Author | Liu Jiayi
In the past, whether a smart car was considered "cool" depended on battery performance and intelligent driving experience. However, users are now more concerned about how various "XX Shrimp" (a metaphor for quirky AI features) and cockpit large models interact and perform.
Especially as urban NOA (Navigate on Autopilot) capabilities of new vehicles tend to converge, the new skills constantly emerging in the cockpit—such as complex intent understanding, multi-person command digestion, and proactive observation of user emotions—have become key differentiators for new vehicles and areas where user experience gaps can be most easily widened.
Today's AI cockpit has long surpassed the limitations of traditional smart cockpits. It has evolved from a passive, tool-like assistant executing commands, to a task-oriented tool with multi-agent collaboration, and now to an AI "co-pilot" that proactively provides services. Looking ahead, it can also collaborate with intelligent driving systems to become the domain controller brain of the entire vehicle, expanding the room for imagination.
This is a booming market that cannot be underestimated.
In its latest report, "China Automotive Cloud Market Tracker," IDC weighed the AI cockpit market for the first time. Data shows that in the first half of 2026, the Chinese AI cockpit solution market size reached CNY 710 million. The full-year figure is expected to reach CNY 1.66 billion, rising to CNY 4.71 billion by 2030, with a compound annual growth rate (CAGR) of 30%.
The players sitting at the AI cockpit table are already positioned differently.
According to the report, in the first half of this year, Volcano Engine held the top position with a 45% share, followed by Alibaba in second place. Baidu, Tencent, iFLYTEK, and StepFun formed the second tier.
The 45% figure reflects an increase in market concentration and indicates that automakers are gradually leaning towards the "better solution" when choosing cockpit partners.
Market feedback supports this trend. Latest statistics show that the Doubao Large Model is currently deployed in over 7 million smart cars, covering more than 50 brands and 145 models, ranking first in industry deployment volume. Furthermore, cockpit solutions built by Volcano Engine have been deployed and implemented on flagship models of leading automakers such as Tesla, Mercedes-Benz, BMW, BYD, and SAIC.
The current AI cockpit market is sending out two concentrated signals:
First, the cockpit is no longer just a configuration for users to try out; it is becoming another "major component" alongside batteries and intelligent driving that both users and automakers take seriously.
Second, AI cockpits are evolving from "passive response" to "proactive service." Major players with mass production experience, ecosystem synergy capabilities, and strong C-end brand recognition will gain more initiative in this shift.
In the era of software-defined vehicles, a competition with the highest "AI density" and the best test of agent capabilities has kicked off in the cockpit sector.
01. How Far Has the "Human-Like Feel" of AI Cockpits Come?
Over the past two years, smart cockpits have collectively shifted gears to AI cockpits, all emphasizing one term: "human-like feel" (or "vitality").
However, the human-like feel emphasized by most cockpit products remains relatively shallow. They can engage in natural dialogue but fail to understand vague commands; they can integrate complex vehicle controls but do not execute them proactively; they can order food or create travel guides but cannot accommodate multiple users' needs simultaneously. In essence, they provide some emotional and practical value, but not enough.
Although many cockpit solutions have achieved rich functional coverage and ecosystem integration, their proficiency in the most fundamental aspects of need understanding and intent parsing still lacks depth.
Tesla's Grok serves as a good control group. Its popularity stems not only from its linkage with intelligent driving capabilities but also from the embedding of an emotional personality during interaction—it can banter, use memes, and understand implied meanings.
This capability is crucial. It is akin to someone growing up in a local cultural environment suddenly listening to American stand-up comedy; often, they fail to grasp why the audience laughs because many jokes are only understood by locals.
Therefore, a cockpit AI's understanding ability must at least reach the linguistic level of "throwing and catching memes" to quickly bridge the distance with users, increasing user stickiness through natural and interesting interactive tension.
Tesla's introduction of the Doubao Large Model in the Chinese market provides a case worth observing.
Setting aside regulatory and data restrictions, the key point is that Grok may not possess the robust "local internet sensibility" of Doubao, which allows it to reliably connect with Chinese users.
We tested the performance of the Doubao Large Model integrated into Tesla. Although currently limited to conversational interaction, its display of "human-like feel" is quite penetrating.
It can use internet slang common among young people, such as "social death" or "vanishing in place," while telling jokes; it can deconstruct the debate trap of "should couples check each other's phones" with high EQ logic; and even when you hum a song on a rainy day, it can select a track with a highly atmospheric vibe based on the external environment. This demonstrates not just intelligence, but importantly, its ability to automatically judge the user's true intent within context, respond cleverly, and maintain appropriate boundaries.
However, the Tesla case only highlights Doubao Large Model's solid foundation in semantic understanding. Volcano Engine's cockpit capabilities go beyond this. The "Doubao Cockpit Assistant" it developed, as a more complete cockpit product form, aims to deliver a comprehensive suite of AI cockpit capabilities.
Simply put, it not only knows what to say but also how to act.
Currently, the Doubao Cockpit Assistant debuted on the SAIC Roewe Jiayue 07, launched on October 23. Its cockpit capabilities translate the concept of the "AI-native car" for the Jiayue 07 into a tangible product form, specifically divided into three layers:
At the execution layer, it decomposes vague intents into continuously executable tasks. The Jiayue 07 integrates over 2,000 SOA (Service-Oriented Architecture) service interfaces across the vehicle. In tests, when the system understood "the passenger seat is tired," it executed a series of actions: raising the lumbar support, lowering the AC temperature, and switching music to a soothing style. Even semi-joking commands like "there is a mosquito on the driver's dashboard, blow it out" triggered a response involving the driver's window.
At the expression layer, the Doubao Cockpit Assistant's feedback has physical channels. The Jiayue 07 features a dedicated "Doubao Screen," similar to NIO's NOMI. Through physical head turning and facial expression changes, the system achieves anthropomorphic expression on an emotional level, such as swaying to music or responding to emotions with actions. The core value of this screen lies in supplementing the "anthropomorphism" of cockpit interactions.
But to interact smoothly in complex real-world scenarios, an anthropomorphic shell alone is insufficient. The reason the Doubao Screen can precisely look at targets relies on the model layer's powerful multi-zone voice recognition and multi-person conversation capabilities. This enables the system to accurately determine who is speaking, from which location, and with what intent in multi-person scenarios, adjusting orientation and gaze accordingly. This hardware-software synergy reduces confusion caused by conflicting multi-person commands and provides concrete physical anchors for the cockpit's proactive service capabilities.
That is, judgment layer capability: completing judgments before the user speaks. Relying on OMS (Occupant Monitoring System), DMS (Driver Monitoring System) cameras, and four-zone independent recognition, the Doubao Cockpit Assistant continuously senses the status, voices, and movements of occupants inside the vehicle, combining context from continuous conversations to judge users' true needs. For example, if a user says "it's stuffy," it does not mechanically execute "open window" or "turn on AC," but rather judges whether to switch to external circulation, adjust temperature, or proactively ask if the user wants to change the fragrance, considering AC status, outdoor weather, and occupant conditions. Meanwhile, the system possesses long-term memory and continuous learning capabilities, allowing users to directly personalize settings, enabling the cockpit assistant to increasingly align with the owner's exclusive preferences over time.
The overlay of these three layers gives the AI cockpit's "human-like feel" its thickness. More precisely, the Doubao Cockpit Assistant truly reaches the "central brain" layer of the car: it no longer responds linearly by function but forms a complete chain from perception, judgment, to execution and feedback, much like a human.
From actual testing experiences, it has already demonstrated what a first-tier AI cockpit should look like.
02. The Battle for AI Cockpits Requires Both Offense and Defense
For Volcano Engine, a 45% share represents not just leadership in deployment numbers, but also its ability to secure designations for flagship models of leading automakers.
This becomes clearer from the automaker's perspective.
For top-tier automakers, the AI cockpit is critical to ensuring the next model holds its ground. However, each faces different situations and problems to solve:
- Tesla needs to compensate for weaknesses in localized interaction in the Chinese market, maintaining "race-grade" intelligent experience standards despite Grok's absence;
- SAIC Jiayue aims to use the AI cockpit to support the product definition of an "AI-native car," carving out a differentiated path in the fiercely competitive price segment;
- Mercedes-Benz hopes to leverage the AI cockpit to align its experience with luxury quality while catching up in intelligentization.
You can see that all three automakers face tough battles, yet they all chose Volcano Engine as their partner.
When new forces, traditional automakers, and luxury brands lock onto the same option, it underscores that Volcano Engine's cockpit product capabilities are merely the entry ticket; engineering stability and service quality are the keys to widening the gap.
Examples in tech history are not rare. For instance, Apple has long entrusted its core chips to TSMC (Taiwan Semiconductor Manufacturing Company) rather than Intel. The reasons include not only TSMC's more advanced process nodes but also its technology, stable manufacturing capabilities, and the trust derived from "not competing with customers."
Similarly, when automakers hand over cockpit AI to external partners—especially players like Tesla who are stubbornly committed to full-stack self-development—concerns extend beyond whether the model is smart enough, to data boundaries, brand definition rights, and mass production delivery rhythms.
Facing these escalating rigorous considerations, Volcano Engine's solution mainly manifests in two aspects.
First, it provides callable capabilities, not a system that must be accepted wholesale.
Currently, Volcano Engine offers two product lines:
Doubao Cockpit Assistant Solution. A complete "turnkey" solution.
AI Cockpit Suite. A set of AI capability modules that can be disassembled and combined on demand. Its foundation is the widely mass-produced Doubao Large Model intelligent cockpit solution, upgraded with an Agentic architecture.
For automakers, the keywords for the AI Cockpit Suite are layered options and on-demand configuration. Automakers can integrate only the Doubao Large Model and build their own cockpit AI system, purchase only engine-like skills, or buy mature interaction components.
Second, guarding boundaries and retaining overall vehicle definition rights for automakers in product design. According to Volcano Engine's design, the Doubao Cockpit Assistant can input interpreted user intents to the intelligent driving system, but the final closed-loop driving decision is completed by the automaker or Tier 1 intelligent driving supplier. This partially addresses automakers' concerns that cockpit suppliers will not extend AI capabilities into the driving domain.
At the whole-vehicle definition level, Volcano Engine provides the cockpit AI experience layer, while brand, user relationships, and product definitions remain in the hands of automakers. Automakers can also configure and manage data access and permissions.
Tesla's introduction of the Doubao Large Model in the Chinese market itself has reference significance. An automaker extremely sensitive to data sovereignty is willing to hand over part of its cockpit AI capabilities to an external partner, provided that boundaries can be clearly defined.
Volcano Engine securing a 45% share is due partly to model capabilities; "ease of integration" and "clear boundaries" are likely reasons equally valued by automakers.
03. New Big Three for Smart Cars: Battery, Intelligent Driving, Cockpit
Market concentration in an industry often results from the interplay of supply and demand.
The fact is, AI cockpits are deeply influencing users' purchasing decisions. A typical case is the positive market feedback for the SAIC Roewe Jiayue 07, which received over 40,000 pre-sale orders, with confirmed orders breaking 10,000 within 1 minute and 53 seconds after launch.
While this achievement cannot be entirely attributed to the cockpit—as price, space, and brand factors also play roles—it at least indicates that positioning the AI cockpit as a core selling point has genuine driving power.
Thus, today's AI cockpit has begun entering the "user attention list" alongside batteries and intelligent driving, with the three forming the "New Big Three" of smart cars.
From B-end automakers to C-end users, everyone cares not only about the specific performance of the "New Big Three" but also about who stands behind them.
If CATL has established distinct market awareness in the power battery sector, and Huawei, Momenta, and Horizon Robotics have done so in intelligent driving, then Volcano Engine is also establishing its competitive position in the AI cockpit sector through market share and mass production scale.
In terms of share, several top players occupy significant proportions in their respective tracks.
CATL's domestic power battery market share in Q1 2026 was 50.1%, returning to half the market after five years, with second-tier manufacturers trailing by an order of magnitude.
Meanwhile, Huawei, Momenta, and Horizon Robotics present a "tripod" pattern among third-party urban NOA suppliers.
Now, Volcano Engine occupies nearly half of the AI cockpit market share.
As the AI cockpit begins to determine a vehicle's differentiated experience and the upper limit of product definition, automakers' choice of cockpit suppliers is effectively selecting the foundation for the next round of product competition.
Viewed this way, Volcano Engine currently holds two advantageous positions:
First is the scale position, corresponding to mass production stability. Deployment in 7 million vehicles, implementation across 50+ brands and 145+ models, and daily average of 30 million interactions mean its suite has been validated by sufficient models and scenarios, keeping the cost and risk for the next automaker relatively controllable.
The second is the benchmark position, representing product advancement. The complete product form represented by the "Doubao Cockpit Assistant," accompanying the delivery and implementation of the Jiayue 07, gives Volcano Engine a window to showcase its capabilities externally.
Of course, unlike batteries and intelligent driving—which have entered stages of supply chain stabilization and experience convergence—the narrative for cockpits is far from concluded. This is precisely where cockpits differ most from the other two "big components."
The landscape for batteries and intelligent driving leans towards "convergence": once a technological route creates a gap, suppliers rapidly concentrate at the top, leaving fewer chances for latecomers to turn the tables.
But cockpits are different. Half of the competitive variables in cockpits lie in the hands of automakers, including how wide to open functional entry points, how much data to release, and to what extent to yield brand definition rights—none of which are determined solely by suppliers.
This means that market concentration in cockpits will not be determined solely by technological convergence like batteries and intelligent driving, but also depends on the degree of openness automakers are willing to accept.
However, the advantage of first-movers remains important in this "openness game," implying richer competitive chips and earlier accumulation of usage data and engineering experience from real-world scenarios.
At least for Volcano Engine, which is already leading by a body length, the trust of top automakers is building a moat for the next round of competition. Winning depends on product and boundaries; defending depends on trust and flywheels. Transforming positional advantage into ecological niche advantage is the answer sheet it must submit next.