BYD, XPeng, Li Auto Unveil Self-Developed In-Vehicle Full-Domain AI Models, Top Car Makers' R&D Exceeds RMB 10

On July 19 at the WAIC Auto AI session, BYD, XPeng, and Li Auto unveiled their respective self-developed full-domain in-vehicle large models. The cumulative R&D investment of leading car companies has exceeded 10 billion yuan.

Full-domain in-vehicle AI integrates cockpit, autonomous driving, and vehicle-road coordination into a cohesive system. Data from cameras capturing road conditions, passenger status, and vehicle parameters flows into a single large model. The model adjusts driving strategies based on road conditions, regulates cabin temperature according to passenger status, and anticipates risks through vehicle-road coordination. Previously, the cockpit handled only infotainment and autonomous driving focused solely on driving; now the two systems work in tandem.

The smart electric vehicle industry has long faced problems such as lack of coordination between cockpit and autonomous driving, limited interaction capabilities, and insufficient on-device computing power. The full-domain architecture unifies all data links within the vehicle, dynamically sensing both internal and external environments. A lightweight model is adapted to the vehicle chip, theoretically achieving integrated scheduling of cockpit and autonomous driving. Past attempts at middleware and multi-system integration yielded limited results. The full-domain large model uses a single unified model base.

A mature self-developed in-vehicle large model requires building an algorithm team, procuring computing power, and long-term road testing for iterative improvement. Ten billion yuan is just the starting point, with continuous annual investment to follow. On the hardware side, older vehicles cannot run the new models, so the new models are only installed in new cars.

Previously, AI large models were developed by technology companies, while car manufacturers acted as hardware producers and purchasers. After leading car companies began self-developing models, they became core players in the AI track, possessing vehicle terminals and real-world road data. Autonomous driving technology companies such as Horizon Robotics and Haomo. ai are promoting in-vehicle solutions, and general-purpose large model providers have also launched lightweight versions for vehicles. Car companies still have ground to make up in algorithm talent and technological accumulation.

The technological direction is to integrate cockpit, autonomous driving, and vehicle-road coordination, turning the car into a mobile intelligent space. Widespread adoption requires overcoming thresholds such as balancing computing power and cost, establishing a safety system, ensuring data compliance, and identifying genuine user needs. The concentrated unveiling at WAIC signals that the in-vehicle AI track has entered an explosive phase, with car companies committing tens of billions of yuan in investment, setting the stage for a competitive race.