PUBLISHER: ResearchInChina | PRODUCT CODE: 2074811
PUBLISHER: ResearchInChina | PRODUCT CODE: 2074811
Research on OEMs' Software Strategies: R&D Focus, Development Strategies and Supplier Building Models of 30 OEMs
In this paper, we adopt a research framework covering 13 subsystems and 48 sub-dimensions to analyze the R&D focus, development strategies and supplier building models of 30 OEMs. The key trend directions are summarized as follows:
Functional Software Layer: Flexible Iteration Driven by Servitization and Atomization
The functional software layer enabled software-defined, modularized common requirements to realize rapid deployment and flexible integrated configuration of specific vehicle functions (such as autonomous driving, cockpit interaction and body control). Its core trends are "servitization" and "atomization", namely functions are split into independently developable, deployable and upgradable software components or services, with underlying capabilities called via standard interfaces.
Great Wall Motor launched the GWM*ONE S Platform, an all-powertrain intelligent super platform compatible with five powertrain forms: fuel, hybrid, plug-in hybrid, battery electric and hydrogen energy. As a native AI all-powertrain base, the GWM*ONE Platform will support diversified market demands for powertrain forms worldwide through modularization (49 core modules). Great Wall Motor plans to launch sub-platforms such as VIS, GWM*ONE A and GWM*ONE Q to cover different computing power levels and vehicle model classes. The platform decomposes the whole vehicle structure into 49 core modules and 329 shared components. By splitting module assemblies and standardizing interfaces between modules, it derives products of different powertrain forms, body types and driving control styles through flexible calling and combination just like "movable type".
At the hardware level, the platform adopts structural tree decomposition to divide 49 core modules and 329 shared components, including engine, transmission, air spring and battery. Reasonable splitting of module assemblies and standardization of inter-module interfaces endow the platform with high scalability. Similar to individual Chinese characters in movable type printing, a limited number of these modules can be combined creatively to build vehicle models of various categories, meeting diversified market demands.
The Super Intelligent Chassis based on Great Wall Motor's GWM*ONE S Platform represents top-tier technology integration under the "GWM*ONE" philosophy. Through full-stack self-developed hardware and AI collaborative control, it achieves in-depth integration of four major systems: powertrain, suspension, steering and braking. The Super Intelligent Chassis includes fully self-developed closed dual-chamber air suspension (5-speed height adjustment, 100-millimeter lifting), fully self-developed EDC damper, fully self-developed bidirectional 20° rear-wheel steering (+-10° ultra-large angle active steering), and virtual wheelbase adjustment function.
In May 2026, Stellantis Group launched the brand-new STLAOne modular platform, integrating five original different platforms into a single scalable architecture covering global B, C and D-class vehicle models. It can boost cost efficiency by up to 20% and shorten product launch cycles. Meanwhile, the platform supports multiple powertrain types including hybrid and battery electric, integrates the STLA Brain software platform, STLA Smart Cockpit and steer-by-wire technology, and supports full-lifecycle OTA updates for vehicles. Among them, the STLA Brain software platform features comprehensive OTA update functions and is capable of processing 30 modules, much more than the previous 10 modules, thus delivering high flexibility. It also serves as a service-oriented architecture fully integrated with the cloud which connects ECUs and the central high-performance computer inside a vehicle via a high-speed data bus.
STLA Brain breaks the current version matching issue between software and hardware, allowing software developers to quickly create and update multiple functions and services without waiting for new hardware releases. The above OTA update function significantly reduces costs for both customers and Stellantis Group, and simplifies vehicle maintenance for car owners, and is more conducive to vehicle value retention.
System Software Layer: AI OS Builds Reusable Underlying Capabilities.
The system software layer acts as a core bridge connecting upper and lower layers, covering system kernel, middleware, virtual machine and more, providing a stable, efficient and secure operating environment for upper-layer functional software. At present, many OEMs have built a "unified operating system kernel + middleware" framework. Based on microkernel, safe and reliable real-time operating systems (such as QNX, AUTOSAR Adaptive), hardware resources are managed, while middleware (such as SOME/IP, DDS, ROS2) provides standard communication, diagnosis, security and other services. With the rise of AI OS, OEMs build reusable underlying capabilities centered on vehicle computing power scheduling, cross-domain collaboration, data governance, security isolation and continuous iteration.
For example, Great Wall Motor's all-new model WEY V9X is equipped with a self-developed native AI cockpit-driving agent, composed of Great Wall Motor's native AI all-powertrain vehicle platform - GWM*ONE Platform, Coffee EEA 4.0, vehicle AI OS, the world's first dual VLA large models and bionic motion control system. As the "brain" of the GWM*ONE Platform, the vehicle AI OS deeply integrates multi-modal perception, autonomous decision-making and multi-agent collaboration capabilities of AI into the underlying layer of the operating system, enabling vehicles to possess human-like perception, decision and execution capabilities. It also embeds the world's first dual VLA large models responsible for vehicle thinking and decision, as detailed below:
The cockpit VLA model handles in-vehicle interaction and personalized services, perceiving passenger status and demands with sensors such as cameras and microphones.
The intelligent driving VLA model serves as the core of autonomous driving, constructing an accurate digital model of the environment surrounding the vehicle via multi-sensor fusion technology. Unlike traditional autonomous driving systems, the intelligent driving VLA model can not only identify objects but also understand scenario semantics.
In addition, the vehicle AI OS contains multiple function-specific agents such as driving agent, energy management agent and entertainment agent, which can collaborate with each other to complete complex tasks. For instance, when planning a long-distance trip, the driving agent takes charge of route planning, the energy management agent recommends charging stations and optimizes energy consumption, and the entertainment agent recommends music or podcasts according to passenger preferences, delivering comprehensive mobility services.
R&D Tools: AI Drives Restructuring of Development Paradigms.
The R&D tool layer covers process and systematic tools, data closed loop and development toolchain, serving as key support for SDV implementation. The core trend in 2026 is that AI enables R&D toolchains in an all-round way, evolving from "single-point efficiency tools" to "agent-driven full-link closed loop".
Chery put forward the concept of AI Equality in early 2026, enabling AI penetration in both vehicle systems and the whole manufacturing industry chain. As of January 2026, AI technology has generated a total value of over 2.3 billion yuan across Chery's entire industry chain, covering 8 major fields including R&D, production, supply, marketing, service, human resources, finance and legal affairs, with more than 20 million API calls and over 100 billion token calculations, becoming a core engine for cost reduction and efficiency improvement.
The industrial-level applications of AI are specified as follows:
R&D: Chery adopts AI4iCME technology to shorten the traditional material development cycle from 3-5 years to less than 1.5 years, reduce component modeling time to 1 minute, and lift coding efficiency by 40% and quality by 45% via AI programming assistant;
Manufacturing: AI completes scheduling in only 1 minute versus 210 minutes required for manual scheduling. For complex defects, the defect detection accuracy reaches 99% with a missed detection rate controlled below 0.05%, basically realizing automatic defect interception;
Marketing & Service: CheryGPT sales assistant and full-lifecycle AI accompanying service both enhance user experience and optimize operational efficiency.
Since the first half of 2025, over 1,500 R&D personnel at Geely have adopted Alibaba Cloud Tongyi Lingma, with AI-generated code accounting for more than 30%. Especially in general code logic development and code inspection & testing links, code development efficiency has increased by 20%. GAC Group and Huawei co-built an AI solution, utilizing computer vision models to identify defects in all production links and boost quality inspection efficiency in the manufacturing sector. Meanwhile, predictive large model and large language model are used to monitor health status of equipment 24 hours a day to reduce manual inspection demands. In the supply chain sector, multi-objective optimization heuristic algorithm and solver combined with sales big data facilitate intelligent dynamic production scheduling, cutting scheduling time by 60%, alleviating supply chain pressure and lowering inventory costs.
Vehicle AI Agents are evolving from passive voice assistants to agents with active perception and decision capabilities, and cockpit-driving integration and on-device large models becoming mainstream directions. Intelligence driven by large models will accelerate the arrival of the digital peak, and Artificial Intelligence Defined Vehicle (AIDV) is emerging as a new high ground in competition. Currently, multiple OEMs have started layout:
Application Software Layer: AI Agent Becomes the Core of Differentiated Competition.
The application software layer directly delivers services and experience to users, covering cockpit software, intelligent driving software, vehicle control software and AI empowerment. In 2026, AI agents will be fully deployed on vehicles, evolving from passive voice assistants to agents with active perception and decision capabilities.
For example, IM Ultra Agent, the super agent of IM Motors, is supported by three technical pillars: IM Fusion Nova cockpit-driving integrated full-domain fusion architecture, IM AD ZETA intelligent driving large model co-developed with Momenta, and Alibaba Tongyi Qwen large model. IM Fusion Nova, the industry's first "cockpit-driving integrated full-domain fusion" intelligent architecture, thoroughly connects three core systems including chassis-by-wire, intelligent driving AI and intelligent cockpit AI at the underlying heterogeneous computing architecture level, constructing a three-layer architecture of "global cerebrum + agile cerebellum + execution body":
Global Cerebrum: Tongyi Qwen large model, responsible for understanding user intentions, global scenario scheduling and multi-task coordination;
Agile Cerebellum: IM AD ZETA intelligent driving large model, responsible for driving scenario decision, risk prediction and vehicle motion control;
Execution Body: Fully wire-controlled digital chassis, responsible for converting AI decisions into precise and rapid physical vehicle actions.
This architecture realizes a full-link closed loop of "user one-sentence intention -> AI global decision -> collaborative execution of intelligent driving + intelligent cockpit + chassis", upgrading vehicle AI from "voice assistant only capable of understanding instructions" to "exclusive driver assistant with comprehension, decision and execution capabilities". It also represents the industry's first complete solution for vehicle embodied intelligence, evolving from concept to mass production and implementation.
Dongfeng Nissan N8 is equipped with the latest AI cockpit where the AI voice assistant Xiao Ni AI agent deeply integrates iFlytek Spark + DeepSeek-R1 large models, boosting vehicle control capability by 25% and supporting more than 700 smart scenarios. It enables free dialect dialogue, scenario prediction (automatically switching to child's voice upon detection of a child getting in the car via the camera), health management (real-time monitoring of driver's heart rate and blood oxygen) and other functions.
FAW Hongqi Lingxi Cockpit is equipped with Qwen agent, realizing recognition of multiple vague intentions, and complex route planning. When the user says "First drive to Peking University, find a convenient and delicious Peking duck restaurant along the road at noon, and send me to Terminal T3 before 5 p.m.", the system can quickly identify three distinct intentions and return precise planning combined with weather conditions.