Over the past decade, the automotive industry has undergone a rapid electrification transformation, and now, the momentum for intelligent transformation is gaining urgency. With the swift advancements in artificial intelligence, 5G communication, and high-performance computing technology, autonomous driving technology is transitioning from science fiction to reality, becoming a pivotal force driving the global automotive industry's transformation. Governments worldwide have issued policies to foster research, development, and testing of autonomous driving technology, giving the green light to industrial development. Massive capital inflows have nurtured numerous startups and innovative projects focused on autonomous driving technology. Consumers' expectations for travel experiences have also risen, eager to liberate themselves from the rigors of driving and enjoy a relaxed and pleasant commute.

The past year has undoubtedly been a "milestone" for the intelligent driving industry. "Luobo Kuaipao" in Wuhan has captured public attention, Tesla's self-driving taxi Robotaxi has unveiled its capabilities, and Chinese and American companies are locked in fierce competition in the intelligent driving race. Companies like RoboSense, Horizon Robotics, Black Sesame Technologies, WeRide, Pony.ai, and others have rushed to go public, attracting significant capital inflows into the intelligent driving industry. The industry's "explosive development" has become a consensus.

In 2024, China's Ministry of Industry and Information Technology and other departments announced that Beijing, Shanghai, Guangzhou, and six other cities have launched pilot programs for intelligent connected vehicles to enter the market. Nine enterprises, including BYD and NIO, have been selected for the L3 autonomous driving technology pilot program. These enterprises will conduct pilots in passenger vehicles, buses, and trucks, aiming to refine technical specifications and regulations.

In 2025, multiple laws and regulations on autonomous driving will also come into effect, with L3 autonomous driving taking center stage. The intelligent driving industry will continue to experience rapid growth.

01

What is L3 Autonomous Driving?

Autonomous driving technology has emerged as a hot topic in the global automotive industry in recent years, with L3 autonomous driving serving as a significant milestone gradually transitioning from concept to reality. To understand the significance of L3 autonomous driving, it's essential to grasp the classification standards of autonomous driving and the technical logic behind them.

Since SAE International (Society of Automotive Engineers) released the driving automation classification standards in 2014, this standard has become the most widely referenced benchmark for describing autonomous driving levels in the industry. The latest version, revised in 2021, divides driving automation into six levels:

L0 (No Automation): The vehicle is entirely controlled by the human driver, with the system offering basic warning functions like lane departure warning.

L1 (Driver Assistance): The vehicle features single-function automation, such as Adaptive Cruise Control (ACC) or Lane Keeping Assist (LKA), but the driver must still monitor the vehicle throughout the journey.

L2 (Partial Automation): The vehicle can control multiple functions simultaneously, including acceleration, braking, and steering, but the driver must remain ready to take over at any time.

L3 (Conditional Automation): Under specific conditions, the vehicle can drive autonomously, and the driver can take over when the system requests. The key feature of L3 is "conditional," meaning the system can fully handle driving tasks in certain scenarios (like highways).

L4 (High Automation): The vehicle can drive autonomously under most conditions without human intervention, though human takeover may still be necessary in extreme situations.

L5 (Full Automation): The vehicle can drive autonomously under all conditions without human intervention.

L3 autonomous driving represents a turning point for "human-machine co-driving" as it allows drivers to hand over driving tasks to the vehicle completely under specific conditions for the first time. However, L3 also imposes higher requirements on system reliability and safety, particularly during the transition phase when the system requests human takeover.

The realization of L3 autonomous driving hinges on the collaborative effort of multiple technologies, including sensors, algorithms, and computing platforms. Among these, computing power stands as one of the core elements supporting L3 autonomous driving. The computing power demand for L3 autonomous driving typically exceeds 30 TOPS, while L4 and L5 may require hundreds or even thousands of TOPS. Therefore, computing power has emerged as a critical bottleneck for implementing L3 autonomous driving.

02

National Standard for L3 Autonomous Driving Implemented, Policy Takes the Lead

The implementation of L3 autonomous driving necessitates not only technological breakthroughs but also the backing of policies and regulations. In recent years, China has taken the global lead in policy formulation and standard construction in the field of autonomous driving, paving the way for the commercialization of L3 autonomous driving.

In 2024, China released the "General Technical Requirements for Autonomous Driving Systems of Intelligent Connected Vehicles" (GB/T 44721-2024), marking China's first national standard for autonomous driving systems. This standard outlines the overall requirements, dynamic driving task execution, dynamic driving task backup, human-machine interaction, and other aspects of L3 and higher autonomous driving systems (ADS).

The issuance of this national standard provides a technical foundation for the commercialization of L3 autonomous driving and offers a clear research and development direction for automakers and suppliers.

In addition to national standards, local governments are actively promoting the implementation of L3 autonomous driving. Many regions have issued relevant policies to support the testing and commercialization of L3 autonomous driving.

On December 31, 2024, the "Regulations on Autonomous Driving Vehicles in Beijing" were passed and will come into effect on April 1, 2025, providing institutional guidelines for L3 and higher autonomous driving vehicles. The regulations clarify the testing and operation rules for L3 autonomous driving vehicles in Beijing, encompassing licensing conditions for test vehicles, data recording requirements, and accident liability division. They also propose that Beijing will gradually open more roads for testing and operation of L3 autonomous driving.

Just a day before the Beijing "Regulations" were passed, the "Regulations on the Promotion of the Development of Intelligent Connected Vehicles in Wuhan" were also officially released and will take effect on March 1, 2025.

Furthermore, Zhejiang Province recently unveiled the "Action Plan for the Development of the Intelligent Connected Vehicle Industry in Zhejiang Province (2025-2027)," aiming to foster high-quality development in the intelligent connected vehicle industry. According to this plan, the goal is to establish 2 to 3 globally renowned automotive brands by 2027, with the penetration rate of intelligent connected fuel vehicles above L2 exceeding 70%, the penetration rate of intelligent connected new energy vehicles above L3 surpassing 40%, and the local supporting rate increasing by 10 percentage points.

The first domestic intelligent connected vehicle management regulation to be issued and implemented was the "Regulations on the Management of Intelligent Connected Vehicles in the Shenzhen Special Economic Zone," which came into effect on August 1, 2022. In the nearly two and a half years since then, no other city has issued relevant regulations. Constrained by policies and regulations, despite major automakers having mature autonomous driving technology, they have hesitated to invest in the commercialization of L3 autonomous driving.

Industry insiders believe that with the successive issuance of L3 autonomous driving regulations in various regions, it signifies strong policy support and promotion for the rapid implementation of L3 and higher autonomous driving technology. The enactment of these local policies provides a practical basis for the implementation of L3 autonomous driving and serves as a reference for policy formulation in other areas.

03

Higher Computing Power Demand for L3: Automakers Invest in Self-Developed Chips

In 2024, the penetration rate of L2 and higher autonomous driving in passenger cars in China was 55.7%, and Zhang Yongwei, vice chairman and secretary-general of the China Electric Vehicle Hundred People Council, predicts that this figure may approach 65% by 2025. The influence of intelligent experience in purchasing decisions has risen to 14%, becoming the third most significant decision-making factor after vehicle quality and performance, on par with vehicle design and surpassing brand and price.

For autonomous driving, hardware forms the foundation, encompassing key components like LiDAR sensors and controllers, particularly high-performance computing chips; software serves as the "soul," encompassing algorithms and systems across multiple levels of perception, decision-making, and control. Currently, autonomous driving is the focal point of competition among major automakers.

The high demand for computing power in L3 autonomous driving has spurred the upgrade of in-vehicle computing platforms, with computing power chips, as the core of the computing platform, becoming the center of competition among automakers. To meet the computing power requirements of L3 autonomous driving, an increasing number of automakers have begun investing in self-developed chips.

XPeng Motors showcased its self-developed "Turing AI Chip" at the "XPeng AI Technology Day" in November 2024. This chip boasts a 40-core processor, with AI computing power equivalent to three NVIDIA Orin X chips, a comprehensive computing power exceeding 750 Tops, and the capability to run large models with up to 30B parameters, supporting two independent ISPs, providing clear vision in night, rain, and backlight conditions. The Turing chip is claimed to be equivalent to three chips, with one chip delivering an L3+ high-level intelligent driving experience and two chips achieving an L4 autonomous driving experience.

In July 2024, Li Bin, founder of NIO, presented NIO's first self-developed intelligent driving chip, Shenji NX9031, at the "NIO IN 2024 NIO Innovation Technology Day" and announced its successful tape-out. The chip employs a 5nm automotive-grade process and contains over 50 billion transistors. Li Bin stated that one of their self-developed chips can match the performance of four industry-leading intelligent driving chips (Orin).

On October 28, 2024, Geely Automobile officially unveiled its self-developed "Xingchen No.1" autonomous driving chip. This chip utilizes a 7nm process and boasts impressive computing power, with a CPU computing power of 250KDMIPS and a single NPU computing power reaching up to 512TOPS. When multiple chips operate together, the maximum computing power can achieve 2048TOPS. Geely plans to commence mass production of the "Xingchen No.1" chip in 2025 and widely apply it to its high-end models, including the Lynk & Co and Galaxy series, in 2026.

Moreover, there are rumors that BYD is internally developing a dedicated chip for intelligent driving in future vehicles. It is reported that this chip boasts 80TOPS of computing power, potentially covering nearly all future BYD models. BYD intends to replace all Orin N and Horizon J6E chips with its self-developed 80 TOPS computing power chip in the future.

Li Auto is also progressing with its self-developed chip project, codenamed "Schumacher," featuring a 5nm design process and production by TSMC. Recently, the Li Auto team has been adjusting its division of labor, with the NPU head's role gradually expanding.

By developing their own chips, these automakers have not only enhanced computing power but also strengthened their supply chain control and reduced dependence on external suppliers.

The significance of automakers investing in self-developed chips lies not only in meeting computing power demands but also in boosting technological autonomy and market competitiveness. Self-developed chips enable automakers to tailor chip functions according to their needs, thereby better supporting the development of L3 autonomous driving systems. Simultaneously, self-developed chips minimize dependence on external suppliers, improving supply chain security. They also provide automakers with differentiated competitive advantages.

The implementation of L3 autonomous driving marks a crucial step towards the automotive industry's intelligence. With the backing of policies and regulations, Chinese automakers are accelerating the commercialization of L3 autonomous driving through self-developed chips and technological innovation. In the future, as technology further matures and policies continually improve, L3 autonomous driving is expected to be widely adopted in more scenarios, offering consumers a safer and more convenient travel experience.

However, the challenges confronting enterprises within China's auto industry chain are universally shared, and there remain substantial weaknesses in core areas such as chips. By 2024, the autonomous driving chip market will still be predominantly controlled by international giants. Notably, NVIDIA holds a commanding 82.5% of the global market share with its high-performance autonomous driving SoC chips. According to Gasgoo data, NVIDIA's Drive Orin-X chip boasts an installed base of 1.554 million units, accounting for 32.6% of the market. Close behind is Tesla, which, despite not yet officially introducing its FSD system in the Chinese market, has an installed base of 1.0131 million units and a 26.8% market share, thanks to its self-developed FSD chip. Together, these two industry leaders occupy nearly 60% of the market share, demonstrating a pronounced concentration effect.