San Francisco — June 23, 2026 — CloudPilot AI today brought its autonomous Kubernetes optimization to Microsoft Azure. Any team running Azure Kubernetes Service (AKS) can now point CloudPilot AI at a cluster and start cutting waste, surviving Spot evictions, and keeping workloads off the OOM line—through a read-only agent that installs in minutes, reads none of your application data, and never edits your cluster configuration. With Azure joining AWS and Google Cloud, CloudPilot AI now covers all three major clouds.

Provisioning is solved on AKS. Efficiency isn't.

This launch lands at a good moment for AKS. With Node Auto Provisioning (NAP)—Azure's managed, Karpenter-based autoscaler—clusters finally get node-group-free provisioning out of the box: a pod goes pending, the right VM size appears, and you've stopped hand-maintaining node pools.

That answers how nodes get created. It doesn't answer how well they run.

NAP will faithfully provision whatever your requests ask for—including the 60–70% of CPU and memory that most clusters reserve and never use. It won't tell you a JVM heap is sized wrong until the pod OOMs. It won't decide which workloads are safe to put on Spot, or how to hand a stateful service from one node to the next without a blip. That gap—between provisioned and optimized—is the layer CloudPilot AI runs on top of NAP (or on a classic cluster autoscaler setup, if you're not on NAP yet).

"Azure teams kept asking when CloudPilot AI would land on AKS," said Jingkang Jiang, co-founder and CEO of CloudPilot AI. "Now that node auto-provisioning is native to AKS, the missing piece is the intelligence above it—knowing what to right-size, what's safe on Spot, and how to replace a node without anyone noticing. That's the part we automate, and it's the same engine our customers already trust on AWS and Google Cloud."

Three things CloudPilot AI does on your AKS cluster

Rather than another dashboard to watch, CloudPilot AI is an automation layer that acts. It organizes around three outcomes:

1 — A smaller bill, without a babysitter

CloudPilot AI continuously measures what each workload actually uses and adjusts CPU and memory requests in real time, then consolidates pods onto fewer nodes so you stop paying for reserved-but-idle capacity. Java gets special treatment: standard rightsizers read container memory and miss the JVM underneath, so they either leave headroom on the table or trigger OOMs. CloudPilot AI is heap- and GC-aware—it right-sizes the heap with HeapDrift handling and JVM-args mutation—so Java services run lean without GC regressions. (Java workload optimization.)

2 — Spot savings you actually get to keep

Azure Spot Virtual Machines run up to 90% cheaper than pay-as-you-go, but Azure can evict them whenever it needs the capacity back—which is why most teams never trust Spot for anything that matters. CloudPilot AI makes Spot survivable instead of scary: it spreads workloads across multiple VM families and generations (Dsv5, Dasv5, AMD-based and Intel-based, plus memory-optimized Esv5 and compute-optimized Fsv2) and across availability zones, so no single capacity pool is a single point of failure. Replicated workloads are kept anti-affined across at least two nodes by default, and every node replacement is a graceful handoff—the new node is Ready and its Pods are Running before the old one is drained.

3 — Faster scale-ups and an honest audit trail

Big container images are the hidden tax on autoscaling: a node can be provisioned in seconds and still sit idle for minutes pulling an image. CloudPilot AI's Image Accelerator attacks this with lazy loading (start the container as soon as the needed layers arrive, not the whole image) and P2P distribution (pull layers from nearby nodes instead of a remote registry), so scale-ups and replacements keep pace with traffic. And because every action is logged—node create / delete / replace, each with a status and a reason—plus live dashboards for spend and events, you get an audit trail instead of scattered kubectl output.

Tuned to how Azure actually runs

CloudPilot AI isn't a generic optimizer pointed at Azure. It understands AKS specifics: it diversifies across Azure VM SKUs and zones the way the Azure capacity model rewards, treats Spot eviction as a first-class event rather than a failure, and sits cleanly alongside whatever you already run—NAP or cluster autoscaler, on-demand or Reservations and Savings Plans for your steady-state baseline, with Spot and rightsizing capturing the variable savings on top.

Connect your AKS cluster in minutes

Onboarding is a single script. In the CloudPilot AI console, choose AKS, then run the generated command from Azure Cloud Shell or any shell with kubectl access to the target cluster. Cluster metadata is discovered automatically from the AKS nodes and context, so there are no Azure IDs to copy in. The agent is read-only—it can't reach your sensitive data and makes no changes to cluster configuration—and CloudPilot AI begins analyzing immediately.

Visit cloudpilot.ai to connect your first AKS cluster, or book a demo.

About CloudPilot AI

CloudPilot AI delivers autoscaling Kubernetes for the most demanding teams—eliminating cloud waste, improving application performance, and reducing operational risk, all without manual tuning. Based in San Francisco and trusted by 100+ enterprises globally, CloudPilot AI has helped customers save more than $500M in cloud spend, with an average savings rate of 67%, while maintaining reliability at scale.