CNCF Graduates Karmada: Multi-cluster Kubernetes Orchestration Reaches Production Maturity

Foundation (CNCF) announced that Karmada — short for Kubernetes Armada — has graduated. Graduation signals that the project met CNCF requirements for technical maturity, governance, and security and is considered production-ready within the CNCF portfolio.

Karmada is designed to run applications across multiple Kubernetes clusters, clouds, and regions without changing the applications themselves. That capability is increasingly important for organizations that need hybrid cloud capacity, multi-region resilience, intelligent traffic distribution, and GPU/CPU scheduling for AI workloads. The project has seen adoption across large enterprises and major cloud, telecom, AI, and travel platforms.

The v1.19 release advances multi-component scheduling for distributed AI training jobs. It also promotes priority-based scheduling to Beta and enables it by default. These changes aim to ensure critical workloads are scheduled first in environments where GPU and other accelerator capacity is constrained.

Karmada lists a wide production adopter base. Named users include Bloomberg and Wellhub outside China and many prominent Chinese companies and cloud providers such as Alibaba Cloud, Bilibili, Huawei, iFLYTEK, JDCloud, Kuaishou, SenseTime, Trip.com, Vivo, WPS, and ZTO. These adopters apply Karmada for hybrid cloud capacity, multi-cluster application delivery, fleet-wide configuration distribution, and multi-cluster AI infrastructure.

Project growth and ecosystem integration

Governance and security steps for graduation

To reach graduation, Karmada completed a third-party security audit, formed a formal steering committee, adopted the CNCF Code of Conduct, and maintains a Core Infrastructure Initiative (CII) Best Practices Badge. Those steps are part of CNCF's standard expectations for graduated projects.

Karmada's 2026 roadmap moves beyond basic workload propagation toward a resource-aware multi-cluster control plane. Planned work includes:

• Priority-based preemption for critical workloads
• Multi-cluster queuing aimed at AI training and batch jobs
• Multi-cluster support for Kubernetes Dynamic Resource Allocation (DRA) across GPUs and other accelerators

CNCF's CTO highlighted the importance of a production-ready way to coordinate fleets of Kubernetes clusters, especially in GPU-constrained AI environments. Karmada's maintainers positioned graduation as a milestone and a starting point for deeper production adoption and broader ecosystem collaboration.

Karmada's graduation marks its arrival as a production-grade solution for multi-cluster orchestration. Organizations looking to run distributed workloads across clusters, clouds, and regions — particularly those managing constrained GPU resources for AI training and inference — can consider Karmada as an option that already has real-world, large-scale adopters and a concrete roadmap for more resource-aware multi-cluster capabilities.