Kubeflow adds everything but the kitchen sink to Kubernetes, then asks CNCF for a diploma

Kubeflow announced a wave of updates at KubeCon Japan 2026, including Kale 2.0 for converting annotated Jupyter notebooks into production pipelines, a ground-up Notebooks v2 redesign with CRD-driven architecture, and a unified SDK with native Spark support and LLM fine-tuning blueprints. The new Kubeflow Trainer expands to cover distributed AI and HPC workloads via MPI, with proposals for hyperparameter optimization jobs and reinforcement learning support for LLM post-training. The project has also officially applied for CNCF Graduation, signaling its maturity as a production-ready ML platform on Kubernetes.