Two years ago, I don't know who else remembers the declaration of "AI on the Car Year". However, at this year's Beijing Auto Show, almost every new car is vigorously showcasing what a true AI intelligent cockpit is. Of course, Volcano Engine has also released a new generation of automotive AI solutions based on the Agentic AI architecture, announcing that over 7 million intelligent cars equipped with the Big Bean Pack model have been produced, covering more than 50 brands and 145 vehicle models. Alibaba's Tongyi Qianwen announced that more than ten leading car companies, including Changan, Dongfeng, BAIC, BYD, Geely, Great Wall Motors, and Ideal, will simultaneously join the market. From foreign brands such as Mercedes Benz pure electric GLC to SAIC Audi E7X, from domestic brands such as Chery Xingtu EX7 to FAW Hongqi HS6 PHEV. These car companies are proving that AI can take over the brain of cars, and human car interaction has evolved from passive to active. The positive significance of AI loading is shifting towards a competitive track centered on experience.

However, this craze will eventually return to calm. As stated by Zhong Xuedan, Vice President of Tencent Smart Mobility, "Today, the interaction efficiency on mobile devices can reach 85 to 90 points, but the interaction between cabin and car systems has been stagnant for a long time at around 40 to 50 points." In the past year, the industry has generally faced the dilemma of "voice assistants being able to chat better," but the travel experience has not fundamentally changed. Instead of fantasizing about the infinite possibilities of an AI cockpit in empty talk, it's better to look down at our phones first. If even mobile terminals with faster iteration speed and lower technological barriers have not yet achieved the comprehensive popularization of AI, how can we expect geometrically complex cars to reach the sky in one step?

AI loading still requires time

The most direct reference to see the real progress of AI loading is the smartphone we use the longest every day. As the consumer electronics product with the largest global shipment volume, the fastest technological iteration, and the highest frequency of user replacement, the AI process of mobile phones itself is a mirror. If even mobile phones are still halfway through the popularization of AI, then in car AI cannot be achieved overnight. First, let's take a look at a set of current data. According to IDC's forecast, the shipment volume of China's new generation AI smartphones is expected to reach 147 million units by 2026, accounting for approximately 53% of the overall market. The first half, this number looks pretty good. But it should be noted that what kind of phone is considered a new generation AI phone? It is said that this label covers a wide range of standards, from the local deployment capability of end-to-end large models to the degree of calling cloud AI functions. You see, the description in this sentence is enough to exclude 90% of consumers from using AI. So for consumers, whether they can make good use of AI is the biggest issue.

Data shows that even among users who have purchased so-called AI phones, a high proportion of them actually use AI functions frequently. Most people still use the phone for social, video, and basic communication, and the penetration of AI remains at the surface level. By comparison, as of March 2026, the monthly active users of the Doubao Big Model reached 345 million. However, this is a purely consumer grade AI application, which is fundamentally different from the generative AI capabilities integrated naturally into mobile terminals. In early December 2025, the Doubao AI smartphone made its debut, priced at 3499 yuan and limited to approximately 30000 units for sale. This mobile phone is positioned as an engineering prototype in the water testing market by ByteDance, and the official website gave a special prompt, saying that it is "only for the industry people who need to experience the beanbag mobile phone assistant", and "does not promise the functional perfection of mature products, ordinary consumers must be careful to choose". However, due to its AI capabilities for cross application automation, the first batch of 30000 units quickly sold out and were once hyped up to high prices of thousands or even tens of thousands of yuan in the second-hand market. However, the vast majority of these 30000 early adopters are developers and digital geeks. Industry insiders commented that "Dou Bao has outlined the prototype of smartphones in the AI era", but its technological frenzy quickly encountered practical difficulties: mainstream apps such as WeChat and Meituan have banned the automatic calling of Dou Bao assistant due to AI simulation operations that touch platform security risks, and some users' WeChat accounts have even been restricted from logging in as a result. It is not difficult to see that this star product that has caused a sensation in the AI industry, with a shipment volume of 30000 units, cannot even be considered a luxury in the multi million level mobile phone market, and the vast majority of ordinary consumers have yet to see or use a real AI phone. It not only did not leave the developer circle, but also did not enter the mass market.

Why is the popularity of AI smartphones so slow? The fundamental reason lies not in the business model or user habits, but in a series of technological bottlenecks that have not yet been completely overcome. These bottlenecks also lie on the path of AI in car cockpits, and have even been magnified several times. A true 'AI car' needs to achieve more than just generating text or recognizing speech. It needs to integrate various functional domains such as vehicle control, intelligent driving, navigation, and cockpit to build an integrated closed-loop. The vast majority of current solutions claiming to be "AI intelligent agents" still have a huge gap from truly executing multi-step, cross scenario tasks. That is to say, the popularization of AI smartphones takes time, and in car AI needs to wait even more. It requires solid engineering accumulation, continuous algorithm iteration, and long-term polishing of user experience from the industry, walking through a long trial and error stage step by step. Rushing for success will only backfire, patience is the attitude that truly respects the laws of technology.

Don't let AI become a new marketing point

If the technological bottleneck determines that AI vehicle installation cannot be accelerated quickly, then the excessive packaging of car companies in promotion and the misplacement of consumers' understanding of AI are another gap that is more difficult to overcome than chip computing power. In fact, the painful lessons in the field of intelligent driving are already crystal clear. When the "intelligence" in marketing vocabulary far exceeds the actual capabilities of the product, and users have a serious lack of understanding of technology, conflicts, accidents, and trust breaches will follow one after another. The AI cockpit is highly likely to repeat this old path.

A survey shows that over 40% of consumers mistakenly equate L2 level assisted driving with autonomous driving. Under the strong promotion of words such as "advanced intelligent driving" and "zero takeover for 100 kilometers" by some salespeople, users have developed an extremely dangerous expectation: a car that cost hundreds of thousands should be able to drive on its own like in science fiction movies. However, the reality is that the system frequently fails to recognize and respond in complex scenes such as construction sections and rainstorm weather. The huge gap between good intelligent driving and actual driving methods has rapidly spread consumer distrust. If the technology is correct and the driver is not at fault, then whose fault is the ultimate result of mutual distrust among the three parties? Returning to the intelligent cockpit, watching any car launch event report, words such as "AI intelligence", "proactive prediction", "deep collaboration", and "reshaping the relationship between humans and vehicles" are everywhere. Car companies use these trendy terms to depict a desirable scene, but real user feedback is not like that. Whether on auto forums, social media or third-party complaint platforms, you can see that countless consumers complain about their car AI experience in different scenarios, which is very inconsistent. For example, if you understand something on the highway, you can only hear it when you enter the tunnel. Please say it again. How can we expect AI to do more if we can't even do basic voice interaction well? And these are only reduced items at the interactive level, the more core issue is that consumers have not regarded these functions as important chips in their car purchase decisions. The weight of intelligent cockpit functions in the influencing factors of car purchase decisions is much lower than traditional options such as exterior design, power performance, safety configuration, and brand reputation.

The real problem is that car companies use "AI" as a tool to upgrade configuration levels and differentiate manufacturing and marketing, but do not truly think from the perspective of users' cognitive level. For the vast majority of ordinary consumers, the term 'artificial intelligence' itself carries a mysterious and omnipotent aura. Once the experience is poor, there will naturally be a sense of distrust towards the AI cockpit. Also, currently intelligent driving requires the driver to take over. Does the AI cockpit also need the driver to take over when faced with choices during execution? In the past, drivers only needed to focus on driving the car and keeping a good direction. Nowadays, with the upgrading and increase of in car functions, it is believed that drivers will also be distracted to confirm the results of these functions while driving, and the risk of accidents will also increase. Open the bean bun, you don't need any tutorials, nor do you need to memorize cold command words. You can express your meaning naturally like sending a WeChat message to a friend. It will break down the intention, call appropriate tools, retrieve necessary information, and ultimately provide a clear execution result. However, as of now, there are still few complete AI solutions that have been implemented, and most car models' intelligent cockpits only have the basic functions of AI large models. So, if car companies really want to do a good job in AI loading, they must put aside the speculative mentality of harvesting the market with new concepts and return to the origin of user value. Ultimately, consumers are not foolish. They won't pay for a feature called 'AI' just because of it, but they will definitely question the brand because some features are not user-friendly or there is a gap in experience. So, those car companies that hype up AI without even doing a good job in basic skills will not ultimately reap sales, but complaints and reputation collapse. On the contrary, whoever can provide a simple, stable, and easy to learn cockpit AI experience like a bean bun on a mobile phone, can truly be considered an AI car.