CASE 73 · GLADE · 2026
Twelve EKS clusters, one application surface.
An AI infrastructure company ran 12 EKS clusters across regions and GPU classes. Deploying anything across the whole estate was 12 separate kubectl applies plus a spreadsheet to track them. We adopted Karmada for cluster federation and turned cross-cluster operations into a single declaration.
AI infrastructure
PLATFORM
2026
RESULTS
What changed, by the numbers.
CROSS-CLUSTER DEPLOY TIME
−83%
CLUSTERS FEDERATED
12
DRIFT INCIDENTS
−100%
NEW-CLUSTER ONBOARDING
2h
HOW IT WENT
The cross-cluster spreadsheet was the smoking gun. Each cluster had drifted slightly from the others — different versions of a CRD here, a missed NetworkPolicy there. Deploys involved checking the spreadsheet, then apologising when a deploy failed because cluster #7 still had v1beta1.
Karmada became the federation control plane. ArgoCD synced application manifests to Karmada; Karmada propagated them to the member clusters with override policies for cluster-specific config (GPU type, region-specific secrets). Cilium Cluster Mesh handled cross-cluster networking for the few cross-cluster service calls.
Drift incidents dropped to zero — the single source of truth made it impossible for clusters to differ unintentionally. A cross-cluster deploy that used to be 90 minutes of careful coordination now takes 15 minutes and is initiated by a single git push.
RELATED · SAME DOMAIN
Other engagements in this space.
READY WHEN YOU ARE
Let's get your AWS bill (and architecture) in order.
The discovery call is free. You walk away with at least one concrete idea — even if we never work together.