automobiles not only need the ability to iterate flexibly but also the ability to learn and evolve on their own, autonomously optimizing their performance and functions according to the constantly changing environment and user needs.
These three keywords, seemingly simple, contain profound implications and permeate the entire development of the automotive industry. From a first-principles perspective, a deep understanding of these keywords helps resolve various conflicts and problems encountered in practice. For example, applying the standards of flexible iteration to traditional automobiles will inevitably lead to cognitive biases, as the R&D and production models of traditional automobiles are ill-suited to the demands of rapid iteration. Similarly, applying the requirements of independent growth to flexible iteration may also face numerous challenges, as the technologies and philosophies involved differ. Therefore, when analyzing and solving problems in the automotive industry, it is essential to fully consider the keywords at different stages and the logical relationships behind them.
II. Evolution of Technology and Architecture in the Automotive Industry
The transformation of the automotive industry is not only reflected in changes in development philosophy and the relationship between subject and object, but also significantly in the evolution of vehicle architecture, core components, chip composition, software technology, and business models.
In terms of vehicle architecture, it has evolved from distributed architecture to centralized architecture, and then to a self-closing architecture. In the era of traditional gasoline-powered vehicles, the vehicle architecture was characterized by its distributed nature, with a large number of components and a relatively complex structure. With the development of automotive intelligence and electrification, centralized architecture has gradually emerged. This architecture integrates components with similar functions, reducing the number of parts, making the overall vehicle structure simpler, and improving system integration and coordination efficiency. For example, in the application of domain controllers, multiple related control functions are centralized in a single controller, achieving centralized management and control of certain vehicle functions.
Further development suggests that a self-closing architecture is becoming the future trend. This architecture not only achieves high integration at the hardware level but also enables the system to self-monitor, self-adjust, and self-optimize through software algorithms, giving vehicles greater autonomy and adaptability. From gasoline vehicles to electric vehicles, and then to robotic extended architectures, the development trend of vehicle architecture is a gradual reduction in the number of parts, making vehicle assembly and maintenance more convenient, and potentially even as simple as assembling a computer in the future. This development process not only improves production efficiency and reduces costs but also provides a better hardware foundation for the intelligent upgrade of automobiles.
The importance of core components in the automotive industry is becoming increasingly prominent. Just as computer assembly occupies a lower position in the overall computer supply chain, the importance of automobile assembly in the automotive supply chain is gradually decreasing, while centralized components and key software are becoming the focus. With the increasing centralization of architecture, the quality and performance of key components and software directly affect the overall quality and functionality of a vehicle. For example, high-performance chips, advanced sensors, and intelligent software systems have become key factors in enhancing automotive competitiveness.
The composition of chips has also undergone significant changes in the development of the automotive industry. Early on, the automotive industry primarily used MCUs (general-purpose CPUs), whose functions were relatively simple, mainly used to implement some basic control functions. With the increasing demands for automotive intelligence, hybrid SoCs (CPU+GPU) have gradually been adopted. These can simultaneously handle complex computing and graphics processing tasks, providing more powerful computing support for intelligent driving assistance systems and other applications. Under a closed-loop architecture, the proportion of ASICs (custom-designed chips) is expected to gradually increase and eventually become dominant. ASIC chips are custom-designed according to specific application requirements, achieving a better balance in performance, power consumption, and cost, making them more suitable for the high-performance, low-power, and high-reliability requirements of automotive intelligence.
The application of software technology in the automotive industry has evolved from simple to complex, and from auxiliary to core. In the traditional automotive era, software mainly existed in the form of rules plus a few models, its function primarily being to implement some basic control logic and auxiliary functions. As the level of automotive intelligence increased, software gradually evolved into a form of models plus a few rules. By introducing machine learning and deep learning models, cars gained a certain degree of intelligent decision-making capabilities. In the era of autonomous intelligence, end-to-end models have become the mainstream, and software can directly output the final decision result based on the input sensor data, achieving more intelligent and automated control.
In terms of cloud platforms, their importance is increasingly prominent with the development of automotive intelligence. Cloud platforms not only provide vehicles with powerful computing and storage capabilities, but also enable information interaction and sharing between vehicles (V2V) and between vehicles and infrastructure (V2I). For example, through vehicle-cloud closed-loop FOTA (Firmware Over-The-Air) and SOTA (Software Over-The-Air) technologies, vehicles can obtain the latest software versions in real time, enabling functional updates and optimizations; data management platforms can collect, store, and analyze the large amounts of data generated by vehicles, supporting intelligent decision-making and personalized services; and multimodal large-scale model platforms provide more powerful algorithmic support for the intelligent development of vehicles, enabling them to better understand and process various types of data, such as images and voice.
The business model has also undergone significant changes in the automotive industry. In the traditional automotive era, revenue was primarily driven by hardware sales, with automakers profiting from the sale of automotive hardware. As software has become increasingly important in automobiles, a model has emerged where hardware sales are the main revenue driver, with some automakers beginning to charge separately for certain software features. In the future, a model where software is charged for while hardware sales can break even or even incur losses is gradually gaining traction. Take Tesla as an example: its software revenue generation is not only for direct economic benefits, but more importantly, it aims to reduce product costs, increase market coverage and user numbers, and thus collect vast amounts of data. This data has become Tesla's core competitive advantage in the field of artificial intelligence, providing strong data support for its subsequent robotics and other industries. In China, although software revenue generation is not yet fully widespread and is mostly included in the price of automotive products, with the development of the industry, software revenue is expected to become one of the important profit models in the automotive sector.
III. Application and Development of Large-Scale Models in the Automotive Industry
Large-scale models play a crucial role in the development of the automotive industry, especially in the field of autonomous driving, where they provide new technological paths and solutions for achieving higher levels of autonomous driving.
In the early stages of autonomous driving technology development, rule-based algorithms were the primary reliance. Engineers formulated a series of rules and logic to enable cars to make corresponding decisions in specific scenarios. For example, stopping at a red light or slowing down when an obstacle is detected ahead. However, these rule-based algorithms have significant limitations. They struggle to cope with complex and ever-changing real-world traffic scenarios, often failing to make accurate decisions when encountering irregular obstacles, special traffic signs, or unexpected events.
With the development of machine learning technology, it has been widely applied in the field of autonomous driving. Machine learning algorithms can automatically extract features and patterns from large amounts of data through learning, thus enabling cars to cope with complex scenarios to a certain extent. In image recognition, machine learning algorithms can identify different types of vehicles, pedestrians, and traffic signs. However, machine learning algorithms also face some challenges, such as strong dependence on data and limited generalization ability of the models.
The emergence of end-to-end algorithms represents a significant breakthrough in the development of autonomous driving technology, while the application of large-scale models provides robust support for their implementation. End-to-end algorithms directly use sensor data as input to the model, which then processes the data and outputs the final driving decisions, such as steering angle, acceleration, or deceleration commands. Large-scale models, with their powerful learning capabilities and ability to process complex data, can better learn and understand driving behavior patterns in various traffic scenarios, thereby achieving more accurate and intelligent driving decisions.
The application of large-scale models in autonomous driving is mainly reflected in several aspects. At the perception level, large-scale models can process and analyze data from sensors such as cameras and radar more accurately, identifying the shape, position, and motion state of various objects, thus improving the accuracy and reliability of perception. Through learning from large amounts of image data, large-scale models can accurately distinguish different types of vehicles, pedestrians, and road signs, and can even identify some blurry or occluded objects. At the decision-making level, large-scale models can comprehensively consider various factors, such as traffic rules, road conditions, and vehicle status, to make more reasonable driving decisions. When encountering complex traffic intersections, large-scale models can select the optimal driving route and speed based on real-time traffic conditions. At the planning and control level, large-scale models can generate smoother and safer driving trajectories and precisely control the vehicle's power, steering, and other systems to ensure the stability and comfort of the vehicle during driving.
From the development history of large-scale models, it has gone through several important stages. In 2015, deep learning began to emerge in various fields, and gradually found applications in autonomous driving. People began to recognize the potential of deep learning in processing complex data and achieving intelligent decision-making. In 2017, AlphaGo's victory over human Go players caused a global sensation, further demonstrating the powerful capabilities of artificial intelligence technology and injecting new momentum into the development of large-scale models. In 2022, the emergence of ChatGPT further propelled large-scale model technology beyond its core, attracting widespread attention and application. ChatGPT demonstrated the outstanding capabilities of large-scale models in natural language processing, providing reference and ideas for the application of large-scale models in other fields. In 2024, related technologies continued to develop, with institutions such as OpenAI continuously releasing new achievements. In the field of autonomous driving, the FSD (Full Self-Driving Capability) large-scale model also made significant progress. These developments have not only driven the continuous progress of large-scale model technology but also accelerated its application and promotion in the automotive industry.
In recent years, large-scale models have seen rapid development in the automotive industry, attracting a surge of capital investment. Nvidia, for example, has witnessed a significant increase in its market capitalization, reflecting strong investor confidence in the future prospects of large-scale models and related technologies in the automotive sector. In the United States, many companies and teams originally focused on autonomous driving research have shifted their focus to research and applications related to large-scale models, demonstrating a clear trend of capital migration. In China, the government also attaches great importance to the development of related technologies. The "mental productivity" mentioned at the Two Sessions is largely related to the development and application of artificial intelligence technologies such as large-scale models, indicating that China is actively planning and promoting the application and development of these technologies in the automotive industry and other fields.
IV. Problems and Challenges Facing the Development of the Automotive Industry
Despite significant progress made by the automotive industry driven by autonomous driving and large-scale modeling technologies, it still faces numerous problems and challenges in its development.
From a technical perspective, the rapid pace of technological updates is a major challenge facing the automotive industry. In the current stage of development, new technologies emerge endlessly, and research and development results are updated extremely quickly. Often, a technology, once developed and put into application, is quickly superseded by a new one. This necessitates that automotive companies continuously invest significant human, material, and financial resources in technological research and development to maintain technological advancement. Simultaneously, engineers need to constantly learn and master new technical knowledge, or risk becoming obsolete. For example, in the research and application of large-scale model technology, engineers need to keep abreast of the latest algorithms and model architectures, continuously optimizing and improving the technology to adapt to rapidly changing market demands.
Data security and privacy protection issues are becoming increasingly prominent. As vehicles become more intelligent, they collect vast amounts of data during operation, including users' personal information, driving habits, and location information. The security and privacy protection of this data are crucial; data breaches not only infringe on user privacy but may also threaten users' lives and property. Automotive companies need to establish robust data security management systems and strengthen technologies such as data encryption and access control to ensure data security and privacy.
While significant progress has been made in the application of technologies such as large-scale models in autonomous driving, several technical bottlenecks remain. The reliability and stability of autonomous driving systems in complex scenarios need improvement. For example, sensor performance can be affected by adverse weather conditions (heavy rain, dense fog, etc.), leading to decreased perception accuracy. In extreme traffic situations, autonomous driving systems may fail to make accurate decisions. Furthermore, the legal and ethical issues surrounding autonomous driving technology require further discussion and resolution, such as the determination of liability in the event of a traffic accident.
From a market and business perspective, the adoption of paid software models in the automotive industry still faces certain challenges. In China, while the concept of paid software is gradually being accepted, it hasn't been fully implemented yet, with most software costs still included in the price of vehicles. This is primarily because consumer acceptance of paid software needs further improvement, and assessing the value of software also presents certain difficulties. Automakers need to explore more reasonable software pricing models to increase consumer acceptance.
Competition in the automotive industry is intensifying, not only among traditional automakers but also from cross-industry competition from technology companies. Leveraging their technological advantages in areas such as artificial intelligence and big data, technology companies are rapidly entering the automotive sector, posing a significant challenge to traditional automakers. Traditional automakers need to strengthen cooperation with technology companies, integrating the strengths of both sides to enhance their competitiveness.
From the perspective of talent cultivation and industry development, the rapid transformation of the automotive industry has placed higher demands on talent. There is a need for interdisciplinary talents who possess both automotive engineering expertise and mastery of emerging technologies such as artificial intelligence and big data. However, there is currently a relative shortage of such interdisciplinary talent, and the pace of talent cultivation is struggling to keep up with the industry's development needs. Universities and vocational education institutions need to adjust their curricula, strengthen the development of relevant majors, and cultivate more talent to meet the industry's evolving needs.
Furthermore, the development of the automotive industry also requires consideration of supporting infrastructure. The development of autonomous driving technology necessitates the support of infrastructure such as high-precision maps and reliable communication networks. Currently, the construction of such infrastructure is not yet fully developed, which to some extent limits the promotion and application of autonomous driving technology. Governments and enterprises need to increase investment in infrastructure construction to create favorable conditions for the development of the automotive industry.
