Optimizing Kubernetes Infrastructure with Karpenter for Cost-Efficient Scaling

Case Study

Optimizing Kubernetes Infrastructure with Karpenter for Cost-Efficient Scaling

Services We Delivered

Devops

Industry

Automotive

The Client

A US-based automotive enterprise operating a high-traffic digital platform for used-car sales, connecting sellers with a nationwide network of buyers and managing the full transaction lifecycle from listing and pricing to financing and final sale. The platform runs on Kubernetes (Amazon EKS) and relies on dynamic node autoscaling to efficiently handle fluctuating traffic.

The Challenge

To address inefficiencies caused by running workloads on pre-defined node sizes (i.e., compute machines with fixed CPU and memory that do not adapt to actual demand), the client adopted Cast AI. While it delivered strong technical optimization, its commercial model began to introduce challenges beyond performance as the platform scaled.

Total costs increased due to the platform licensing fees and underlying EC2 infrastructure expenses
Licensing fees grew with each infrastructure addition, without a proportional increase in value.
Cast AI's pricing, features, and roadmap were not in the team's control.
The optimization logic operated as a black box, with no ability to inspect, audit, or customize it.
A single-vendor dependency for core node provisioning meant there was no community fallback if anything changed.

The Solution

The migration to Karpenter was a deliberate move toward a community-backed, fully owned infrastructure model. This is how the migration happened:

Cast AI audit: All active policies, instance preferences, and scaling behaviors were documented to create a clear migration checklist.
Karpenter setup: Installed on EKS with NodePools mirroring Cast AI's instance types, availability zones, and resource constraints.
Non-production validation: Provisioning behavior confirmed against Cast AI baselines in staging and dev before any production change.
PDB configuration: Pod Disruption Budgets are set for all critical services to ensure zero disruption during node consolidation events.
Production cutover: Karpenter progressively took over provisioning in a phased rollout. Cast AI was then fully disabled, and the subscription was canceled.
Observability setup: SigNoz dashboards were updated with Karpenter-native metrics for full real-time infrastructure visibility.

Results

Karpenter matched Cast AI's capabilities in full while eliminating the licensing overhead, vendor dependency, and opacity that had prompted the move.

Total annual infrastructure cost reduced from approximately $20,000 (platform + EC2) to around $12,000 (EC2 only)
Full feature parity, right-sizing, scale-down, and spot support all matched.
Vendor dependency removed, reducing operational risk across the infrastructure.
Long-term confidence secured through CNCF community backing and a public roadmap.

Paying licensing fees for Kubernetes optimization?

If your team is running a commercial Kubernetes tool and wondering whether open-source can match it, we can help you evaluate, migrate, and validate without losing any capabilities.