What can new forces in car manufacturing do in the second half to find differentiation?

Content/Awen

Editor/Yibai

Proofreader/Yonge

Car companies are accelerating towards intelligence.

Previously, there were two news stories that swept the screens. One was that Lei Jun personally recruited Luo Fuli, one of the key developers of the DeepSeek open-source large model who is in her late 20s, with a salary of tens of millions of yuan to lead Xiaomi's AI large model team.

The other was an AI Talk event held by Li Auto, where CEO Li Xiang revealed that he was transforming from a car company CEO to an AI company CEO, and that half of the 10 billion yuan in R&D expenditure would be invested in AI.

After carefully studying car companies' product introductions in the past two years, it is found that they no longer focus on promoting driving range, but instead emphasize smart cockpits and intelligent driving. For example, within the vehicle cabin, voice and gesture interactions are possible, and AR technology can be used to enjoy movies on a large screen. The vehicle also possesses advanced intelligent driving functions and integrated intelligent driving capabilities for both driving and parking.

Take Xiaomi SU7 Ultra, which was temporarily launched by Lei Jun, who became the richest man yesterday, as an example. It comes equipped with Xiaomi HAD end-to-end full-scenario intelligent driving right out of the factory. Meanwhile, Xiaomi's Pentai Smart Cockpit, combined with the full ecosystem of human-vehicle-home technology, provides an integrated experience encompassing smart navigation, entertainment, vehicle control, and more functions. It also introduced features such as "Roaming and Position-Finding Parking" and "Parking Out of Extremely Narrow Spaces".

For car companies, it is not easy to occupy a high ground in the field of intelligence. If they truly want to achieve their goals, strong computing power and continuously iterated algorithm support are the keys.

Currently, car companies are divided into two factions in their approach to large models. One faction chooses to cooperate with existing large model providers, such as cooperating with large model companies like Baidu, iFlytek, and SenseTime. The other faction chooses to develop large models independently, such as Li Auto releasing its self-developed large model MindGPT.

2024 is hailed as the "first year of intelligent driving." This year also marks the tenth year of the electric vehicle wave. Smart cars are accelerating into the AGI era. Whether it's self-developed large models or collaborations with these tech giants, "end-to-end" has become the current benchmark for intelligent driving competition.

What can new forces in car manufacturing do in the second half to find differentiation? The final outcome is uncertain. Will the new forces that continue to rush ahead meet again at the next decade's milestone?

Part.1

New forces are starting to "compete" in large models

As one of the few "profitable" enterprises among new energy vehicle companies, Li Xiang believes that "the future development direction of Li Auto is not to make cars intelligent, but to make artificial intelligence automotive." Li Auto's goal is to become a leading AI company.

In this way, the car is the means, and intelligence is the goal. That is, the "car" is a carrier for Li Auto to achieve intelligence. As he said, "Large models are the operating system and programming language of AI, and Li Auto must master this core technology."

In Li Xiang's view, Li Auto's growth path is divided into two major stages. First, the short-term goal is to become the number one in the field of spatial intelligence in China and enter the top three in the field of large language models. The medium- and long-term goal is to achieve deep integration of linguistic intelligence and spatial intelligence to build a unified VLA model (Visual Language Action model).

Almost keeping pace with Li Auto is Lei Jun and Xiaomi.

Thanks to the aura of Lei Jun and Xiaomi, Xiaomi SU7 has undoubtedly been the king among the "new forces" in terms of marketing and traffic over the past year. In December last year, Xiaomi Group announced that it was actively recruiting top talent in the AI field, including Luo Fuli, a core developer of the DeepSeek open-source large model DeepSeek-V2.

Xiaomi's investment in AI large model research and development is also continuing to increase. On December 26 last year, it was revealed that Xiaomi was building a GPU 10,000-card cluster to further enhance its AI large model training and optimization capabilities.

It is reported that Xiaomi's large model team already had 6,500 GPU resources at its inception, and the newly built GPU 10,000-card cluster will significantly enhance its competitiveness in the AI field.

In fact, Xiaomi has already made attempts in the AI large model race. As early as July 2016, Xiaomi had already begun to lay out the AI field. In April 2023, Xiaomi established a large model team within its AI lab; in November last year, Xiaomi even established a dedicated AI platform department, led by Zhang Duo, who was personally appointed by Lei Jun as the "Xiaomi God."

Looking at XPeng Motors, it is reported that its AI XiaoP, which integrates large model capabilities, can respond to questions about encyclopedic knowledge and obtain the latest information on the internet through online searches. AI XiaoP can also create fairy tales, generate images, make appointments related to the user's speech, and even help users identify vehicle information, road sign information, etc., through camera input.

The AI Tianji system of XPeng's AI can mimic the driving style of the car owner to handle practical problems, enabling "one-click to work" and "one-click to go home." Its AI bodyguard can recognize more than 50 targets, including manhole covers and hanging road signs, to provide early warnings of potential risks.

The level of intelligence in cars is becoming increasingly attractive to consumers. Looking around, there are hundreds of electric vehicle companies in China, but very few of them are building base models. This also provides opportunities and markets for AI companies that provide basic large model automotive industry solutions, such as SenseTime, Megvii, and iFlytek.

For example, SenseTime currently has a trinity of general artificial intelligence (AGI) technology architecture and product system comprising "driving-cockpit-cloud." On November 27 last year, SenseTime announced multiple new AI automotive products, including the Jueying mass-production intelligent driving product system for high-level intelligent driving and end-to-end intelligent driving covering highways, urban areas, and parking, as well as a new upgrade to the industry benchmark-level "Kaiwu" world model. For smart cockpits, the industry's first large model innovative product "A New Member For U" (Your New Family Member) was launched, giving cars an "interesting soul."

It is worth noting that SenseTime is also the company that proposed the industry's first end-to-end intelligent driving solution integrating perception and decision-making.

Public data shows that the cockpit large model product matrix of the smart car platform "SenseTime Jueying" has been installed on models such as Xiaomi SU7, IM Motors, and LEVC L380. Among them, the Jueying large model fully supports the application of Xiaoai voice assistant in the vehicle voice scenario (Xiaomi SU7), and the "Miaohua" product is installed on the entire IM Motors model lineup.

Looking at Megvii, which had waited for years to go public but ultimately chose to withdraw its IPO, founder Yin Qi welcomed a new identity - diving into smart cars, serving as chairman of Lifan Technology.

Tracing the starting point of Megvii's involvement in the smart car business can actually be traced back to the beginning of 2021, when it established the Maishi Zhixing intelligent driving brand and quickly entered the technology route dominated by vision, becoming one of the earliest domestic manufacturers to conduct research and development along the BEV vision-based technical direction.

In 2023, three mass-production intelligent driving solutions with different configurations were officially released, all targeting L2+ autonomous driving, and mass production was achieved on multiple Geely models.

Previously, Chongqing Liangjiang New Area, Geely, and Megvii announced a strategic cooperation plan, where the three parties will jointly build an AI intelligent open platform and construct an "AI + car + robot" industrial innovation development model.

In the automotive field, iFlytek released the Spark large model in 2023, actively promoting the application and landing of large model technology in the automotive field; in December of that year, the first Chery Star Era ES equipped with the iFlytek Spark large model was officially launched and won the "Best Sedan" award in January last year.

In the process of large models being applied to cars, iFlytek also launched the automotive edge-side large model for the first time.

Compared to cloud-side large models, the edge-side 1.3B large model has an effect loss of ≤1%, with a first response time of 40ms and an edge-cloud fusion interaction time of as low as 1.3s, enhancing the human-computer interaction experience. In the first half of 2024, iFlytek successively reached cooperation with companies such as FAW Group, Ruilan Automobile, Alipay, and China Automotive Technology and Data.

Today, the new forces are standing on a broader "battlefield," facing new competitors, strategies, and market competition. They all want to find differentiated competitive advantages over their peers and "survive" and "thrive" in the fierce market elimination tournament. Therefore, such a scene has emerged - car companies are eager to get large models on board, and large model companies are also eager to recommend their large models to car companies.

Part.2

AI large models, the next "battlefield" for new forces

Today, intelligent driving and smart cockpits have become the main battlefields of AI competition.

First, intelligent driving, as the focus of AI technology applications in the automotive field, is the sector with the most rapid technological changes in recent years. In the field of intelligent driving, the application of AI large models can significantly enhance the perception accuracy of cars, gradually reduce dependence on high-precision maps, and thereby improve driving safety and convenience, which is what consumers are concerned about and looking forward to.

Looking back over the past few years, intelligent driving technology has undergone several important adjustments in technical routes.

From solutions that rely on high-precision maps to shifting to mapless modes, to companies represented by Tesla trying the pure vision (abandoning LiDAR) route. However, this has not truly achieved intelligence. These so-called intelligent driving systems still require a large number of programmers to invest a lot of effort in writing code and fixing bugs.

However, the birth of end-to-end technology has completely overturned this traditional model.

Specifically, it promotes the transformation of the intelligent driving experience from "usable" to "user-friendly," enhancing users' willingness to pay. Taking Tesla as an example, the company released the FSD V12.3 version in March 2024, adopting an end-to-end neural network algorithm that reduces information transmission loss between modules in traditional modular systems; through training on over 10 million Clips, FSD has achieved better simulation of human driving behavior and reduced takeover willingness.

Another in-depth application of AI in the automotive context is the smart cockpit. Currently, car companies represented by Li Auto, Jiyue, NIO, and XPeng have embedded large models into smart cockpits, enabling functions such as speech recognition, sentiment analysis, and personalized recommendations. The smart cockpit can not only help users achieve convenience in vehicle driving, management, leisure, and entertainment but also provide assistance for users' work.

Data from Gaoshi Auto Research Institute shows that the current overall penetration rate of smart cockpits in China is close to 60%. Next, as the entire industrial chain further matures and driving costs continue to decline, it is expected that the adoption rate of smart cockpits will continue to increase, with the overall penetration rate expected to exceed 70% in 2024.

Although the introduction of large model technology has brought powerful processing capabilities to intelligent driving, there is still a significant gap between technology adoption and actual user scenario applications, such as challenges in stability and accuracy. For example, the in-car system sometimes has issues with precise execution of voice commands and slow system response, which may affect the user experience.

In future smart cockpits, "doing subtraction" may become the main optimization path for car companies.

AI will not only be a technical tool but will become the car owner's "intelligent assistant," enhancing driving safety, convenience, and entertainment while also providing more personalized services through precise understanding of the car owner's needs. For example, the voice output of the smart cockpit should be more concise and accurate to ensure that the driver can quickly complete tasks and focus their attention on the road.

From car companies to intelligent system suppliers and technology companies, they are all competing in AI, which in turn forces car companies to continuously expand the boundaries of application and make smart cars "smarter."

Part.3

Car companies vs. large model suppliers, allies or competitors?

Voice assistants have evolved from primarily playing the role of a "conversation tool" to later possessing logical thinking and general task processing capabilities, becoming "little assistants." However, this still cannot truly satisfy the diverse and personalized needs of humans.

Large models can achieve better generalization performance at lower costs, adapting to different customers and environments. They can also help car companies achieve efficiency in data collection, processing, model training, etc., quickly respond to market demands, and shorten product development cycles.

After using large models, car companies will possess differentiated competitiveness, which is one of the reasons why new forces are eager to enter the field of AI large models.

Compared with advanced foreign large models, domestic large models still have a gap in data annotation, computing power construction, and model architecture, despite rapid development. "No company can make such a large language model by rushing through a few months of work. It is the result of Baidu's massive R&D investment accumulated over more than a decade," Li Yanhong said previously.

The application prospects of AI large models in the automotive industry are broad, but computing power construction requires huge capital investment. Car companies need to comprehensively consider various factors when choosing between self-development or cooperation, including technical capabilities, resource investment, capital investment, market demand, competitive pressure, and more.

For new forces, the large language models of Baidu, Huawei, and iFlytek have significantly enhanced the personalized services and user interaction experience of smart cockpits through technologies such as natural language processing, speech recognition, visual recognition, and multimodal fusion, enabling more natural and efficient intelligent interactions.

However, from the perspective of the market ecosystem, the exploration of AI large model application and commercialization continues. In the future, car companies and large model suppliers will not be absolute competitors, and the cooperation mode between them will deepen, perhaps jointly exploring and innovating application scenarios.

Previously, Yu Chengdong posted on Weibo, saying, "The era of universal intelligent driving has arrived!" For new forces in car manufacturing that are at the halfway point of the game, intelligent driving is where they need to make all-in investments in technology research and development and marketing offensives. They are well aware that if they want to "snatch" market share from traditional car companies, they must find their own differentiations, and AI large models are one of them.