CoreWeave Joins CNCF as a Platinum Member

Kubernetes is central to CoreWeave’s cloud

CoreWeave said Kubernetes powers its cloud platform, from the control planes behind customer clusters to bare-metal GPU nodes, with no hypervisor in between. The company built CoreWeave Cloud on Kubernetes and uses it to run large AI models at production scale, according to the announcement.

Its CoreWeave Kubernetes Service, or CKS, is a managed Kubernetes environment designed for building, training, and deploying AI applications. Customers can use Slurm on Kubernetes, known as SUNK, to run training jobs and Kubernetes services from the same pool of GPUs.

CoreWeave said its Platinum membership will provide stronger support for the CNCF community, particularly in the AI-native cloud space. The commitment builds on work already carried out by its engineers, who contribute to Kubernetes, Kueue, and CloudNativePG.

Engineers contribute to Kubernetes and AI projects

Rita Zhang, a Principal Software Engineer at CoreWeave, is a Kubernetes Steering Committee member, SIG Auth co-chair, and one of the leads of the Working Group on AI Conformance. At the event, she is scheduled to present “Authorized? It Depends.” The session covers conditional authorization and how the API server proves its identity to admission webhooks.

CoreWeave also participates in the Certified Kubernetes AI Conformance Program. The company is a founding contributor to llm-d alongside Red Hat, IBM, Google, and NVIDIA. llm-d is now a CNCF Sandbox project.

Amy Chen, a Senior Software Engineer at CoreWeave, is presenting sessions on batch workloads and GPU scheduling. One focuses on how Kueue’s Topology Aware Scheduling accounts for NVLink domains and rack boundaries. Another covers updates to Job, Kueue, and JobSet, as well as plans for the working group through 2027.

What CoreWeave is showcasing for AI workloads

At booth 147, CoreWeave is demonstrating training and inference at scale, agent development, and CoreWeave Forge. Visitors can speak with software engineers, product managers, and cloud architects about batch scheduling, AI conformance, and GPU cluster design.

CoreWeave said CKS provides managed Kubernetes on bare-metal GPUs and integrates with Kueue for batch and training job queueing. Its inference options include Serverless Inference, Dedicated Inference, and inference on CKS, where users can bring runtimes such as vLLM, SGLang, llm-d, or NVIDIA Dynamo.

CoreWeave also offers Sandboxes, which provide an isolated environment for agents that run code, and Forge, which connects the build, train, and deploy workflow. The company said keynotes and breakout sessions are available through the CNCF YouTube page for people unable to attend in Salt Lake City.