Source | Bohu Finance (bohuFN)

In the early days of electric vehicles (EVs), when they were still struggling to gain traction, intelligence was a key differentiator from internal combustion engine vehicles. Tesla's Model S pioneered the replacement of physical buttons with a central touchscreen, marking the beginning of the electronic cockpit's evolution into an intelligent one.

However, the industry has yet to produce a mature product that defines an era, akin to the iPhone. Intelligence has largely been confined to more screens, simple in-vehicle systems, and the elusive promise of autonomous driving.

Yet, the emergence of large models has begun to give intelligence a more concrete form. Following Grok's integration, the recent 2026 Beijing International Auto Show showcased the mass adoption of large models in vehicles.

Automakers such as XPeng, Li Auto, Geely, IM Motors, and even Volkswagen have all introduced their own solutions. Intelligent driving suppliers like Horizon Robotics and Momenta, along with players such as Volcano Engine and Tencent, have also made their presence felt.

This signals a new trend: the automotive industry is transitioning from 'building four good wheels' to 'installing a superior brain.' This is not just a competition among automakers; whoever can create a smarter 'brain' and orchestrate a more formidable 'body' will stand at the new divide leading to the era of intelligent vehicles.

As Zhu Jiangming, Chairman of Leapmotor, puts it, intelligence capabilities will determine a company's survival in the next two to three years.

01 'Reinventing' the Vehicle Brain

Elon Musk stated at this year's Q1 earnings call, 'Memory bandwidth is one of the key elements for achieving unsupervised FSD (Full Self-Driving), a fundamental requirement for artificial intelligence in general.'

As is well known, just because Musk believes something will work doesn't guarantee its success—he's a master at painting a grand vision. However, when he says something won't work, it's often true, given his top-tier engineering capabilities.

In the past, the lackluster performance of EV intelligence was indeed related to factors such as technology and cost. However, these are no longer significant issues. Not only has the computing power of mainstream cockpit chips surged from 8 TOPS to 30 TOPS, but key breakthroughs have also been made in lightweight, low-power deployment technologies for edge models.

From a cost perspective, the cost of intelligent driving chips per vehicle has decreased by approximately 30%. The unit price of LiDAR has dropped from the range of 3,000-5,000 yuan to the four-digit level. As the engineering and licensing fees for large models continue to decline, the cost barrier for large-scale applications has been largely eliminated.

From traditional giants to new entrants, nearly everyone has introduced their own solutions simultaneously.

Last year, Tesla integrated Grok into its vehicles. In Musk's vision, this will enable the creation of a thinking car, a cross-platform AI system, and the mastery of data dominance in human-machine interaction.

IM Motors unveiled the 'Super Intelligent Agent IM Ultra Agent'; Geely introduced the Super Eva Intelligent Agent; Volkswagen released a roadmap for its 'Omni-Intelligent Agent AI.'

The Li Auto L9 Livis made its global debut, emphasizing the vehicle's human-like perception and coordinated movement capabilities. XPeng's second-generation VLA intelligent driving system will achieve deep cross-domain integration with the VLM large model, rivaling the Tesla FSD+Grok experience.

Automakers are taking diverse approaches. Some focus on creating a smarter 'brain'; others emphasize the coupling of intelligent driving systems with chassis decision-making; still, others start from the chassis system, integrating power, braking, and body functions across the board...

Regardless of the path, automakers share a clear consensus—enabling the vehicle's 'brain' to think proactively and control vehicle movements more precisely.

From the most primitive 'voice activation' to 'model integration,' automakers are now pursuing a true 'single brain.'

Yang Liwei, Vice President of Volcano Engine, provided an example: 'If a user is concerned about whether their child in the back seat is asleep, a traditional large model might adjust the lighting and air conditioning. However, the Doubao Cockpit Assistant goes further. If the child is asleep, it will adjust the cockpit to a more comfortable state. If the child wakes up, it will proactively engage, such as telling stories or playing music.'

The distinction here is that vehicles are evolving from 'passively following instructions' to 'proactively getting things done.'

02 The First Year of 'Large Model' Integration in Vehicles

However, achieving this is far from simple. The challenges lie not only in algorithms and computing power but also in enabling the 'brain' to develop eyes, ears, hands, and feet, achieving perception and control of the physical world—this is the essence of cockpit-driving integration.

The first challenge is enabling AI to truly understand user intent.

The second challenge is developing 'hands and feet.' An intelligent vehicle is a vast hardware system, and understanding and being able to control each vehicle function is an extremely complex task. If the problem is complex and the task chain is long, accumulated errors increase, making it easier to fall into a deadlock.

Different types of companies offer various solutions:

Volcano Engine's approach is to integrate the cockpit and driving models and data. The cockpit large model translates the user's generalized instructions into inputs that the intelligent driving system can understand, while the intelligent driving system determines whether and how to execute them.

Horizon Robotics, on the other hand, has chosen to use a single chip to run both intelligent driving and intelligent cockpit functions, reducing hardware costs. Li Auto's full-wire-controlled chassis eliminates mechanical transmission structures, transmitting steering instructions through electrical signals. Huawei and NIO follow a similar approach.

Additionally, there are safety challenges. As intelligent vehicles delegate certain decision-making powers to AI, the possibility of unpredictable behavior arises. Currently, there is no mature solution to this issue.

However, players are not avoiding this problem. Horizon Robotics' Kaka Shrimp operating system has pioneered the 'Castle Safety Physical Isolation Architecture,' achieving physical isolation between the cockpit and intelligent driving systems. Volcano Engine divides vehicle control permissions into three layers, with braking and steering belonging to AI-restricted zones.

After resolving these challenges, the vehicle itself becomes an intelligent entity capable of seeing, hearing, and performing tasks. This explains why automakers are flocking to invest in this area: firstly, there is more room for exploration in human-vehicle interaction; secondly, the anxiety surrounding AI intelligence compels everyone to take an early lead.

From a technical perspective, vehicles are currently a more suitable platform for AI agents than smartphones.

Attempts with Doubao on smartphones have demonstrated that large models struggle to break through existing ecological boundaries on mobile devices. However, vehicles are different. They have various driving scenarios, such as commuting, traveling, and business trips, offering far greater potential for AI assistance than smartphones.

Moreover, the information density within a vehicle cockpit is extremely high, and the functions that can be invoked are far more numerous than those on a smartphone. In the future, further connections to parking, autonomous driving, and external services can further expand the imagination.

From a market perspective, the competition among new energy vehicles is becoming increasingly homogeneous, with no significant differences in overall hardware capabilities. Intelligent agents are seen as the key variable to break the deadlock.

This is a two-way game. Automakers are eager to find differentiated advantages in the second half of the intelligence race, while large model companies and intelligent driving suppliers aim to seize industry influence, accelerating the integration of large models into vehicles.

03 Various Players Vying for a Share

In the past, the automotive industry focused on vehicle control, intelligent driving, and hardware, with automakers dominating the 'product is king' era. However, with the strong intervention of AI, vehicles will shift from being 'hardware-defined' to 'ecosystem-defined,' and automakers will no longer be the sole dominators.

Various players have already made their moves:

The first category is automakers.

For example, XPeng has built a unified AI middleware platform to support its intelligent driving, cockpit, and robotics businesses. Li Auto is not only deeply integrating MindGPT large models and intelligent agents with Alibaba Cloud but also promoting the technological convergence of intelligent cockpits and intelligent driving systems.

Furthermore, the 'binding' between automakers and large model/intelligent driving suppliers is becoming increasingly evident. Examples include Geely and StepFun; BMW, IM Motors, and Alibaba Cloud; Volcano Engine and Roewe; Great Wall Motors and Tencent; Dongfeng Motor and SenseTime Absolute.

Overall, new entrant automakers primarily pursue in-house research and development routes, viewing it as part of the brand experience, with embodied intelligence as their ultimate goal. Traditional automakers, on the other hand, are more likely to catch up on intelligence progress through collaborations.

The second category consists of large model companies and intelligent driving suppliers.

For instance, Volcano Engine offers two forms of cooperation: an AI cockpit suite solution and the Doubao Cockpit Assistant. Alibaba Cloud's intelligent cockpit can invoke Alibaba ecosystem services such as dining reservations, navigation, and payments. Tencent has released the 'Omni-Scenario Intelligent Agent Open Platform for Travel.'

There is also SenseTime Absolute's New Member Native Intelligent Agent; Banma Zhixing's Yuan Shen AI Automotive Robot Brain and its intelligent agent collaboration solution, Auto Claw; and Youjia Innovation 's (Youjia Innovation) cockpit intelligent butler, the 'BamBam Lobster Assistant.'

As the daily functions demonstrated by intelligent agents become increasingly perceptible to users, these infrastructure providers enabling automakers to build intelligent agent capabilities are also beginning to transition from the B-end to the C-end, vying for their own brand recognition.

The third category comprises chip and foundational platform companies, such as Huawei, Horizon Robotics, and MediaTek. MediaTek has integrated intelligent agent AI technology into automotive cockpit chips, while Horizon Robotics' cockpit-driving integrated whole-vehicle intelligent agent chip, Horizon 'Xingkong,' etc.

Their core logic is that regardless of which large model or ecosystem platform automakers ultimately choose, they can secure a ticket to the underlying hardware in advance.

There's a classic metaphor in the automotive industry: electrification is the first half, and intelligence is the second half. Electrification has brought everyone to the same starting line, while intelligence will create even greater disparities among competitors.

Currently, the race for intelligent agent integration in vehicles has just begun, and the final outcome is still far off.

The cover image and accompanying illustrations of the article are owned by their respective copyright holders. If the copyright owners believe that their works are not suitable for public viewing or should not be used free of charge, please contact us promptly, and our platform will make immediate corrections.