Li Auto Unveils AI Embodied Intelligence Strategy, Launches Mach VLA System and In-House Chip Initiative

At the Li Auto Livis Day event, Li Auto unveiled an AI-generated short film featuring a digital avatar of its founder, Li Xiang, and formally announced its strategic shift toward becoming an "embodied intelligence enterprise." The company argued that current smartphones and conventional smart vehicles are function-driven and lack true human-centric intelligence. To address this, Li Auto introduced the concept of a "non-traditional intelligent vehicle," characterized by four core elements: electric powertrain, professional driver capabilities, an AI computer, and a life assistant. To realize the "professional driver" capability, Li Auto launched the Mach VLA (Vision-Language-Action) system, powered by its self-developed Mach M100 chip. Built on a 5nm process, the M100 delivers 1,280 TOPS of computing power—not only supporting advanced driver assistance but also serving as a general-purpose AI computing platform. Through full-stack optimization, the Mach VLA system reduces emergency response time to just 0.28 seconds, approaching the limits of human reaction speed. Equipped with a 3DViT model, the system can understand physical environments in real time. During a live demonstration, the L9 accurately recognized audience gestures and spatial layouts. Li Auto outlined three key OTA upgrade milestones: - In July: 30% improvement in ADAS efficiency; - In September: full-scenario autonomous parking and road surface learning; - In December: Livis reaction speed 56% faster than humans, with support for recognizing traffic police hand signals. The company aims to deliver a user experience comparable to Tesla’s FSD v14 by Q4 2024. Technologically, Li Auto has opted for a VLA architecture that emphasizes direct perception of the physical world and action generation. It stressed that in-house chip development is critical to breaking through existing framework limitations. Because the new chip requires training on new data, current L9 Livis deliveries still rely on legacy vehicle data. A significantly improved model, trained on new data, is expected by the end of Q3. Li Auto plans to collect data from new models first, then distill the optimized models onto older vehicles equipped with NVIDIA Orin or Thor chips. In the domains of "AI computer" and "life assistant," Li Auto introduced the Mach Mind-Pro and Mind-Edge models. The latter shifts focus from merely fulfilling user requests to proactively anticipating needs—such as automatically adjusting cabin temperature and sunshades, and coordinating with active suspension to enable a "rocking cradle" feature. Li Auto is progressively deploying the Mach chip in new models like the all-new L8. Despite strong sales, the company faces significant time pressure to catch up with Tesla and Huawei’s Qiankun ADS—both of which use proprietary chips and have accumulated over 10 billion kilometers of ADAS driving data—within just six months. Li Auto has become the first automaker to explicitly state that switching to a self-developed chip necessitates rebuilding the entire data feedback loop. Overall, Li Auto’s product philosophy has evolved from traditional vehicle manufacturing to developing embodied intelligent agents, aiming to redefine the form of intelligent vehicles by understanding human intent and surpassing human operational efficiency.