Currently, the deep integration of large AI models, agents, and automobiles is accelerating the development of AI cockpits.
Looking at the market, an increasing number of automakers are positioning AI cockpits as a key "selling point" to enhance product competitiveness and user experience, gradually deploying them in new vehicles.
At the same time, upstream supply chain players are also accelerating their deployment. From chip and OS vendors to Tier 1 suppliers, cloud services, large model providers, and application ecosystems, various suppliers are ramping up investment, attempting to secure a leading position in this wave of intelligentization.
Among them, iFLYTEK has leveraged its profound accumulation over the past two decades in intelligent voice and artificial intelligence to build a software-hardware integrated AI cockpit technology stack across multiple dimensions, including models, hardware, and Harness engineering. It successfully achieved mass production immediately upon the release of Spark Wisdom Cockpit 3.0, becoming one of the first samples to close the loop from "technology to mass production" in this industrial race.
Intelligent Cockpit Evolution Enters the AI-Native Era
Reviewing the past decade, the evolution of intelligent cockpits can be roughly divided into three stages.
The first stage was the Infotainment Cockpit, lasting approximately from 2010 to 2015.
During this period, cockpits primarily featured physical buttons and small central control screens. Functions such as navigation, Bluetooth, and music were gradually introduced but remained basic. These features were largely fixed at factory shipment, making it difficult to add new functions later.
The second stage was the Intelligent Connected Cockpit, beginning around 2016 and continuing until approximately 2023.
In this stage, touchscreens gradually replaced physical buttons, central displays grew larger, and the number of in-car screens increased. Voice interaction, vehicle connectivity, and OTA (Over-the-Air) updates were "brought on board," shifting the cockpit from closed to open systems with more applications related to mobility services and infotainment.
For example, iFLYTEK's Feiyu system supported multi-zone and multimodal interaction as early as 2018.
By 2020, the Feiyu system supported multiple languages, with each language covering seven major business areas—vehicle control, music, radio, navigation, commands, Bluetooth calls, and lifestyle services—involving over 200 interaction scenarios, including answering calls, playing music, and setting navigation routes via voice. This initially realized the concept of "using your mouth instead of your hands." Consequently, the cumulative user base for iFLYTEK's pre-installed products exceeded 30 million, with partnerships covering over 90% of mainstream Chinese independent and joint venture automakers.
However, the underlying interaction logic of cockpits during this stage did not fundamentally change; it remained "users finding functions." To perform a task, users had to know which app to open. At this time, voice acted more like a shortcut, capable of recognizing standard commands but struggling to understand complex natural expressions or complete cross-application tasks.
The third stage is the AI Cockpit, currently in its initial implementation phase.
The most fundamental change in this stage is not just an increase in feature quantity, but that AI transforms from an "external plugin" capability into the native foundation of the intelligent cockpit. Under this premise, AI cockpits exhibit five significant differences compared to previous generations:
In terms of service mode, they shift from passive command response (like "chatbots") to proactive service. The system combines user status and driving scenarios to provide relevant services before the user speaks, rather than waiting for instructions.
In terms of processing logic, they move from simple stacking of functions and apps to task-driven closed loops. Users no longer need to remember which app to open or switch between interfaces; stating what they want to do allows the system to decompose tasks, allocate resources, and execute them autonomously.
In terms of perception, they evolve from single-modal voice or vision to multimodal fusion. The system does not just listen; it integrates images, user status, and in-car environment to comprehensively judge user intent, even identifying and understanding the emotional states of drivers and passengers, truly transforming from a "tool" into a "warm emotional companion."
In terms of underlying architecture, they shift from fragmented domains to unified scheduling. Previously, navigation, air conditioning, and media functions operated independently without interoperability. Now, a unified AI agent hub coordinates these functions, allowing linkage so that a single command can mobilize multiple systems.
In terms of operation, edge-cloud collaboration has become the mainstream architecture. The edge side handles fast response to high-frequency commands, privacy protection, and offline availability, while the cloud manages complex reasoning and multimodal interaction. This combination ensures that vehicle capabilities continue to strengthen through updates after delivery.
iFLYTEK's recently released Spark Wisdom Cockpit 3.0 is a typical representative of the AI cockpit.
Positioned as the industry's first full-modal human-like interactive cockpit, Spark Wisdom Cockpit 3.0 is based on the SOPHIA architecture, comprising four systems—Full-Modal Perception, Thinking & Decision-Making, Multi-Dimensional Expression, and Memory Evolution—to form a closed loop.
Specifically, Full-Modal Perception enables the car to both understand user intent and perceive in-car conditions. The Master Agent in the decision-making layer serves as the unified scheduling hub for the entire vehicle, responsible for understanding intent, planning tasks, allocating capabilities, and supervising execution. In terms of expression, a customized "Spark Family" allows automakers to define brand personas. The memory system not only continuously accumulates user habits but also distinguishes preferences among different passengers, truly realizing "personalized experiences for every individual."
From the user perspective, Spark Wisdom Cockpit 3.0 currently delivers three main values: a more understanding emotional companion, a capable automotive manager, and an immersive audio-visual master.
In fact, as early as 2023, iFLYTEK released automotive application results based on the Spark Large Model and simultaneously upgraded its intelligent cockpit solution. As an early prototype of the AI cockpit, Spark Cockpit achieved freer, more anthropomorphic human-machine interaction based on multi-turn, multi-person, and multimodal capabilities. This established iFLYTEK as one of the "pathfinders" who completed early technical positioning in the AI cockpit field.
In the long run, the goal of AI cockpits will be to evolve automobiles into "Super Vehicle Intelligent Agents" with comprehensive cognition, proactive service, and continuous evolution.
Software-Hardware Integration Is the True Entry Ticket
While bringing large models onto vehicles has indeed ushered intelligent cockpits into a new stage of AI-native development, AI cockpits are not simply a superposition of large models and intelligent cockpits. Realizing these capabilities requires a complete system of coordinated software and hardware.
First, consider the software layer.
A complete AI cockpit software suite generally includes four parts: The bottom layer consists of AI development platforms and Agent orchestration, responsible for model development, deployment, and organizing/assigning tasks to various agents. Above that is Model-as-a-Service (MaaS), handling the collaborative invocation of large language, multimodal, and voice models across cloud, vehicle, and edge sides, solving "where the capabilities come from." Further up are AI applications, including intelligent voice, AI assistants, navigation Q&A, and content generation used directly by users. The top layer is AI operational services, supporting continuous updates and maintenance after models and applications go live.
Internet and cloud service providers such as Volcano Engine, Alibaba, Baidu, and Tencent primarily enter the AI cockpit track from the software and service layers, though with different focuses: some excel in underlying platforms and computing power, others in large model technology, and others in application ecosystems.
Overall, this layer determines the intelligence level of the AI cockpit and the richness of in-cabin applications.
However, scaling these capabilities "onto vehicles" depends heavily on hardware, including computing chips, cockpit domain controllers, various sensors, displays, and acoustic hardware.
On one hand, hardware is the physical carrier for software and models.
For instance, cockpit-side large models range from hundreds of millions to billions of parameters. Without matching computing power, bandwidth, and memory, smooth operation is difficult. Similarly, rich in-cabin applications and ecosystems cannot complete interaction and service loops without hardware like screens and acoustics.
In other words, the upper limit of the AI cockpit experience is ultimately constrained by hardware conditions.
On the other hand, hardware performs two tasks that software cannot replace: collecting information from the physical world and executing digital commands in the physical world.
No matter how smart the model is, it relies on cameras and microphones to capture images and sound to "see and hear." Commands issued by the model rely on motors to drive seats, air conditioning, and windows to truly land in the physical world. Inputting information from the physical world and outputting commands to the digital world cannot be substituted by writing any amount of code.
In this sense, software-hardware integration is not optional but an entry ticket for AI cockpits. Separating them means neither end constitutes a complete AI cockpit.
The parameters of edge-side large models must match the chip's computing power, bandwidth, and memory. Perception and execution must land on physical devices. On this basis, challenges regarding automotive-grade adaptation, cost control, and mass production rhythm must be solved to ensure software runs stably and efficiently on automotive-grade hardware. Only when both aspects are addressed can AI capabilities truly be delivered within the cockpit.
But software-hardware integration is merely a threshold, not the endpoint.
Crossing this threshold, the industry's competitive focus shifts further to mass production and ecosystems: the larger the scale of mass production and the more effective data accumulated, the faster the model evolves; the richer the application ecosystem, the more likely the cockpit is to retain users.
Ultimately, software-hardware integration determines "whether it can get on board," while mass production and ecosystems determine "how far it can go once on board."
From Technology to Implementation, Mass Production Reveals True Value
After more than a decade of development, the intelligent cockpit market has formed a relatively clear landscape: Desay SV Automotive leads in cockpit domain controllers, Qualcomm leads in domain controller chips, iFLYTEK leads in vehicle voice, and Huayang Multimedia excels in HUDs.
However, entering the AI cockpit era, the rules of competition are changing: without hardware, even powerful models and application ecosystems remain confined to the digital world; without software, even reliable hardware cannot generate intelligence on its own.
Therefore, facing the new trends in AI cockpit development, iFLYTEK chose to deeply integrate software and hardware along its strongest links—voice and large models—building a software-hardware integrated AI cockpit technology stack.
It is worth noting that in the AI cockpit field, iFLYTEK is one of the few full-stack players possessing R&D capabilities in large models, agent orchestration, intelligent voice, acoustic systems, and edge computing hardware simultaneously.
Specifically in the software layer, iFLYTEK's core product is the SparkAuto automotive large model. Based on the Master Agent as a unified kernel, technically corresponding to the four systems of the SOPHIA architecture (Full-Modal Perception, Thinking & Decision-Making, Multi-Dimensional Expression, Memory Evolution), it is divided into Cloud SparkAuto and Edge SparkAuto-EMM series.
The Edge SparkAuto-EMM covers sizes ranging from 2B to 35B and over 100 cockpit scenarios. It can be flexibly configured for different automakers and vehicle levels, balancing experience, cost, and security through edge-cloud collaboration.
iFLYTEK also possesses the SparkAudio voice foundation large model, as well as industry-specific large models for education and healthcare, providing more content and capability sources for cockpit applications.
Taking healthcare as an example, leveraging the Master Agent and iFLYTEK Xiaoyi, Spark Wisdom Cockpit 3.0 brings the "test, ask, diagnose" loop into daily commutes, connecting cars, homes, smartphones, clouds, and wearable devices from Huawei and Apple.
At the hardware level, iFLYTEK has deployed edge computing hardware and acoustic system products such as microphone arrays and speakers around cockpit scenarios, forming a配套 (supporting system) with edge models.
Regarding edge computing, iFLYTEK recently released two AI Boxes designed to directly host edge-side large models—SparkAuto BOX, available in Lite and Ultra series. SABOX-Lite is built on the RK1828 platform, offering 20 TOPS of computing power and hosting a 4B model. SABOX-Ultra offers 200 TOPS of computing power and hosts a 35B large model.
By packaging large models and computing power into standardized boxes, iFLYTEK transformed models from "algorithms running in the cloud" into "products shipped with vehicles, integrating software and hardware."
Moreover, centered on the SOPHIA architecture, iFLYTEK designed and developed the Spark Wisdom Cockpit 3.0 Harness Engineering, turning the SOPHIA architecture from an internal diagram into an industry-level engineering foundation capable of stable mass production and continuous evolution.
However, no matter how comprehensive the technology stack, it must ultimately be tested in real vehicle projects. From jointly defining requirements with automakers, completing automotive-grade validation, to mass production delivery and subsequent iterations, each link determines whether this software-hardware integrated capability can truly be installed in cars and used reliably.
This is precisely where iFLYTEK's strength lies.
Built on over twenty years of deep cultivation in vehicle voice, iFLYTEK has accumulated mature engineering implementation experience and extensive customer relationships in automotive validation and mass production delivery. Installation volume is the most direct proof.
According to GaoShi Auto Research Institute's intelligent cockpit configuration data, as early as 2024, iFLYTEK ranked first in installation volume among vehicle voice suppliers with 8.276 million sets of standard configurations, exceeding the second-place player by over 5 million sets.
In the following two years, iFLYTEK maintained a commanding lead in the vehicle voice market. In the first eight months of this year, its voice installation volume in the domestic passenger car market reached 4.361 million sets, accounting for a 42.7% market share. For comparison, AISpeech and Cerence, ranking second and third respectively, had installation volumes of 2.595 million and 1.056 million sets, corresponding to market shares of 25.4% and 10.3%.
Overall, iFLYTEK's automotive pre-installed products have cumulatively landed 81 million sets, partnering with over 50 automakers, covering more than 60 countries and regions globally, supporting 32 core languages, and delivering over 500 mass-produced vehicle models.
It should be noted that vehicle voice has now become the core interaction entry point for intelligent cockpits.
Traditional cockpit interactions centered on physical buttons and touchscreens, requiring users to locate controls and navigate menu hierarchies to complete operations. As cockpit features multiply, app-style hierarchical menus only increase interaction costs.
In contrast, voice interaction naturally fits driving scenarios: hands-free, eyes-free, interruptible, and continuous, perfectly matching high-frequency needs such as navigation, vehicle control, communication, music, weather, charging, and parking.
Especially in the last two years, the rapid evolution of AI large models has upgraded voice from "understanding commands" to "understanding intent," pushing intelligent cockpits towards "de-appification." Users simply state what they want to do, and the system executes directly, eliminating the need to switch between menus and apps.
Voice has thus transformed from a functional entry point into the primary entry point for scheduling whole-vehicle tasks. Occupying this entry point means controlling the first touchpoint of user intent.
More importantly, large-scale real-world vehicle interactions generate rich scenario data, feeding back into model iteration to make models more adaptable to in-car environments. Meanwhile, extensive adaptation across brands and vehicle types has accumulated engineering experience in running the same model stably on different hardware and vehicle types.
These factors collectively constitute iFLYTEK's first-mover advantage in the AI cockpit competition.
Leveraging such abundant mass production accumulation and superior product capabilities, Spark Wisdom Cockpit 3.0 achieved mass production immediately upon release. It has been率先 (took the lead in) deployed in relevant vehicle models from Chery, Great Wall Motors' WEY, and Dongfeng Nissan, quickly securing a key position in the AI cockpit transformation.
Next, as intelligent cockpits fully transition into the AI era, true influence will ultimately be established through installation scale and mass production delivery. iFLYTEK's task is to continuously convert its first-mover advantage into delivery and experience advantages, maintaining its lead in this long-distance race.