Chinese smart-EV maker Li Auto has detailed its in-house Maha M100, a 5nm automotive-grade AI inference chip that delivers 1,280 TOPS of computing power per unit at an 82% utilization rate. The company unveiled the silicon at its "Livis Day" software and AI launch event, marking another step by Chinese automakers toward self-developed driving compute.
A dynamic dataflow architecture
Built on a dynamic dataflow architecture, the M100 supports autonomous driving, large language models and AI agents. Li Auto says it offers roughly three times the effective performance of NVIDIA's Thor-U while cutting end-to-end latency by 40%, and describes it as the world's first mass-produced dynamic data-flow AI chip after entering production in May.
Dual-chip L9 Livis
The new Li L9 Livis SUV is fitted with two Maha M100 chips for a combined 2,560 TOPS, paired with a 3DViT perception model that extends visible range by 50% and the Maha VLA 2.1 visual-language-action stack. A full drive-by-wire chassis enables millisecond-level response across the control chain.
China's smart-driving chip race
Li Auto joins rivals Nio and Xpeng in designing proprietary autonomous-driving silicon to reduce dependence on foreign suppliers. The trend mirrors broader compute moves such as OpenAI and Broadcom's first inference chip and Xpeng's own Turing-powered robotics roadmap, while reshaping the autonomous-driving supply chain.
Reporting based on coverage from CnEVPost and Internet Info Agency.