Huawei Unveils Flex:ai AI Container Tech, Empowering Single Card to Tackle Multiple AI Workloads Concurrently
2025-11-21 / Read about 0 minute
Author:小编   

On November 21, 2025, Huawei officially introduced its Flex:ai AI container software. This innovative software leverages computational - power partitioning technology, which can finely divide the computational capacity of a single GPU (Graphics Processing Unit) or NPU (Neural Processing Unit) card into multiple virtual computational units. The partitioning precision is as high as 10%, allowing a single card to handle multiple AI workloads simultaneously. In the context of computing, this kind of multi - tasking capability on a single card is a significant breakthrough. In traditional setups, each AI workload often required a dedicated card, which could lead to inefficiencies, especially when some cards had idle computational power. Moreover, Flex:ai has the ability to aggregate the unused XPU (a general term for various processing units, similar to GPU and NPU) computational power from each node within a cluster. By doing so, it creates a 'shared computational power pool'. This is akin to how in a large - scale data center, instead of having individual resources lying idle, all available resources are pooled together for more efficient utilization. It has been reported that immediately following its release, Flex:ai will be made open - source in the Moqing Community, enabling developers and researchers worldwide to contribute to and benefit from this cutting - edge technology.