KServe becomes a CNCF incubating project, for AI inference on Kubernetes

2025-09-29 · Publisher : CNCF · Domain : Cloud & DevOps

On 29 September 2025, the CNCF accepts KServe into incubation: the generative and predictive AI inference platform on Kubernetes, born in 2019, integrates with llm-d, Envoy AI Gateway and Gateway API.

### Background On 29 September 2025, the CNCF technical committee votes to accept **KServe** at the **Incubating** level; the CNCF publishes the detailed announcement on 11 November 2025. KServe is a generative and predictive AI inference platform for multi-framework deployments on Kubernetes. The project was born in 2019 from joint work by Google, IBM, Bloomberg, NVIDIA and Seldon, before going through LF AI & Data in 2022. ### Key facts - History: spun out of Kubeflow (KFServing), donated to LF AI & Data in February 2022, renamed KServe in September 2022, then accepted by the CNCF in September 2025. - Features cited: autoscaling with scale-to-zero, multi-node inference for large models, inference pipelines, routing and explainability. - Integrations: llm-d (disaggregated serving, prefix caching, new `LLMInferenceService` resource), Envoy AI Gateway since KServe v0.15, Gateway API and Envoy Gateway. - Users cited: Bloomberg, Red Hat, Cloudera, CyberAgent, Nutanix, SAP and NVIDIA. - Figures in the file: more than 300 contributors, 19 maintainers and more than 30 adopting companies. ### Why it matters For AI platforms on Kubernetes, KServe is one candidate building block for serving models in a standard way, without vendor lock-in. Incubation signals a project that has passed the first maturity criteria, but not yet the top level. IT services firms building inference platforms should look at its alignment with llm-d and Gateway API, which sketch a cloud native inference stack. ### Takeaway KServe joins the CNCF: AI inference becomes a front-line topic for the foundation, beyond training.

Tags : Kubernetes, IA, Open Source, Cloud natif, NVIDIA