TLDRs;
- Huawei launches Flex:ai, an open-source tool to improve AI chip efficiency by 30%.
- The software manages GPU, NPU, and accelerator workloads using Kubernetes orchestration.
- Researchers from three Chinese universities contributed to developing Flex:ai’s core framework.
- Flex:ai aims to overcome China’s restricted access to advanced semiconductor technology.
Huawei has announced Flex:ai, a new open-source software platform designed to enhance the efficiency of AI chips.
The tool aims to improve processing utilization by optimizing how workloads are allocated across GPUs, NPUs, and other accelerators, promising significant performance gains for AI workloads.
Open-source software targets AI chips
Flex:ai allows AI processors to divide a single physical card into multiple virtual computing units, enabling more flexible and efficient use of hardware.
Huawei claims the software can improve processor utilization by roughly 30%, although this figure has not been independently verified. The company plans to release the platform through its ModelEngine developer community, inviting developers and researchers to explore and build on the technology.
The move reflects a broader trend among Chinese tech companies to develop software-driven solutions that maximize performance despite limited access to the latest advanced chips due to trade restrictions and export controls imposed by the U.S.
Huawei has rolled out new OPEN SOURCE software called Flex:ai, which is built on Kubernetes and designed to boost AI chip usage by about 30%.
It can split a single GPU or NPU into multiple virtual units and run several AI workloads at the same time, and it includes a scheduler… pic.twitter.com/0eBmINu6Z0
— Wall St Engine (@wallstengine) November 21, 2025
Collaboration with top Chinese universities
Huawei worked closely with researchers from Shanghai Jiao Tong University, Xian Jiaotong University, and Xiamen University to develop Flex:ai.
The collaboration highlights the growing intersection of corporate and academic expertise in AI software and hardware innovation within China. By pooling resources and knowledge, Huawei and its academic partners have crafted a system that aims to rival similar orchestration platforms developed abroad.
For comparison, Nvidia acquired Run:ai in 2024, a company that provides similar AI workload management software. Huawei’s Flex:ai seeks to offer a domestic alternative to such tools, ensuring that Chinese developers have access to advanced AI chip orchestration without relying on foreign solutions.
Virtual computing units enhance utilization
One of Flex:ai’s core innovations is its ability to partition a single processing card into multiple virtual units, effectively multiplying the chip’s computational capacity for AI tasks.
This approach enables researchers and companies to run multiple experiments simultaneously on a single GPU or NPU, significantly reducing hardware waste and improving throughput.
Flex:ai uses Kubernetes to manage resource allocation dynamically, distributing workloads efficiently across all available hardware. This approach aligns with global trends in cloud and AI infrastructure, where software orchestration is increasingly critical to maximize chip utilization.
Flex:ai responds to chip access limits
The launch of Flex:ai comes amid ongoing challenges for Chinese companies in obtaining advanced chips. Recent reports show Huawei’s Ascend AI chips still incorporate parts from foreign suppliers, including TSMC, Samsung, and SK Hynix, despite U.S. export restrictions.
By focusing on software optimization, Huawei can extract more performance from existing hardware while mitigating supply chain limitations.
The company is also ramping up AI chip production. In 2025, Huawei aims to double output of its flagship 910C Ascend chips, with plans to produce up to 1.6 million dies by 2026. Flex:ai’s software capabilities complement this expansion, ensuring that the growing number of chips are used efficiently across various applications in Chinese tech firms, including Alibaba and DeepSeek.
