// the take
CAPA is the official Cluster API infrastructure provider for AWS, letting you declare EC2-based self-managed clusters or EKS clusters as Kubernetes CRDs instead of hand-rolling Terraform or CloudFormation. It's for platform teams who've already bought into Cluster API and want one declarative control plane across clouds rather than AWS-specific tooling like eksctl or kops.
Bootstraps nodes without SSH, using cloud-init/Ignition through the bootstrap provider, which cuts off a whole class of access-key and open-port problems. clusterawsadm generates the exact IAM policies the controller needs instead of making you guess scopes by trial and error. Supporting both self-managed EC2 control planes and EKS under the same CRDs means you can move a fleet between the two without changing your GitOps layer. It's a k8s-sigs project with CodeQL scanning and an OpenSSF best-practices badge, so the security posture is taken seriously, not bolted on.
The API has gone through v1alpha4 to v1beta1 to v1beta2, each a breaking change, so upgrading a cluster you provisioned two years ago is its own project, not a version bump. Bootstrapping still leans on a separate CloudFormation stack for IAM/VPC groundwork alongside the Go reconcilers, so you're debugging two provisioning systems, not one, when something goes wrong. The required IAM policy set is broad (VPC, EC2, ELB, IAM role creation) and a misconfigured trust policy here is a real blast-radius risk, not a lint warning. Running this in production means understanding Cluster API's reconciliation model on top of AWS networking — there's no shortcut version for a team that just wants a cluster up this afternoon.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →