In a presentation at the Li Auto Software and Embodied AI Day event, the company's Chief Technology Officer Xie Yan delivered a keynote address.

Discussing the internally developed Mach M100 chip, Xie Yan stated that the goal is not merely to create a faster chip than previous versions, but to engineer an entirely different type of chip. He explained that AI computation is inherently parallel, with data as tensors, deterministic relationships, and clear data flow paths. The design philosophy is to let data flow drive computation and to architect the chip's design around the nature of AI computing. This, he said, is the dataflow architecture, and the Mach M100 is the world's first dynamic dataflow AI chip.

He detailed that the Mach M100 utilizes a 5nm automotive-grade process, delivering a single-chip computing power of 1280 TOPS. This makes it the most powerful mass-produced automotive-grade inference chip currently available worldwide. Due to its dataflow architecture design, the actual operational efficiency of the Mach M100 exceeds 82%. In direct comparison with the globally recognized and most mainstream chip in the intelligent driving domain, the Nvidia ThorU, the Mach M100 demonstrates comprehensive leadership across every test, including CNN-based backbone networks, UniAD, and Li Auto's core MindVLA model. Xie Yan emphasized that this is not a slight lead, but a performance gap measured in multiples.