Beijing Xingyun Integrated Circuit Co., Ltd., a domestic innovator in GPGPU technology, has successfully completed multiple rounds of financing, raising over 400 million RMB. The investment was led by Wuyuan Capital, SAIF Partners, and Primavera Capital, with participation from local state-owned assets and industry capitals. Established in August 2023, Xingyun is focused on developing new-generation inference chips for large AI models, using non-3D DRAM architecture to create high-memory GPGPU products compatible with CUDA. This approach aims to address the increasing memory demands of large AI models, which have grown from gigabytes to terabytes, surpassing the cost of computing power chips.
Founder Dr. Ji Yu, a Tsinghua University graduate and former Huawei “Genius Youth” program participant, emphasizes that reducing costs lies in memory rather than computing power. Xingyun’s strategy involves bypassing high-cost HBM in favor of more affordable storage media such as LPDDR and NAND, significantly reducing memory costs. To counteract the lower bandwidth of these media, Xingyun employs a multi-channel, parallel design to meet data throughput needs.
Xingyun’s technical strategy focuses on system-level design capabilities, enabling flexibility in adapting to rapid AI application changes. The company’s “Brown Ant All-in-One Machine” has demonstrated the feasibility of sparse models on non-high-end hardware, with plans to bring self-developed chips to market soon. Investors praise Xingyun’s innovative approach and potential to revolutionize AI inference computing costs, with expectations for their new chips to drive widespread adoption of large-model inference.
