With the dawn of the AI large model era, AI-powered vehicles are poised to become the ultimate game-changer in the automotive landscape.

Just over two months into 2025, a subtle shift has emerged from the flurry of automotive industry press conferences. Instead of boasting about range and charging speeds, automotive executives now frequently spotlight "AI." As the transition from traditional fuel vehicles to electric cars gains momentum, "AI-defined cars" have arrived with a roar.

"The second half of the new energy competition revolves around intelligence, or more specifically, AI intelligence," said Gan Jiayue, CEO of Geely Automobile Group, at Geely's AI Smart Technology Conference on March 3rd.

From ChatGPT to DeepSeek, AI large models have emerged as a new force in productivity. Ouyang Minggao, academician of the Chinese Academy of Sciences and vice chairman of the China Electric Vehicle Hundred People's Council, predicts: "The 'AI +' revolution in the automotive field will permeate every aspect of the industry chain, from design and manufacturing to sales and maintenance, in a comprehensive and all-encompassing manner."

"Within the next two years, large model technology will undoubtedly find its way into the automotive sector. Autonomous driving solutions relying on extensive training data will gradually supplant pre-programmed modes based on predefined rules, ushering in a new era of true vehicle intelligence," said Zhou Hongyi, member of the National Committee of the Chinese People's Political Consultative Conference and founder of 360 Group. The application of large models will become a crucial determinant of competitiveness in the automotive market. "With the integration of increasing digital technologies, today's cars have transformed into 'supercomputing centers' on four wheels."

AI large models are continually accelerating the evolution of smart cockpits and the widespread implementation of high-level autonomous driving, stirring up excitement in the terminal market.

The True 'Smart Cockpit' Has Arrived

The concept of a 'smart cockpit' is not new, but the leap from executing simple commands like 'open the sunroof' to accurately anticipating user needs is significant.

"Many consumers have expressed dissatisfaction with the current level of intelligence in smart cockpits. You have to give it a clear, single instruction without contextual understanding for it to complete even a simple task," an insider from the smart electric vehicle evaluation industry told China News Weekly. "With AI large models reshaping the DNA of smart cars, smart cockpits will soon understand a series of actions in the context of drivers and passengers, enabling 'human-like reasoning' and making interactions smoother and more seamless."

This year, the popularity of DeepSeek has ignited enthusiasm among automotive companies, with over 20 carmakers announcing partnerships with the tech giant.

For instance, BYD's intelligent vehicle architecture 'Xuanji' is integrated with the DeepSeek-R1 large model. A BYD spokesperson told China News Weekly, "On the cockpit side, we use our self-developed open AI architecture to achieve deep integration between our proprietary large model and commercial large models. Leveraging DeepSeek's superior reasoning capabilities, we can better comprehend users' vague intentions and implicit needs, providing them with more precise and personalized services."

Li Auto announced that its AI assistant, Li Xiang Tong Xue App, is fully integrated with the DeepSeek-R1&V3 671B full-blooded version. Dongfeng Nissan's N7 Cloud Enjoy Cockpit also integrates DeepSeek, accurately understanding user needs in areas such as voice control, navigation planning, and entertainment system recommendations. Geely Auto's proprietary large model has completed deep integration with DeepSeek. For instance, a simple statement like "I'm a bit cold" will prompt the smart car's AI to automatically adjust the interior temperature, turn on seat heating, and other functions.

"DeepSeek's popularity domestically and internationally, coupled with the introduction by over 20 car companies, will undoubtedly accelerate intelligent development and intensify market competition," Ouyang Minggao told China News Weekly. However, Cui Dongshu, secretary-general of the Passenger Car Market Information Joint Meeting of the China Automobile Dealers Association, cautioned that no large model is omnipotent. While car companies' access to DeepSeek promotes intelligence levels, they must possess strong data and analysis capabilities. No matter how powerful a large model is, if the raw data is flawed, the conclusions drawn will never be accurate.

The Year of the Great Explosion of Intelligent Driving

In 2025, Tesla's FSD finally made its debut in China, while BYD sparked a wave of competition for 'driving equality.' Geely's Xingrui Intelligent Computing Center 2.0 boasts over 10,000 card-level computing power resources. Traditional automakers, which previously lagged in intelligent driving, have now embraced domestic intelligent driving suppliers... The field of intelligent driving is witnessing a year of "Cambrian Explosion."

On February 25th, Tesla announced the optimization and upgrade of its existing Autopilot system, bringing driving assistance functions similar to FSD (Full Self-Driving) on the American market to China.

As the pioneer in applying end-to-end large models in intelligent driving, Tesla currently has a 'shadow mode' for over 7 million vehicles worldwide. Combined with its supercomputing power, algorithms, and data, Tesla has established an infrastructure barrier that significantly outpaces Chinese car companies. In terms of computing power alone, by the end of 2024, Tesla Dojo's total computing power had reached 100 EFLOPS, equivalent to the combined computing power of approximately 320,500 NVIDIA A100 GPUs, surpassing the combined computing power of all domestic car companies.

"The world's most formidable player has arrived in China, pushing technology to the next level, which is a positive development," Cao Xudong, CEO of China's high-level intelligent driving supplier Momenta, commented on Tesla's FSD entering the Chinese market. Industry analysts generally agree that Tesla's entry will accelerate technological iteration among local car companies.

In fact, in the first half of 2024, Baidu Apollo announced the world's first end-to-end large model for fully autonomous driving, Apollo ADFM, progressively realizing fully autonomous driving. The sixth-generation driverless vehicle launched by Robotaxi last year has been integrated with Baidu Apollo ADFM, also the world's first large model supporting L4 autonomous driving. Under China's complex road conditions, the actual accident rate of Robotaxi is only 1/14 of that of human drivers. A Baidu spokesperson provided China News Weekly with data: As of January 2025, Baidu's autonomous driving services have provided over 9 million rides nationwide.

Compared to Tesla, how can local car companies bridge the 'data gap'? "Mining gold data from massive data," said a BYD spokesperson. "First, leveraging the VLM multimodal large model, we automatically obtain perception cases and experienced drivers' behaviors while mining massive data, thereby rapidly accumulating high-value training data. Second, based on extensive intelligent driving data, we train an industry-leading intelligent driving world large model through unsupervised reconstruction."

Chen Qi, chief intelligent driving scientist at Geely Holding Group, bluntly stated, "Geely possesses 23.5 EFLOPS of computing power, 7.5 million vehicles, 10 billion kilometers of driving data, and deeply integrated large models. Geely's intelligent driving experience and data accumulation are also rapidly growing, and we will continue to maintain this momentum."

"For safer and more inclusive intelligent driving, AI capabilities must be incorporated," said Gan Jiayue. The advent of large models has transformed the logic of algorithms. "Previous intelligent driving was 'two-stage intelligent driving,' involving recognition and control processes. Now, we empower intelligent driving with AI technology, replacing traditional algorithms and manually written rules with advanced AI algorithm models, upgrading to 'one-stage intelligent driving.' This approach can generate and continuously train more scenarios through fused large models, achieving greater safety."

In the market for independent intelligent driving solutions, domestic companies such as Momenta, Huawei, and Yuanrong Qixing have become key players. Data shows that in 2024, Momenta's market share reached 60.1%, while Huawei accounted for 29.8%.

To cope with the competitive pressure in the era of intelligence, many foreign-invested and joint-venture car brands have begun embracing mature solutions from China's technological ecosystem. For example, on March 1st, Dongfeng Nissan officially unveiled its new Tianyan architecture and introduced the first pure electric model, Nissan N7, developed based on this architecture. In terms of intelligent driving, Dongfeng Nissan's intelligent driving system, developed in collaboration with Momenta, adopts an end-to-end solution, supporting high-speed and urban expressway navigation assistance, enabling intelligent driving in complex scenarios and enhancing driving safety and convenience.

Prices Accelerate 'Downward'

Technological advancements are so rapid that they seem 'uncatchable,' and once lofty technologies are now entering thousands of households at continuously decreasing prices.

In February 2025, BYD unveiled a 'super move' for intelligent driving - the Tianshenzhiyan High-Level Intelligent Driving System, ushering in the era of universal intelligent driving. BYD Chairman Wang Chuanfu stated that Tianshenzhiyan High-Level Intelligent Driving can achieve full-speed highway driving without human intervention. A total of 21 BYD models will be equipped with Tianshenzhiyan High-Level Intelligent Driving, with standard configurations across models priced at 200,000 yuan, 150,000 yuan, and 100,000 yuan, and most models priced below 100,000 yuan will also be equipped.

"Future car models will undoubtedly all be intelligent driving models, and driving equality is a very normal thing," said Feng Xingya, NPC deputy and secretary of the Party Committee, chairman, and general manager of GAC Group, during this year's Two Sessions. "However, in the early stages of autonomous driving, due to the relatively high cost of technology, consumers may not accept this cost if applied to low-priced models. But I believe that the cost of intelligent driving technology will decrease very rapidly, faster than imagined, and soon driving equality will become very prevalent."

Feng Xingya said that driving equality has always been GAC's goal, and the next step will be to apply intelligent driving functions to more models. Currently, L2-level technology (such as automatic braking, following, lane keeping, etc.) is already prevalent in GAC's models, and more models will apply L2++ (such as highway intelligent driving) in the future.

Cao Xudong said that China's entire market is advancing faster than the global market in terms of high-level autonomous driving. "Our feeling is that we are at least three years ahead of the international market, and it might be even faster," Cao Xudong explained. On the one hand, competition among China's outstanding car companies has driven rapid technological development in the entire market; on the other hand, the age of Chinese consumers purchasing cars is younger than that in Europe and the United States. "Chinese consumers are extremely fond of and accepting of new technologies, which also benefits Chinese enterprises, including technology companies like us."

Continental's 2024 Future Mobility Trends Survey Report provides a consistent judgment: Up to 80% of Chinese respondents hope their cars are equipped with the latest technology, a proportion significantly higher than any other country participating in the survey (about 43% in Germany and France, and 51% in Japan and the United States).

Currently, City NOA (Navigation Assisted Driving) is gradually becoming mainstream, shifting from a niche, high-end product. More and more consumers are making it a must-have option when purchasing cars, and car companies are significantly increasing the configuration of high-level intelligent driving to meet market demand. Cao Xudong predicts that by the end of 2025 to 2026, City NOA may even become a standard configuration for models priced above 150,000 yuan.

"The driving factor behind the standard configuration of high-level intelligent driving is Moore's Law of intelligent driving, which includes Moore's Law of software and Moore's Law of hardware," Cao Xudong told China News Weekly. "Moore's Law of hardware states that the BOM (Bill of Materials) of high-level intelligent driving hardware will halve every two years." Cao Xudong introduced that a couple of years ago, to achieve City NOA, the cost of the entire BOM for dual Orin-X lidars might have been over 20,000 yuan, while now, the entire BOM for a single Orin-X lidar is around 10,000 yuan. By the end of 2025 and 2026, the BOM for achieving City NOA can be within 5,000 yuan.

"Moore's Law of software states that the product experience will improve tenfold every two years, which is the industry standard," said Cao Xudong. The pace of progress for leading companies in the industry may be even faster, achieving a tenfold improvement in just one year. "With the product experience improving exponentially and the hardware BOM cost halving every two years, we predict that the penetration rate will increase to 70% - 80% within a relatively short five years."

Getting Closer to High-Level Intelligent Driving

Intelligent driving seems to be one step closer to ordinary consumers, but there is still a long journey ahead before true 'autonomous driving' becomes a reality.

The advent of Tesla's Full Self-Driving (FSD) capability has ignited a surge of comparative tests for domestic intelligent driving systems. A review blogger lamented, "One day of testing resulted in a full deduction of 12 points." Both Tesla's FSD and several leading domestic intelligent driving models still have ample room for improvement in their current performance. Ouyang Minggao emphasized that the high-level intelligent driving functions currently available on the market still fall under the L2+ category of assisted driving. Despite the incorporation of AI technology, there remains a significant gap from achieving fully autonomous driving.

For autonomous driving, the challenge lies in achieving efficient and safe vehicle control in complex and dynamic traffic environments. Existing autonomous driving systems generally lack multi-agent collaboration capabilities, efficient decision-making and interpretation abilities, and struggle to effectively comprehend the behaviors and intentions of surrounding traffic participants in complex scenarios. Li Bin, CEO of NIO, bluntly stated at CES Asia 2025, "In the realm of autonomous driving, there's a human life separating 99% reliability from 99.9%."

Currently, Tesla relies solely on public data to train simulation scenarios and subsequently uses these to train models. This means that its local data in China is still far from sufficient, rendering Tesla's FSD somewhat 'unsuited to local conditions' in China.

"The generalization capability of the large model technical path is exceptionally strong," said Cao Xudong. "We previously trained a highly effective intelligent driving large model in China. When we aimed to expand into overseas markets, such as Japan, despite the difference in steering wheel positions, we achieved good results. This underscores the universal law of large models. If you can score 100 points in China, you can expect to score around 80 points globally."

"L4-level intelligent driving is undoubtedly the future trend and will inevitably return to artificial intelligence," stated Gan Jiayue. "If we continue to adhere to the recognition + decision-making logic, we will always believe that no matter how many configurations are stacked or how powerful the algorithm is, its scenarios will conform to a normal distribution. The scenarios on both ends of this distribution are endless, and these endless scenarios pose a substantial challenge to intelligent driving. Therefore, the integration of AI technology is the sole path for the development of intelligent driving."

Lang Xianpeng, Vice President of Li Auto's autonomous driving research and development, believes that with the current end-to-end + VLM (Vision Language Model) system, if technical capabilities continue to evolve, Li Auto is poised to achieve L3-level autonomous driving by 2025. L3-level intelligent driving marks the initial stage of supervised intelligent driving and serves as a precursor to L4 and L5-level autonomous driving. In the supervised intelligent driving stage, the product delivered to users by Li Auto will be a full-scenario, integrated end-to-end product.

"Xiaomi aims to become the industry leader in intelligent driving and plans to invest over 6 billion yuan in this field within three years," said Lei Jun, Chairman of Xiaomi Group. Through the optimization of large AI models and intelligent algorithms, Xiaomi can further enhance the autonomous driving capabilities of its vehicles, improve its full-scenario intelligent ecosystem, and identify new value growth points. According to the plan, Xiaomi Automobile will focus on launching full-scenario intelligent driving functions in 2025, encompassing highways, urban roads, and parking scenarios, striving to enter the L3 stage.

"To achieve L3 and L4 autonomous driving in the future, a shift in the technological route of autonomous driving is necessary, transitioning to an end-to-end model based on large models," said Ouyang Minggao. He added that autonomous driving employs the multimodal DeepSeek-V3. "Current large language models primarily process text, whereas multimodal models can understand and generate various types of data, including text, images, audio, and video. Most sensors capture video and images, and this multimodal DeepSeek can serve as the foundational large model for end-to-end algorithms, empowering autonomous driving technology."

Knockout rounds enter a new phase

"The development of intelligent vehicles will intensify market competition," believes Ouyang Minggao. "The current stage of industrial development is characterized by the imminent end of the influx of new carmakers. Competition in the stage of intelligent transformation will become more intense, and preparations should be made for the transitional stage of mergers and acquisitions, as well as the preparatory stage of industrial model transformation."

The competitive landscape in the automotive industry has undergone a transformation. Currently, new players in the automotive industry, such as Tesla, XPeng, NIO, and others, fiercely compete in technological innovation and functional experience, with intelligent driving as their core selling point. For instance, Tesla's FSD leads the way in data-driven end-to-end intelligent driving, while XPeng Automobile was the first to achieve mass production of end-to-end large models for vehicles. Meanwhile, BYD has adopted a fully self-developed vertical integration model, forming a closed ecological loop. Traditional automakers such as Volkswagen and Mercedes-Benz are accelerating their intelligent transformation and investing significant resources in the development of intelligent driving technology. Technology companies and automakers are developing in a competitive yet cooperative manner. Technology companies like Baidu and Huawei leverage their AI technology and algorithm advantages to provide intelligent driving solutions for automakers.

However, the development of intelligent driving driven by large AI models requires substantial capital investment for the purchase of computing power equipment, data collection and annotation, talent recruitment and training, among others. Inevitably, enterprises with insufficient financial strength will be forced out of the market amidst fierce competition. For example, Zongmu Technology, once valued at 9 billion yuan as an "unicorn" in intelligent driving, experienced a collapse within just a few months from late 2024 to early 2025, including power outages, building closures, unpaid salaries for all employees, disruptions in social security payments, and collective resignations of senior executives, ultimately succumbing to capital chain issues.

Moreover, with rapid technological iteration, enterprises that fail to swiftly apply large model technology to intelligent driving for functional upgrades and performance improvements will lag behind. For instance, enterprises that solely focus on low-level intelligent driving functions and fail to break through high-level technologies will see their survival space compressed.

In the view of He Xiaopeng, CEO of XPeng Automobile, with the advent of the era of large AI models, AI cars will be the entry ticket for future competition finals. "When I entered the industry, there were 300 new carmaking enterprises and 100 original carmaking enterprises (including commercial vehicles). Now, there are only 40 enterprises with significant sales volumes in China, and this has only taken seven years," estimated He Xiaopeng. If another 10 years pass, the number of mainstream Chinese automotive brands, encompassing state-owned, foreign-invested, joint ventures, and private enterprises, will be less than seven. "Selling at least 1 million strong AI cars a year is the entry ticket to the finals. Selling cars alone is one thing, but how to integrate cars with AI is the most crucial aspect in the next ten years."

As we stand at the time node of 2025, the transformation of the automotive industry has entered uncharted waters. With Tesla FSD users having accumulated over 1 billion miles of driving and Huawei's Kunlun Intelligent Driving ADS 3.0 system achieving full-scenario connectivity from parking space to parking space, many industry insiders believe that the future of cars transcends being merely a cold machine for transportation; they will transform into a mobile terminal capable of thinking and understanding emotions. This industrial revolution fueled by AI has the potential to become the starting point of the third travel revolution.