The battle of automotive intelligence is no longer controversial: comprehensive

tech | 2024-05-24

As 2024 has already reached its midpoint, looking back on the first half of the year in the automotive market, everyone would probably use the term "intense competition." However, beneath this crude and simplistic summary, most people's attention has been drawn to price wars and the verbal disputes among car manufacturers about "unfair competition," while overlooking some of the developments and changes in the automotive industry this year.

If we were to discuss the changes in China's automotive market in 2024, the most notable point is that no car manufacturer is "resisting" the trend of automotive intelligence. This applies to joint ventures, domestic brands, and foreign investments alike. Moreover, the earlier "neglect" of intelligent experiences in fuel vehicles by joint ventures has been completely "rectified" in 2024. Any newly released fuel vehicles now focus on localized intelligent experiences and the core concept of "same intelligence for both fuel and electric."

Advertisement

On the other hand, the new products of 2024 seem to have encountered a significant bottleneck—when all car manufacturers are "pushing for intelligence," it means there is increasingly less differentiation.

From smart cockpits to autonomous driving, it appears that leading car manufacturers from before now struggle to articulate their competitive advantages, while the majority of other manufacturers are following the path set by the pioneers. If you have it, I have it too, and as for the subtle differences in experience, the average consumer actually finds it difficult to perceive them in a brief trial.

It can be said that all the "intense competition" in the first half of 2024 essentially stems from the collective failure of differentiating intelligent experiences across various products. From the leading new forces to the chasing contenders, all are anxious about lacking differentiation.01

The Chill of Urban NOA in Cities

If we were to discuss the most significant difference in intelligence between the 2024 Beijing Auto Show and the 2023 Shanghai Auto Show, it would be the "ebb" of urban NOA (Navigation on Autopilot).

Last year, Li Xiang, the CEO of Ideal Automobile, once proposed that there would be two indicators for luxury new energy vehicles in 2024: one is 800V, and the other is urban NOA. However, when it really came to 2024, the competition for urban NOA actually remained among a few players—Huawei, Baidu, and Xiaopeng (counting as half)—while the rest of the automakers stopped talking about "city launch plans" and downplayed urban NOA as a long-term configuration.

It is possible that on the surface, everyone is still actively showcasing their urban NOA, through live broadcasts, internal testing applications, and various other methods. However, the enthusiasm of consumers for this feature is clearly not high.Over the past six months, even the most adventurous consumers have lost patience with the elusive urban NOA (Navigation on Autopilot). A prime example is Hongmeng Zhixing, which has been offering substantial subsidies to consumers to purchase the advanced driver assistance features of ADS 2.0, focusing on no-map NOA.

Similarly, AITO, which also emphasizes Huawei's intelligent driving, has not seen a sales increase following the launch of the AITO 12, despite the NOA feature. The case of XPeng is even more telling; with XNGP as a high-end configuration, XPeng has always been "tight-lipped" about its sales proportion.

In other words, the general public, who have just crossed the chasm of electric vehicle consumption, are hardly willing to spend extra money on urban NOA, making the marketing efforts feel like hitting a cotton pad.

The ebbing of urban NOA features in marketing, apart from the consumer's indifference, may also be due to the new forces realizing they can no longer compete. After all, to develop urban NOA, creating large models and algorithms is not just a matter of hiring more people; the time and computational power costs are very real needs, not to mention that purchasing computational power has now become a "luxury."

Furthermore, with this year's sales not being particularly strong, the intelligent driving teams of various car manufacturers are significantly downsizing, and a slight misstep could lead to mass unemployment.The young and passionate executives of automotive companies have found that, without money or computing power, this time the "Chinese speed" seems to be less effective.

However, new forces always manage to "ride the wave" of trends. Last year's buzzword was "driverless driving," which has now evolved into "end-to-end" for this year. Although consumers and most media cannot fully comprehend what "end-to-end" means, it sounds sophisticated and high-end—so new forces and those traditional car companies attempting to craft their "new image" always have "end-to-end" on the tip of their tongues.

Some critics mock: "Everyone is talking about 'end-to-end,' but each company has its own interpretation. These unfamiliar and seemingly nonsensical 'tech jargon' is already difficult to excite consumers again.

In the meantime, while new forces struggle to advance in urban NOA (Navigation On Autopilot), traditional car manufacturers have launched a new round of attacks on NOA. For instance, the newly formed smart driving team at Zeekr is attempting to enable high-speed NOA capabilities for the older models of 001, while Great Wall Motors, considered one of the most traditional car companies, is leading the charge under the guidance of "Wei Jianjun's Weekend," showcasing their pursuit of urban NOA capabilities through frequent live broadcasts, time and again.Even so, Xiaomi's car just started this year, and Lei Jun said that Xiaomi's 3000-person smart driving team can already launch a test version by the end of the year, which is astonishing progress.

The slight cooling of new forces in intelligent driving and the acceleration of traditional car companies have started to narrow the gap between the two in smart driving.

The "newness" of new forces is increasingly insufficient.

02

The castle in the air of AI large models

Like urban NOA, the progress of smart cockpits has also reached a bottleneck period.In the current era where all car manufacturers start with the 8155 chip and the high-end mainstream 8295 chip, the competition among all car companies is only about the UI design of large screens, and this UI interface is essentially copied from two sources—either Tesla or Huawei's HarmonyOS.

Moreover, the suppliers of in-cabin voice systems are also just a few well-known companies, along with more microphone arrays and more internal DMS (Driver Monitoring Systems) and cameras; the rear screens are also just ceiling-mounted screens and magnetically attached Pads, or Huawei's in-car projection systems. It can be said that, at present, the intelligent cockpits of new energy products in China, when only looking at the static definition, are almost indistinguishable from each other.

So, where is the direction for the advancement of intelligent cockpits? It seems that the new forces have chosen AI large models as their answer.

Now, companies like NIO, Xpeng, and Li Auto have been rushing to put AI large models into their vehicles, and other tech companies such as Xiaomi, Huawei, and Baidu have also put large models into their corresponding products. However, in terms of experience, these large models are mainly focused on cockpit interaction, even though they are all called multimodal cognition/perception large models, they essentially convert voice input into text and then obtain answers through the large models.In fact, as Lei Jun said, merely creating encyclopedias or image-to-image models does not constitute a large model for a smart cockpit. A truly intelligent cockpit model should integrate cameras, positioning, and other genuine multimodal capabilities to be meaningful. Therefore, what the actual application scenarios for large models in vehicles are has not yet been fully agreed upon within the automotive companies.

At the same time, according to Baidu, large models on the edge are the future. For instance, deploying a true large-scale voice model on the edge to replace the existing multiple small models can better achieve offline interactions, and at a faster speed.

However, the problem that arises is that for most in-car systems using the 8155 chip, the NPU computing power is quite weak, only 4 TOPS, while the 8295 chip's NPU computing power can reach around 30 TOPS. Moreover, by 2025, NVIDIA's Thor chip will have a computing power of 2000 TOPS that can be uniformly allocated between the cockpit and intelligent driving.

This means that from this year to the next, there will be an explosive growth in the computing power of the integrated cockpit and driving system. What can be done with so much computing power? It's not something that can be planned with just a few or dozens of TOPS of chips.For automotive manufacturers, since the next generation of vehicle models can utilize computing power in the thousands of TOPS, maintaining the functionality of existing low-computing power chips will be a high-cost endeavor. The development of new features for current chips such as the 8155 and 8295 seems to hold little significance.

Additionally, the number of suppliers providing AI large models is also very limited, with iFlytek and Baidu being almost the only options. The so-called self-research by new forces is essentially modifications and self-training on open-source models.

Compared to the large models developed by overseas giants like Open AI, Google, and Meta, there is a noticeable gap with the large models in China. This results in the functionality of domestic large models being limited to vague voice queries and some storytelling, search queries, etc., with truly multimodal capabilities being quite limited upon implementation, not to mention revolutionary innovative experiences.

Therefore, although everyone acknowledges that AI is the direction for the entire smart cockpit, currently, everyone is merely copying the cat to draw a tiger with new AI features, and the scenarios where they can be truly used are still very limited. This is also why, when looking at the new cars at this year's Beijing Auto Show, the interactive functions on the new cars are actually quite limited. The core reasons are, first, there are no more features on the mobile side to "copy," and second, the existing chip computing power is insufficient, leaving no room for the development of new AI features.

03

Smart Popularization for Low-Cost Vehicle ModelsIn fact, the real significant progress made by Chinese automotive companies in intelligent technology in the first half of the year lies in the large-scale deployment of smart cockpits and assisted driving to entry-level models. For instance, we can observe that models like the Galaxy E5, priced at around 120,000 yuan, have started to adopt the Flyme Auto system, while a large number of BYD models have begun to popularize smart car machines and assisted driving. Additionally, models in the 100,000 yuan category, such as Baojun, have started to integrate DJI's smart driving system.

Behind the large-scale adoption of intelligent systems in low-priced entry-level models is the fact that China's automotive industry chain has successfully developed domestic chips and systems. With low computing power chips, they have unlocked the integration of cabin, driving, and parking functions, fully achieving a demand-appropriate nationalization on entry-level products.

At the Beijing Auto Show, it was noticeable that the installation volume of domestic chip manufacturers began to rise sharply. Taking Siwei Tuxin as an example, its subsidiary, JieFa Technology, has developed a dual-domain fusion hardware that integrates Horizon's J3 chip with its own JieFa AC8025, creating a development platform for the integration of cabin, driving, and parking functions. The core SoC has achieved a 100% domestication rate, and the cost has been significantly reduced. Furthermore, JieFa's AC8025 + JieFa AC784X integrated cabin and parking solution can reduce costs to a few thousand yuan, and it can even provide solutions with a single chip split.

Of course, in terms of high-end driving and parking chips, Horizon has become increasingly leading, having released the Journey 6 chip, while previous chips like the Journey 3 have begun to be more widely popularized, with user experience fully capable of meeting the basic needs of L2+ level.In the realm of domestic SoC suppliers, we can observe that domestic chip manufacturers represented by Xingchi have developed solutions covering mainstream products, including the X9U, X9H, and have even begun to build a new generation of central processors based on the X9 chip, which can realize the next step in the domestication process of central domain controllers. It is said that the X9 chip has achieved a larger scale of vehicle installation, with shipments reaching millions of pieces, and has begun to appear in some models priced around 100,000 yuan, enabling the infotainment systems of vehicles in this price range to operate mainstream vehicle interaction methods.

Additionally, Unigroup, which has collaborated extensively with BYD, has also developed a 6nm process cockpit chip supporting 5G, and Unigroup also has 4G vehicle chips targeting a more mainstream market, with shipments continuously increasing. Not to mention that Huawei has also begun to ship its own chips on a larger scale to meet the growing scale of Huawei's vehicle business unit.

It is now also noticeable that Qualcomm has also started to widely deploy the 8155 chip to more affordable products. Although the NPU computing power seems to lag significantly behind its own 8295 chip, for entry-level products, the 8155 chip is also sufficient to meet the current cockpit experience needs. After all, as mentioned above, AI large models are still a castle in the air, and even the computing power of the 8155 is more than enough for the daily use of the vehicle infotainment system.

Due to the cost breakthrough achieved by the domestic SoC supply chain, it can be seen that the infotainment and intelligent driving systems of a large number of joint venture car manufacturers have also begun to become popular. For example, Toyota has provided a new system developed based on the Harmony OS system for the new generation Camry and the annual Highlander, Nissan has also started planning to cooperate with domestic suppliers to develop new intelligent systems, and Volkswagen has already clarified the new domestic development model.The reduction in chip computing power costs has led to another example of an upgraded experience, which is the enhancement of in-car navigation functions.

Map providers such as Baidu, Siwei Zhenxin, and Gaode have successively released new 3D navigation maps, upgraded their human-machine co-piloting navigation capabilities, countdown functions for traffic lights, and have even begun to transplant lane-level navigation from mobile phones to in-car maps, significantly advancing the in-car navigation experience.

It is noteworthy that another group of companies vying for the in-car intelligent system experience is mobile phone manufacturers.

In addition to the well-known Apple's upgrade of the new CarPlay system, which allows navigation functions to be projected onto the driver's dashboard, companies like Xiaomi, OPPO, and Vivo have also begun to form a new mobile phone mapping alliance—three companies jointly established the "Smart Car Connect Open Alliance (ICCOA)," with other participants including Changan Automobile, Geely, SAIC Group, Qualcomm, and so on—The goal of ICCOA is to enable each company's mobile phones to uniformly project onto the in-car system in a manner similar to CarPlay.

Although the domestic mobile phone alliance is still in its infancy, once the alliance is formed, entry-level models can fully achieve an intelligent experience upgrade of the in-car system through mapping methods, without the need for car manufacturers to develop their own intelligent in-car systems. From the perspective of traditional car manufacturers, the mobile phone mapping method is sufficient for most consumers, and CarPlay has proven its ease of use and widespread adoption. In this case, traditional car manufacturers need not worry about the so-called in-car experience advantages of new forces, and their active participation is inevitable.04

Summary by Jia Shi

Looking back at the development of intelligent technology in China's automotive market in the first half of 2024, we have a fundamental consensus that:

High-end intelligent experiences have encountered development bottlenecks, while the intelligent experience of entry-level products has begun to popularize, and the experience gap between high-end and low-end models has been significantly narrowed.

The reason behind this phenomenon is that domestic chips have achieved autonomous control, and costs have been greatly reduced; while global flagship chips lack corresponding scene support, and the high cost of LiDAR required for high-level assisted driving also makes these high-end intelligent experiences fall into a "chicken ribs" state.A more critical change is that new force car manufacturers, which previously focused on intelligence as their main selling point, can no longer maintain a leading advantage—neither in cost nor in experience. Traditional car manufacturers are quickly catching up, especially with almost no difference in the smart cockpit experience.

In the current situation, the only advantage of new forces is the ability to integrate a global intelligent experience, such as NIO's attempt to integrate its own facilities like smartphones, vehicles, and battery swap stations to form an ecosystem similar to Apple's, and Huawei has also formed a similar ecosystem. For more new forces, they don't even have an ecological moat, only a simple intelligent car machine and so-called advanced driver assistance, which makes it difficult to form a clear difference with traditional car manufacturers.

What we can see now is that perhaps Jidu and XPeng are betting on whether they can run a pure vision-based solution without maps. If they can, then these two companies may have the advantage of balancing cost and experience.

The intelligence war in the second half of 2024 may become more peaceful, as everyone has played their cards. The real change may have to wait until the end of 2025 or 2026 to see if NVIDIA can really bring the Thor chip to various companies.