The company's independently developed first-generation AI server based on heterogeneous architecture completed R&D and entered the trial operation stage, providing the compute foundation for subsequent LLMs and industrial scenarios.

The server adopts a heterogeneous architecture of CPU+GPU+proprietary accelerator cards, supporting high-density deployment, with single-rack compute density increased by 40% and energy efficiency optimized by 25%. The server has been put into use on the internal training platform, running stably for over 1,000 hours.

This is an important breakthrough for the company in the computing hardware field, providing valuable system integration experience for the subsequent productization of proprietary AI silicon.