Cast AI for Karpenter is a suite of optimization features designed to work alongside open-source Karpenter. Rather than replacing Karpenter, Cast AI for Karpenter extends it with capabilities that help you capture additional savings, improve operational stability, and gain visibility into cluster efficiency.

Why Cast AI for Karpenter

Open-source Karpenter handles Pod scheduling and Node provisioning effectively. However, organizations running Karpenter at scale often encounter gaps that the open-source project doesn't address:

Cost visibility gaps
Karpenter provisions Nodes based on Pod requirements, but has no awareness of your organization's Reserved Instances, Savings Plans, or negotiated discounts. This can lead to underutilized commitments and missed savings opportunities.

Consolidation limitations
Karpenter's built-in consolidation works at the Node level but doesn't coordinate with workload rightsizing or account for the actual resource utilization patterns of your applications.

Spot reliability concerns
While Karpenter supports Spot Instances, it treats all Spot pools equally and relies on the standard two-minute interruption warning. Teams often limit Spot adoption due to unpredictable interruptions.

Operational overhead
After Spot fallback to On-Demand, clusters can accumulate expensive Nodes that remain in place long after Spot capacity becomes available again. This requires manual intervention to recover cost savings.

Cast AI for Karpenter addresses these gaps by layering Cast AI's optimization capabilities on top of your existing Karpenter deployment.

How Cast AI for Karpenter works

Cast AI for Karpenter uses a non-intrusive approach that preserves your existing Karpenter configuration while adding enterprise capabilities through Cast AI components.

Architecture

When you connect a Karpenter-managed cluster to Cast AI, the platform deploys Castware components alongside your existing Karpenter installation:

• Cast AI Agent collects cluster metrics and workload data
• Cluster Controller coordinates optimization actions with Karpenter
• Evictor handles intelligent workload consolidation
• Spot Handler manages Spot Instance lifecycle and fallback recovery
• Pod Mutator automates Pod spec adjustments for optimal placement
• Workload Autoscaler handles continuous rightsizing

These components work with Karpenter's existing CRDs and node provisioning logic. Cast AI guides Karpenter's decisions through CRD modifications and policy adjustments rather than replacing its core functionality.

CRD-first approach

Karpenter users expect to manage configuration through Kubernetes Custom Resource Definitions. Cast AI for Karpenter maintains this pattern—Cast AI modifies Karpenter CRDs to influence provisioning decisions while keeping configuration in the Kubernetes-native format you're already using.

For example, when Cast AI identifies under-utilized Reserved Instances in a specific instance family, it can adjust your NodePool CRDs to prioritize that family for new capacity. Once those commitments are fully utilized, the configuration shifts to prefer the most cost-effective alternatives available.

Concept mapping

If you're familiar with Cast AI's standard Autoscaler, the following table shows how Karpenter concepts map to Cast AI equivalents:

With Cast AI for Karpenter, you continue using NodePools and EC2NodeClasses. Cast AI reads and, when optimization features are enabled, modifies these CRDs to steer provisioning decisions—but your existing resources remain the source of truth.

If you later migrate to Cast AI Autoscaler, your NodePools will be replaced by Node Configurations, and your EC2NodeClasses will be replaced by Node Templates. For details, see Migration from Karpenter.

Key capabilities

Savings analysis and reporting

Before you enable any optimization, Cast AI analyzes your Karpenter-managed cluster and generates a savings report showing specific optimization opportunities.

This report shows the difference between what Karpenter delivers and what additional optimization is possible with Cast AI. You can use the free reporting tier indefinitely without enabling active optimization.

Consolidation with minimal disruption

Cast AI's Evictor works with Karpenter's consolidation to improve Node utilization while minimizing workload disruption. Unlike standard eviction, this integration:

• Coordinates with Workload Autoscaler to account for actual resource usage, not just requests
• Respects Pod Disruption Budgets and workload constraints
• Uses Container Live Migration (when available) to move workloads without restarts
• Consolidates Nodes progressively rather than aggressively

The result is tighter bin-packing with reduced disruption to running applications.

Cluster rebalancing

While Karpenter optimizes at the Node level, Cast AI's Rebalancer operates across the entire cluster to identify optimization opportunities that span multiple NodePools:

• Identifies Nodes that could be replaced with more cost-effective alternatives
• Coordinates replacements to maintain workload stability
• Accounts for constraints like availability zones, architecture requirements, and capacity types
• Works alongside Karpenter's provisioning rather than competing with it

Spot intelligence

Cast AI enhances Karpenter's Spot Instance handling with two key capabilities:

Spot reliability model
Cast AI maintains global telemetry across connected clusters to identify which Spot pools have historically stable capacity. By steering Karpenter toward reliable pools, interruptions can be reduced substantially compared to random pool selection.

Interruption prediction
Beyond AWS's two-minute warning, Cast AI's prediction model identifies instability patterns before interruptions are announced. This provides additional lead time to gracefully migrate workloads off at-risk Nodes.

Spot fallback recovery

When Spot capacity is unavailable, Karpenter falls back to On-Demand instances. The problem is that these fallback Nodes often remain in place long after Spot capacity returns, accumulating unnecessary costs.

Cast AI's Spot Handler monitors market conditions and automatically recovers Spot capacity when it becomes available. This bidirectional fallback ensures clusters don't get stuck on expensive On-Demand Nodes.

Workload rightsizing

Most Kubernetes workloads are over-provisioned. Requests are set conservatively, leaving clusters with significant waste that Karpenter can't address because it provisions based on what Pods request, not what they actually use.

Workload Autoscaler continuously learns actual resource utilization and adjusts requests to match reality. This tighter fit enables better bin-packing and reduces the total compute capacity your cluster requires.

Workload Autoscaler integrates with Evictor and Rebalancer so that rightsizing decisions flow through to Node-level optimization automatically.

Pod mutations

For teams onboarding existing clusters to Cast AI, Pod mutations automate the configuration changes needed to take advantage of optimization features. Rather than updating every Deployment manifest, you define mutation rules that automatically apply labels, tolerations, and NodeSelectors to matching Pods.

For teams that have already onboarded and seek to take advantage of Spot Instances, Pod mutations allow for proportionally distributing Pods across On-Demand and Spot Instances automatically based on assigned labels.

What Cast AI for Karpenter doesn't change

Cast AI for Karpenter is designed to be non-intrusive:

• Your NodePools remain yours – Cast AI works with your existing NodePool definitions
• Karpenter continues provisioning – Node creation and deletion flow through Karpenter
• Rollback is straightforward – Removing Cast AI components returns your cluster to its original Karpenter-only state
• No vendor lock-in – Your Karpenter configuration remains portable

Next steps

Ready to see what optimization is available for your Karpenter cluster?

Get started with Cast AI for Karpenter – Connect your cluster and view your optimization potential

Related resources

• Cast AI for Karpenter features – Feature details and Karpenter integration
• Migration from Karpenter to Cast AI – For teams considering full Cast AI Autoscaler adoption