AWS is previewing a feature that adjusts CPU and memory requests for running Amazon EKS Auto Mode workloads based on their actual usage. Teams can inspect recommendations before applying changes, then resize Pods in place rather than rebuilding them, giving operators a practical way to tackle over-provisioning without hand-tuning every workload.
AWS AI Watch analysis
What happened
AWS says resource optimisation is available on EKS Auto Mode clusters running Kubernetes 1.36 or later in seven regions, including Ireland and Frankfurt. It is available at no additional charge. Operators create a ClusterResourceOptimizationPolicy, review recommendations in Recommend mode, and switch to Apply mode when ready to make changes.
The feature can keep memory requests above a Java heap and raise them after an out-of-memory kill. It does not change the CPU or memory signals used by Horizontal Pod Autoscaler or KEDA, and workloads already managed by Vertical Pod Autoscaler can be excluded. AWS’s feature announcement has the full region list and setup details.
Why it matters
CPU and memory requests help determine which EC2 instances EKS provisions. Requests left untouched after deployment can mean running the same workloads on larger or more instances than their observed usage calls for. AWS says this feature can rightsize them without recreating Pods, while leaving autoscaler signals alone.
That makes the preview relevant to teams looking to reduce wasted capacity without editing deployment templates by hand. The practical question is whether the recommendations fit the quirks of real workloads, especially those with memory needs that change sharply rather than politely following an average.
Our read
This is a useful operator feature, not a magic cost-cutting button: AWS provides the recommendations and the Apply switch, but teams still need to decide whether each change is safe for their workload. Start in Recommend mode, check what it proposes, and be particularly attentive to memory-sensitive services. Kubernetes housekeeping has never lacked knobs; a pub-carpet calamity that measures first is a welcome change.
What to watch
- Whether teams see meaningful reductions in instance use after applying recommendations.
- How the feature handles workloads with uneven or sudden resource spikes.
- Whether AWS expands the preview’s regional availability.
Discussion spark: Would you trust resource recommendations enough to apply them automatically, or should Kubernetes rightsizing always require an operator’s review?
Sources and evidence
- Amazon EKS Auto Mode now rightsizes Pods to reduce compute costs (Preview) (17 November 2026, 08:00 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)