Huawei has developed the Unified Cache Manager (UCM), an advanced AI software platform designed to optimize memory resource management across high-bandwidth memory (HBM), DRAM, and SSDs, significantly boosting AI inference performance.

Unveiled in 2025, the UCM aims to enhance AI hardware efficiency, reducing inference latency by up to 90% and increasing system throughput by as much as 22 times, according to Huawei's Vice-President Zhou Yuefeng.

The UCM has been successfully tested in real-world applications with China UnionPay, demonstrating its ability to accelerate AI inference and cut operational costs for large-scale AI tasks.

Huawei plans to open-source the UCM platform in September 2025 to promote widespread adoption, development, and collaboration within the industry.

This initiative is part of China’s broader strategic effort to reduce dependence on foreign memory chips amid US sanctions that restrict access to advanced HBM hardware from vendors like Samsung, SK Hynix, and Micron.

The global demand for HBM chips is rapidly rising, with revenues expected to nearly double by 2025 and reach $98 billion by 2030, making Huawei’s innovations strategically important for China.

While UCM improves efficiency and helps mitigate hardware shortages, experts acknowledge it cannot fully replace high-end HBM hardware but provides a vital software-based solution amid geopolitical supply constraints.