Provision and scale a CoreWeave Kubernetes Service GPU node pool and obtain a kubeconfig

Documented steps

1. Create a CoreWeave API access token in the console, then call the control plane at https://api.coreweave.com with header 'Authorization: Bearer <API_ACCESS_TOKEN>'.
2. Cluster lifecycle uses the CKS REST surface: GET /v1beta1/cks/clusters to list, POST /v1beta1/cks/clusters to create, GET /v1beta1/cks/clusters/{id} to poll, PATCH /v1beta1/cks/clusters/{id} to update, DELETE /v1beta1/cks/clusters/{id} to delete.
3. Obtain a kubeconfig by creating an access token scoped to the target cluster in the console; the managed-auth kubeconfig is generated and displayed at that moment only. Save it and export KUBECONFIG to point at it.
4. Node pools are NOT created through api.coreweave.com. They are a Kubernetes custom resource applied to the cluster's own API server with kubectl using the kubeconfig from the previous step.
5. Author a NodePool manifest with apiVersion compute.coreweave.com/v1alpha1. The required fields are instanceType (the CoreWeave GPU instance type, for example gd-8xh100ib-i128) and targetNodes (desired node count).
6. Optional fields include autoscaling: true with minNodes and maxNodes, nodeLabels, nodeTaints, and lifecycle.scaleDownStrategy (IdleOnly or PreferIdle) to control which nodes are removed on scale-down.
7. Apply it: kubectl apply -f gpu-nodepool.yaml.
8. Scale it by editing targetNodes and re-applying, or with kubectl scale nodepool <name> --replicas <N>.
9. Check state with kubectl get nodepool <name> (target/current/in-progress/queued counts) or kubectl describe nodepool <name> to read Conditions - Validated, Capacity, Quota, and AtTarget.
10. Verify nodes joined with kubectl get nodes -l <your nodeLabel>, then schedule GPU workloads with resource requests for nvidia.com/gpu.
11. Official documentation: https://docs.coreweave.com/products/cks/reference/node-pool | https://docs.coreweave.com/products/cks/reference/cks-api | https://docs.coreweave.com/support/cks/articles/how-do-i-download-a-kubeconfig-file-to-access-my-cluster