AI takes over from the "three electric systems" and becomes a new battleground for automakers.

After years of popularizing new energy, a quiet revolution is taking place in the automotive industry. The competition among car - makers seems to be gradually shifting from traditional vehicle - manufacturing skills to a fierce competition in AI technology.

Not long ago, at the new product launch event, Li Auto didn't focus too much on the vehicle's hardware parameters. Instead, it spent a great deal of time introducing the new progress in vehicle - machine intelligence and the evolution path of intelligent driving under VLA technology.

Meanwhile, Geely Auto, in collaboration with Jieyue Xingchen, jointly released the next - generation intelligent cockpit Agent OS (preview version) "Intelligent Egg Cabin", which is natively built for AI Agents. The innovative AI interaction experience and powerful functions have also attracted wide attention both within and outside the industry.

The core messages of these press conferences clearly convey that AI is no longer an "optional" feature for vehicles. Instead, it has become the core selling point for defining product experience, building brand moats, and attracting consumers.

Gao Rui, the deputy general manager of GAC Group, also stated bluntly at the China EV 100 Forum (2025): "Without intelligent driving capabilities, there is no ticket to participate in future competition. This has become a common consensus in the industry."

These phenomena inevitably make people wonder: Have car - makers shifted from a simple vehicle - manufacturing competition to a comprehensive competition in AI capabilities?

01

Betting on AI - powered "Intelligent Manufacturing" Upgrade

With the popularization of the electrification wave, the core technological barriers in the automotive industry have been significantly lowered. Car - makers represented by BYD have greatly reduced the threshold for vehicle manufacturing by virtue of their mature supply chains for the three - electric systems. Against the backdrop of similar hardware and excessive performance, simply relying on "stacking features" can hardly build a long - lasting competitive advantage.

Moreover, as the popularity of cars increases and consumers' awareness deepens, users' demands are undergoing profound changes. They are no longer merely satisfied with the basic functions of vehicles as a means of transportation. Instead, they pursue higher - level emotional experiences and personalized services. The car is transforming from a "moving machine" to a "third living space", an intelligent terminal where people can work, entertain, and rest.

In this context, the rise of AI technology provides an excellent opportunity for car - makers to leap from "manufacturing" to "intelligent manufacturing". AI can not only reconstruct users' travel experiences through intelligent driving and intelligent cockpits but also penetrate the entire life cycle of R & D, production, marketing, and service, achieving cost reduction and efficiency improvement. More importantly, mastering AI means mastering the right to define vehicles with data and software, and seizing the entrance to the next - generation travel ecosystem.

In fact, the maturity of technology and the decline in cost are driving the rapid popularization of AI functions in automotive products. The year 2024 is regarded by many industry analysts as the "Year of Intelligent Driving". Marked by Tesla's first implementation of an end - to - end autonomous driving solution, domestic car - makers such as NIO, XPeng, Li Auto, and Hongmeng Zhixing have also successively launched similar technologies.

For example, Li Auto's "end - to - end + VLM (Visual Language Model) dual - system" has become its unique technological advantage. The end - to - end system can achieve a rapid response to the environment, while the VLM visual language model is responsible for high - level analysis. The organic combination of the two significantly improves the safety and scene generalization ability of autonomous driving.

Similarly, Geely Auto's Qianli Haohan system is constantly evolving. It proposes the concept of "AI training AI, AI testing AI" and plans to implement the technical architecture for L3 in the fourth quarter of this year, promoting the actual implementation and application of L3 - level technology and enabling users to enjoy a higher - level autonomous driving experience earlier.

Empowered by AI technology, the intelligent cockpit has also achieved a major transformation from an "instruction executor" to an "emotional intelligent agent". The super - anthropomorphic in - vehicle AI intelligent agent of Geely Galaxy M9, based on Jieyue Xingchen's end - to - end large - scale AI voice model, can not only achieve multi - modal interaction and accurately perceive users' emotions but also actively provide services for users according to different scenarios.

SAIC - GM globally launched the Qualcomm 8775 cockpit chip and built an AI central hub integrating the cloud and the terminal, achieving cross - scenario intention understanding and making the in - vehicle interaction experience more smooth and natural. Li Auto's intelligent agent, Li Auto Mate, has even achieved a magnificent leap from a "vehicle control assistant" to a "mobile life butler".

Undoubtedly, large - scale AI models, especially multi - modal large - scale models, enable vehicles to have the ability to leap from "perceptual intelligence" (recognizing objects) to "cognitive intelligence" (understanding scenarios and intentions), laying the foundation for achieving a higher - level autonomous driving and a more user - friendly intelligent cockpit interaction.

An obvious trend is that high - level intelligent driving functions such as high - speed NOA and urban NOA are gradually shifting from being optional features for high - end models to becoming standard features in the mainstream price range below 200,000 yuan. The application of large - scale AI models in intelligent cockpits is also becoming more and more widespread. From providing more natural voice interaction to actively recommending services according to users' habits, AI is comprehensively enhancing users' experiences. This transformation from "optional" to "standard" and then to a "quasi - core selling point" indicates that AI has become a key variable in determining the competitiveness of automotive products.

02

AI Also Needs to Compete in Differentiation

Currently, leading enterprises are competing for the high - ground through differentiated technological routes. The core battlefields are focused on three major areas: intelligent driving, intelligent cockpits, and the integration of AI throughout the entire life cycle.

First of all, intelligent driving is the most intense and most - watched battlefield in the car - makers' AI competition. Major manufacturers have invested huge amounts of money, either through self - research or cooperation, to promote the evolution of intelligent driving technology from L2 - level assisted driving to L3 - level and higher - level autonomous driving. The technological routes also show a diverse trend, from multi - sensor fusion to pure - vision solutions, from modular architectures to end - to - end large - scale models. Each company is exploring the optimal path to the ultimate goal of autonomous driving.

Li Auto's layout in the field of intelligent driving reflects its strategic determination to "go all in on AI". Its proposed VLA (Visual Language Action Model) technology aims to enable vehicles to perceive the environment through vision, understand intentions through language, and finally translate them into driving actions, just like humans.

The core of this technological route is to train an end - to - end large - scale model with a large amount of data, enabling it to handle complex and unstructured road scenarios. Lang Xianpeng, the vice - president of intelligent driving R & D at Li Auto, once said that the company expects the training computing power to exceed 8 EFLOPS by the end of 2024, with the cumulative training mileage exceeding 3 billion kilometers. He also believes that the computing power required for autonomous driving training will ultimately reach the level of 100 EFLOPS. This huge investment in computing power is precisely to support the rapid iteration and evolution of its VLA model, ultimately achieving a safer and more anthropomorphic autonomous driving experience.

In addition to Li Auto, XPeng is one of the new forces in China that first laid out intelligent driving. Its technological route is centered on "full - stack self - research". At the hardware level, XPeng has invested in self - researching AI chips in order to achieve in - depth collaborative optimization of hardware and software. BYD also takes advantage of its three - electric systems in its vehicle - wide intelligent strategy. Through the Xuanji architecture, it has achieved in - depth integration of electrification and intelligence, enabling all parts of the vehicle to be uniformly scheduled and controlled by AI.

If intelligent driving liberates users' hands and feet, then the intelligent cockpit is committed to liberating users' minds, providing a more emotional and personalized human - machine interaction experience. The application of large - scale AI models is driving the evolution of intelligent cockpits from the simple stacking of functions in the past to "emotional intelligent agents" that can understand users and provide active services. Many large companies are also following up.

The Agent OS (preview version) "Intelligent Egg Cabin" jointly released by Geely Auto and Jieyue Xingchen, and NIO's intelligent cockpit system NOMI are representatives of intelligent cockpit systems natively built for AI Agents. According to the official description, this kind of AI intelligent agent no longer passively executes instructions but can actively perceive users' needs and provide personalized services. For example, it can plan travel routes in advance according to users' schedules and recommend suitable music or ambient lights according to users' emotional states.

Picture/NIO's intelligent cockpit system NOMI

This transformation from "functions" to "intelligent agents" indicates that the intelligent cockpit is entering a new stage of development. It also sends a signal to the industry: In the future, only those who can build an AI Agent ecosystem that attracts developers and deeply engages users can build a moat.

Of course, in addition to the parts that consumers can clearly perceive, the entire automotive field, including R & D and design, production and after - sales, marketing and service, is also actively promoting AI - enabled transformation. Whether it is AI simulation test - driving, AI technology detection, or AI customer service, they will all be important parts of enterprises' foundation - building in the future.

Undoubtedly, AI and large - scale models are becoming one of the important selling points for car - makers. This forces car - makers to quickly build their AI capabilities and enrich the concept of "full - domain AI". However, although AI depicts a beautiful blueprint for the automotive industry, the road to the future is not all smooth.

03

Cold Thinking Behind the Frenzy

It is not difficult to see from the increasing investment of major car - makers in AI that almost all of them are making attempts on mid - to - high - end models, which may be related to the current real - world difficulties.

Technology is the cornerstone for realizing all visions. The huge gap in current computing power infrastructure has led to bottlenecks for car - makers in the field of AI. Wang Yao, the deputy chief engineer of the China Association of Automobile Manufacturers, once pointed out sharply: "The total number of AI (chip) graphics cards of all domestic car - makers is less than that of Tesla's Dojo."

This statement reveals the "generational gap" in computing power between domestic car - makers and global top players. Tesla's self - developed Dojo supercomputer is specifically designed to process about 160 billion frames of video data collected by its global fleet every day, providing powerful computing power support for its pure - vision autonomous driving solution. Although domestic car - makers such as XPeng and Li Auto are also actively building their own super - computing centers, there is still a large gap in overall scale and investment.

Picture/Tesla's self - developed Dojo supercomputer

In addition to computing power, the construction of a data closed - loop is also a huge challenge. The training and optimization of AI models rely on a large amount of high - quality data. How to efficiently collect, annotate, process, and apply data to form a benign data closed - loop is the key to determining the evolution speed of AI capabilities.

Sun Hui, the technical director of the Intelligent Connected Vehicle Center of the Suzhou Automotive Research Institute of Tsinghua University, believes that in today's increasingly homogeneous algorithms, data will become the next key competitive point. Enterprises with millions of real - vehicle data can solve the "interaction game" problem and optimize "long - tail scenarios" through training with a large amount of real - world road conditions, thus building a competitive advantage that is difficult to replicate.

Moreover, in most cases, the hardware and software of intelligent driving and intelligent cockpits are independent of each other. The in - depth integration of the two to build a fusion solution represented by "cockpit - driving integration" can combine the computing platforms of the intelligent driving domain and the cockpit domain into one. However, this also means more excellent algorithm iterations, which also require continuous capital investment in the early stage.

Another aspect is that the development of AI technology is promoting the transformation of the automotive industry's business model from the traditional "one - time product sales" to "continuous service sales". The value of software and services will become increasingly prominent.

However, whether it is algorithms, computing power, or service value, they all come with extremely high costs. These costs also need to be spread across products, especially in the early stage of development. Owners who focus on mid - to - high - end models are often less sensitive to price, so these products bear the heavy responsibility of bringing AI into vehicles.

However, when we shift our focus from the cloud - based AI models to the solid asphalt roads and from the core business districts of first - tier cities to the broader second - and third - tier markets, we will find that for the vast majority of consumers who regard cars as "means of transportation", the tangible physical experiences - space and handling - are still the most important factors in their purchasing decisions.

This seemingly contradictory phenomenon precisely reveals the deep - seated logic of the automotive industry's intelligent transformation: AI is not intended to subvert the essence of cars as "means of transportation" but to inject new possibilities into travel on the premise of consolidating the basic experience.

When high - level intelligent driving functions penetrate into the mainstream market of 200,000 yuan, when the AI assistant in the intelligent cockpit can accurately remember the preferences of each family member, and when the AI - optimized supply chain continuously improves the cost - performance of vehicles - all these technological advancements should ultimately return to the essence of "making travel more beautiful".

This shift from "vehicle - manufacturing" to "AI - manufacturing" is essentially a leap for the automotive industry from "product defined by hardware"