On August 12, Huawei introduced its groundbreaking AI inference technology, Unified Cache Manager (UCM). UCM is a comprehensive suite focused on KV Cache, which integrates diverse cache acceleration algorithms and tools to manage the KV Cache data generated during the inference process in a hierarchical manner. This innovative approach is designed to enhance the inference context window, bolster inference efficiency, and minimize the cost per token inference. According to reports, UCM has already been successfully tested in three key business scenarios of China UnionPay, demonstrating its capability to accelerate AI inference for smart finance applications. Additionally, Huawei announced its intention to officially open source UCM within the MoEngine community in September 2025.