Huawei introduced Flex:ai, an open-source software tool designed to optimize the use of AI chips.

The tool manages and allocates processing power across GPUs, NPUs, and other accelerators using Kubernetes.

Huawei claims it can improve processor utilization by 30%, though this is not independently verified.

Flex:ai can divide a single processing card into multiple virtual computing units, and researchers from Shanghai Jiao Tong University, Xian Jiaotong University, and Xiamen University contributed to the project.

Huawei will open-source the software through its ModelEngine developer community.

The release comes as Chinese companies push software solutions to address limited access to advanced chips, and it is similar to orchestration tools developed by other firms, including Run:ai, which Nvidia acquired in 2024.

🔗 Source: South China Morning Post

Recent Huawei developments