The second 'bend' of Chinese automobiles

In the 1990s in Shanghai, the first Santana sedan drove off the production line, which was a Sino German joint venture product and was extremely popular. That is a representative work of 'market for technology'. Subsequently, more international car brands entered China, and joint venture brands became the dominant players in the Chinese car market.

The drawbacks are gradually exposed. The foreign party has adopted strict control strategies over the technology research and development of the joint venture company under the pretext of ensuring the quality of automobiles. The joint venture company does not have the right to modify automobiles, and all issues in the localization process require approval from the foreign party. By 1990, the domestication rate of several major joint venture car brands in China was as high as 60% and as low as 13%, far below expectations.

The market has gone out, but technology has not been exchanged back. Moreover, multinational car companies are keen on dumping outdated models in the Chinese market. So, technology is controlled, the market is divided, and profits are distributed. Chinese cars were once in a state of confusion.

By 2024, the total number of new energy vehicles in China will reach 31.4 million, increasing from an annual output of one million to becoming the world's first country to achieve an annual output exceeding 10 million. Chinese brands that were once criticized for being "low-quality and cheap" are now dominating the global electric vehicle market.

So, the market does not exchange for technology, on the contrary, only by mastering technology independently can we win the market. The competition in the market is essentially a competition for technological leadership.

Now, for the global automotive industry, a new battlefield has emerged: intelligent assisted driving.

Curve reproduction

The automotive industry is undergoing a critical stage of transition from electrification to intelligence. In the first half of 2025, the installation rate of L2 level and above intelligent assisted driving 

vehicles in China will reach 67.8%, and intelligence will become the core driving force of the industry.

We have reached the second bend - intelligent assisted driving, but this time the track is even more difficult, "Chinese automotive industry practitioners hold a similar view.

However, in the face of higher requirements for L3 level conditional intelligent assisted driving, the collection and management of EB level data is like a "frame" in the ocean fishing;

The demand for 100 Eflops of computing power leaves even the most advanced servers struggling to catch their breath; Simulate testing scenarios of billions of kilometers to verify system reliability; 

Ensuring the safety of millions of intelligent vehicles operating simultaneously in the future;

There is also continuous innovation and collaboration in the technology ecosystem These are the towering mountains that stand on the road of "intelligent assisted driving technology overtaking".

The former students have now become competitors to the teachers, and China's automotive industry has transformed from a pursuer to a leader. The Chinese automotive industry needs to put in more effort to maintain its leading position.

According to global market observations, the overseas automotive industry has regarded intelligent assisted driving as the "high ground" of future industry competition. The United States has opened up fully autonomous testing licenses through the Autonomous Driving Act in California, Nevada, and other states, and the federal government has incorporated intelligent assisted driving into its national transportation strategy.   

The European Union has mandated that new cars be equipped with L2 level assisted driving functions as standard under the General Safety Regulation, and has set up special funds to support L4 level technology research and development; Japan accelerates the commercialization of L3/L4 through the revision of the Road Transport Vehicle Law; The South Korean government is investing billions of Korean won to build an intelligent transportation demonstration zone.

Traditional car companies such as Toyota, Volkswagen, General Motors, etc. invest billions of dollars in research and development budgets every year to build technological barriers through independent research or cooperation. The Internet giants Google, Apple and Amazon are also crossing the border, forcing the transformation of traditional automobile enterprises with AI and data advantages. Not to mention in car chips, LiDAR, high-precision maps, etc., there are strong enemies lurking in every link.

The bend is coming, but I don't know who will overtake?

Why do we need clouds?

In the era of intelligent assisted driving, the competition dimension of the traditional automotive industry has shifted from mechanical performance and supply chain efficiency to intelligent competition based on data as fuel, algorithms as engines, and cloud platforms.

Shi Jianhua, Vice President of Chebaihui Research Institute and President of Chebaiku Research Institute, said that with the deep integration of automobile and AI technology, the demand for AI computing power in the automobile industry has significantly increased. The rapid iteration of assisted driving technology has driven exponential growth in demand for flexible and adjustable centralized cloud computing power, which has become an important growth point in the national cloud computing market, second only to large models. The continuous deepening of AI transformation in automotive companies is driving the acceleration of computing power to the cloud, gradually building the core assets of enterprises in the AI era.

Intelligent assisted driving has spurred the demand for car companies to go to the cloud, and cloud computing power has also driven the evolution of intelligent assisted driving in reverse. Taking data as an example, in order to improve the intelligence level of automobiles, car companies continue to increase the accuracy of onboard perception terminals, with camera pixels increasing from 8 million to 64 million, generating a large amount of data on the automotive end.

At the same time, the sales of intelligent connected vehicles are expected to increase to 5 million by the end of the year, with a large number of cars driving on the road every day, continuously generating data. In order to make the in car model more intelligent, the model parameters have also evolved from billions to billions, shifting from end-to-end to VLA (Visual Language Action) and world models, which further exacerbates the growth of data volume.

The data accumulated by car companies will range from PB level to EB level.

Taking the corner case scenario of intelligent assisted driving as an example, there are numerous scenarios such as a sudden drop of a steel pipe on the road and whether a car can intelligently avoid obstacles. These scenarios can aggregate into tens of millions, and they all need to be trained by models to achieve reasonable perception, prediction, planning, and control, requiring powerful computing power to support the training cluster.

In the future, cars will soon evolve from L2 level to L3 and L4 level intelligent assisted driving. In this process, the requirement for the model is to quickly precipitate the difficult scenarios encountered on the road every day, train in a timely manner, clarify the update direction of the model, and the model update cycle will be measured in days. This places extremely high demands on the real-time and sufficient computing power.

The computing power on the vehicle side is increasingly unable to meet the growing demand for intelligence mentioned above. To solve this problem, car companies have different investment directions. One is to place two chips on the car to increase computing power, but this increases the overall vehicle cost by 25%. The other is to seek external computing power support.

In contrast, once the in car chip is solidified, its upper limit of computing power is locked, unable to keep up with the rapid evolution of algorithm models and business scenarios. Cloud computing power has almost unlimited scalability and powerful iterative flexibility, and can continuously integrate the latest and most powerful AI computing chips. Vehicles can continue to receive globally leading computing power support throughout their entire lifecycle, and can enjoy the latest and most complex intelligent assisted driving and intelligent cockpit applications without replacing hardware, truly realizing the vision of "software defined cars".

A well-known intelligent assisted driving assistance company in China has made extensive use of Huawei Cloud's Ascend Cloud service. It is reported that the use of cloud computing power by Yinwang has already supported the intelligent flying of 1 million vehicles. Taking the current typical parking scenario as an example, by linking cloud computing power, on the one hand, the success rate of difficult parking cases can be increased by 15%, and on the other hand, with the help of cloud large models to optimize end-to-end parking efficiency, users can obtain a better experience.

Systematic Transcendence

The first cornering overtaking solved the "heart" and "body" problems of the car, achieving a leap in the power system. This is not a simple market victory, but a systematic overtaking driven by independent technological innovation, from a pursuer to a parallel runner or even a leader.

So, the second overtaking on the bend is to endow the car with "brain" and "nerves" to achieve a leading decision-making ability. The core of this competition has shifted from hardware manufacturing to software iteration and data-driven, and its foundation is the independent innovation system centered on cloud computing.

The essence of intelligent assisted driving is a closed loop of "real-time perception decision execution", which requires processing massive multimodal data (sensor data such as cameras, radar, LiDAR, etc.), training complex deep learning models (such as BEV perception algorithms, path planning models), and achieving low latency cloud collaborative control. The ability of cloud computing platforms directly determines the intelligence limit of intelligent assisted driving systems.

Therefore, Huawei has consolidated its deep accumulation in communication, cloud computing, chips and other fields, providing full stack intelligent solutions for every car, and ultimately outputting them in the form of Huawei Cloud CloudVeo intelligent driving cloud services.

Huawei started the research and development of in car modules in 2009, and laid out the vehicle networking laboratory in 2013. Over the years, it has continued to deepen its cultivation in the field of intelligent automobiles and increased investment. As of now, Huawei has ranked among the top providers of automotive cloud services in China: over 1 million intelligent assisted driving vehicles are flying on Huawei Cloud nationwide, and 50 million intelligent connected vehicles are serviced by Huawei Cloud. According to IDC data, Huawei Cloud has been ranked first in the Chinese automotive cloud market share for three consecutive years from 2022 to 2024.

This also indicates that the Chinese automotive industry has the opportunity to transform the disadvantages of local market size and scene complexity into advantages in data and iteration speed, and the core competitiveness of cloud vendors lies in this.

Taking computing power as an example, Nvidia's H20 chip customized for the Chinese market has been exposed to have security backdoors, with doubts about chip security and data sovereignty. No car company wants to be controlled by others. The performance of Huawei Cloud CloudMatrix384 per card can reach three times that of Nvidia H20, making it a more suitable computing platform for intelligent assisted driving model training. The actual test results show that the performance of Huawei Cloud CloudMatrix384 supernode has exceeded H100 on E2E and VLA models.

Compared to the computing power on the vehicle side, cloud computing power generally has high latency. Huawei Cloud's approach is simple and direct, but requires high resource investment. The Huawei Cloud Gui'an Automotive Zone has officially launched, echoing the Ulanqab Zone and the future Wuhu Automotive Zone. Huawei Cloud is about to complete the industry's first three zone launch, with each automotive zone carrying thousands of servers and supporting millisecond level services.

Cars are getting closer to the cloud. In Huawei Cloud's automotive zone, intelligence can emerge after training and provide a local experience at all times. The latency of the car cloud is significantly reduced by 60%, and the availability reaches 99.999%. The intelligent assisted driving experience is also smoother and more reliable.

The newly restructured "New Changan Automobile" has already established a deep cooperation with Huawei Cloud, which supports the efficient training of Changan Automobile's intelligent assisted driving model. Both parties have adapted various intelligent assisted driving models such as VLA and end-to-end. In addition, behind Changan Tianshu's intelligent assisted driving is the CloudMatrix384 super node, and Changan is also the first car company to apply this super node and conduct intelligent assisted driving research and development with domestic computing power.

Whether to build a good car or operate an intelligent system that can continuously learn and evolve, more car companies have seen the future of Chinese automobiles, with comprehensive competition based on cloud native platforms, data as fuel, and AI algorithms as engines, and the curtain has been lifted.

From looking up to Western technology to leading independent innovation, the Chinese automotive industry has gone through a tortuous and glorious path. The technological barriers of the past are being overcome one by one, and the dreams of the past are becoming a reality.