According to a report by Late Auto, Li Auto is rumored to be delving into the in-house development of cloud-based inference chips. This endeavor follows the data flow architecture employed in its intelligent driving chips, though the project is still in its nascent stages. The objective of these chips is to offload some of the inference tasks that are currently managed by GPUs. This includes tasks like processing data for intelligent driving models and addressing requests for large language models. From a technical standpoint, the company may leverage the NPU (Neural Processing Unit) design from its in-vehicle systems. By integrating multiple AI computing units and incorporating high-bandwidth memory (Note: The term 'matching' has been contextualized as 'incorporating' for clarity and accuracy in the subsequent description), Li Auto aims to construct larger-scale inference chips, thereby sharing the research and development costs.