Shanghai – September 10, 2026 -- The Cloud Native Computing Foundation (CNCF) has graduated Karmada, an open source engine that coordinates applications across multiple Kubernetes clusters, clouds, and regions without requiring changes to the applications themselves. The milestone was announced at KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026, where three foundations shared a single stage in China for the first time.
Karmada v1.19 advances scheduling for distributed AI training workloads
The release adds multi-component scheduling for distributed AI training jobs and promotes priority-based scheduling to Beta status, enabled by default, ensuring critical workloads are scheduled ahead of lower-priority tasks. Short for Kubernetes Armada, the project extends the standard Kubernetes API with centralized placement, propagation, failover, and multi-cluster autoscaling.
Bloomberg, Trip.com and DaoCloud confirm production-scale reliance on the project
Bloomberg's Streaming Platform engineering lead, Michas Szacillo, said Karmada automates disaster recovery and simplifies management of individual clusters for the company's platform engineering teams. Trip.com's Honghui Yue said the tool lets the firm operate multiple clusters as a unified resource pool, support cross-cluster elasticity and failover, and perform large-scale workload migration with minimal application disruption. DaoCloud Chief Architect Kay Yan said the company is extending Karmada into its AI Token Factory architecture as multi-cluster inference across data centers grows in importance.
Adopter base spans Bloomberg, Wellhub and major Asian cloud and AI platforms
Production users include Alibaba Cloud, Huawei, Bilibili, iFLYTEK, JDCloud, Kuaishou, RedNote, SenseTime, Vivo, WPS, and ZTO, alongside global enterprises Bloomberg and Wellhub. Organizations deploy Karmada for hybrid cloud capacity, multi-region resilience, intelligent traffic distribution, AI training, GPU/CPU scheduling, and fleet-wide service-configuration distribution.
Contributor base reaches 1,214 developers across 292 organizations
Karmada's first commit dates to November 2020; it entered CNCF as a Sandbox project in September 2021 and advanced to Incubating status in December 2023. Since joining CNCF, the project has grown to more than 1,214 contributors across 292 contributing organizations and surpassed 5,600 GitHub stars. It exports Prometheus metrics across its control-plane components, packages an etcd instance for control-plane state, and ships Helm charts for installation.
Graduation required a third-party security audit and formal governance structure
To meet CNCF graduation criteria, Karmada completed a third-party security audit, established a steering committee, and adopted the CNCF Code of Conduct. The project also maintains a Core Infrastructure Initiative (CII) Best Practices Badge. CNCF CTO Chris Aniszczyk said graduation demonstrates the project has achieved the technical maturity, governance, and security work required for enterprise deployment, particularly as organizations scale Kubernetes across GPU-constrained AI environments.
2026 roadmap targets resource-aware scheduling for GPU and accelerator clusters
Karmada's broader roadmap moves multi-cluster Kubernetes from basic workload propagation toward a resource-aware control plane. Planned capabilities include priority-based preemption, multi-cluster queuing for AI training and batch jobs, and multi-cluster support for Kubernetes Dynamic Resource Allocation across GPUs and other accelerators.