Adoption & Ecosystem
Who uses it
The repo has no ADOPTERS.md. The named organizations below come from the CNCF Incubating announcement and a Bloomberg engineering story; only cited adopters are listed.
| Organisation | Use case | Source |
|---|---|---|
| Bloomberg | Built its ML inference platform on KServe; also an original co-creator | Bloomberg story |
| Red Hat | Listed adopter; ships KServe in its AI platform | CNCF blog |
| Cloudera | Listed adopter | CNCF blog |
| CyberAgent | Listed adopter | CNCF blog |
| Nutanix | Listed adopter | CNCF blog |
| SAP | Listed adopter | CNCF blog |
| NVIDIA | Listed adopter; also an original co-creator | CNCF blog |
Adoption signals
From the GitHub API for kserve/kserve, observed 2026-06-24: 5,611 stars, 1,542 forks, 70 watchers (subscribers), and 636 open issues plus PRs; the repo was created 2019-03-27 (GitHub API). The contributors API paginates to roughly 360 entries including anonymous authors, so the contributor base is on that order.
The CNCF Incubating announcement reported its own figures at announcement time: 4,600+ stars, 2,400+ PRs, 300+ contributors, 19 maintainers, and 30+ company adopters (CNCF blog). KServe entered CNCF Incubation on 2025-09-29 with TOC sponsors Faseela K and Kevin Wang.
Ecosystem
KServe sits on top of and alongside several projects:
- Knative for the serverless (
Knative) deployment mode. - Istio for ingress and
VirtualService-based routing. - KEDA and the Kubernetes HPA for autoscaling.
- Kubeflow, the original parent, which still bundles KServe in its distribution.
- ModelMesh for high-density multi-model serving.
- Runtime backends: vLLM (generative AI), Hugging Face, NVIDIA Triton, and TorchServe, wired in as
ServingRuntimeresources.
Alternatives
KServe is the right pick when you want a Kubernetes-native CRD API, the Open Inference Protocol, and automatic provisioning of routing, autoscaling, and storage download. The others trade differently.
| Alternative | Differs by |
|---|---|
| Seldon Core | Inference graphs with ROUTER/COMBINER for multi-armed-bandit and ensembles; KServe centers on the isvc CRD and V2 protocol with auto-provisioned infrastructure |
| BentoML | Code-first packaging of any Python framework; often paired with KServe (package in BentoML, deploy on KServe) |
| NVIDIA Triton | A GPU-optimized server, not a competitor; KServe runs it as a ServingRuntime |
| Ray Serve / vLLM / MLflow / SageMaker / Vertex AI | Distributed serving, LLM engines, or cloud-managed platforms that KServe either integrates with or replaces on self-managed Kubernetes |