Full Breakdown
Huawei Launches Flex:ai to Enhance AI Chip Utilization
11/26/2025, 3:03:56 PM
Overview of Flex:ai Technology
Huawei has officially launched Flex:ai, an open-source orchestration tool aimed at improving the utilization rate of AI chips within large-scale compute clusters. Announced on November 21, 2023, during the 2025 AI Container Application Implementation and Development Forum, Flex:ai addresses the simultaneous issues of "insufficient computing power" and "wasted computing power." The platform is built on Kubernetes, a widely used container management system, and is designed to optimize the deployment of AI workloads across heterogeneous hardware types, including GPUs and NPUs.
Core Features and Innovations
Flex:ai incorporates three primary technological innovations to enhance computing power utilization:
1. XPU Pooling Framework: Developed in collaboration with Shanghai Jiao Tong University, this framework allows a single GPU or NPU to be divided into multiple virtual computing units with a precision of 10%. This capability aims to increase the average utilization rate of computing power in small-model training and inference scenarios by 30%.
2. Cross-Node Remote Virtualization: In partnership with Xiamen University, this technology aggregates idle computing power from various machines within a cluster, creating a "shared computing power pool." This allows general-purpose servers without intelligent computing capabilities to access GPU/NPU resources remotely.
3. Hi Scheduler: Jointly developed with Xi'an Jiaotong University, this intelligent scheduler can sense the status of diverse computing power resources. It automatically selects the most suitable local or remote resources based on task priority and computing power requirements, facilitating optimal scheduling and resource allocation.
Open-Sourcing and Collaboration
Huawei has committed to fully open-sourcing Flex:ai, allowing developers from industry, academia, and research to access its core technological capabilities. This initiative aims to promote the establishment of standards for heterogeneous computing power virtualization and AI application platforms, ultimately leading to a standardized solution for efficient computing power utilization.
Context and Implications
The launch of Flex:ai comes amid ongoing U.S. export restrictions on high-end GPU hardware, prompting a shift within China towards enhancing software efficiency as a workaround for limited silicon supply. By focusing on software-side improvements, Huawei's Flex:ai seeks to provide a competitive edge in AI computing, particularly in environments utilizing Chinese silicon, such as Ascend chips.
Criticism and Concerns
While the introduction of Flex:ai has been met with interest, there are concerns regarding the lack of available documentation and benchmarks for the tool. Questions remain about the granularity of resource slicing, its interaction with standard Kubernetes schedulers, and compatibility with widely used GPU types. The open-source code has yet to be released, and the effectiveness of Flex:ai in real-world applications will need to be evaluated once it becomes available.
Verbatim Quotes
The introduction of Flex:ai represents a significant step in addressing the challenges of AI chip utilization and reflects Huawei's commitment to innovation in the face of external pressures.
