Cut Container Costs with Spot Instances and Kubernetes

If you run containers in production, you're probably paying too much. In 2025, the CNCF annual survey found that 82% of container users run Kubernetes in production (CNCF Annual Cloud Native Survey). That means most of you are also staring at a cloud bill that's climbing faster than your revenue. But here's the thing: you don't have to.

The Cost Problem

Picture this: you're the platform lead at a mid-sized SaaS company. You've got a Kubernetes cluster humming along, running your API, background workers, and a few ML inference jobs. You started with a few On-Demand instances, but as you scaled, your monthly cloud spend ballooned. You're now paying full price for compute that sits idle half the time. It's painful.

The Scenario

Imagine you're rolling out a new microservices architecture. You need to run 50 containerized services, each with modest CPU and memory needs. Your initial instinct is to pick a bunch of general-purpose instances and let them run 24/7. That's the default, but it's also the most expensive way to run containers, especially for workloads that are stateless and fault-tolerant. And a lot of containerized workloads are exactly that.

The Spot Option

That's where Spot Instances come in. AWS offers unused EC2 capacity at up to a 90% discount compared to On-Demand (AWS EC2 Spot). Azure has a similar offering with Spot Virtual Machines, and Google Cloud calls theirs Spot VMs, with up to 91% savings (Google Cloud Spot VMs). The catch? They can be interrupted. AWS gives you a two-minute warning, while Azure gives you 30 seconds (Azure Spot VMs). That sounds scary, but if your containers are designed to be resilient, it's a non-issue.

Kubernetes to the Rescue

Kubernetes is the perfect platform for this. It's designed to handle node failures gracefully. If a spot instance gets reclaimed, Kubernetes simply reschedules the pods elsewhere. You just need to ensure you have enough capacity in your cluster to absorb the loss. That's where a mix of spot and On-Demand (or even reserved) instances comes in. You run your critical, stateful services on On-Demand or reserved capacity, and your stateless, bursty workloads on spot.

A Concrete Example

Let's get specific. Say you need a cluster with 20 vCPUs and 80 GiB of memory for your batch processing jobs. On AWS, you could use m5.large instances (2 vCPU, 8 GiB) at $0.096/hour each (AWS EC2 Price List API). For 10 instances, that's $0.96/hour On-Demand. Now, if you switch to spot, you could save up to 90%, bringing it down to roughly $0.096/hour for the entire cluster. That's a massive win. But you have to be smart about it.

Mitigating the Risks

To make this work, you need to architect for interruption. Use pod disruption budgets, set up node affinity to spread pods across spot pools, and maintain a buffer of on-demand instances to absorb spikes. Tools like the Kubernetes Cluster Autoscaler can help you scale up spot nodes when they're available and scale down when they're not. Also, consider using spot for dev/test environments, which are inherently tolerant of interruptions (Azure Spot VMs).

My Recommendation

If you're running containers on a public cloud, you should be using spot instances for any workload that can tolerate interruption. That's not a radical idea; it's just good financial sense. The savings are too large to ignore, and the orchestration tools are mature enough to handle it. Start with a single non-critical service, measure the savings, and then expand. You'll thank yourself when you see the bill.

Sources

• AWS EC2 Spot - https://aws.amazon.com/ec2/spot/
• Azure Spot VMs - https://learn.microsoft.com/en-us/azure/virtual-machines/spot-vms
• Google Cloud Spot VMs - https://cloud.google.com/compute/docs/instances/spot
• CNCF Annual Cloud Native Survey - https://www.cncf.io/announcements/2026/01/20/kubernetes-established-as-the-de-facto-operating-system-for-ai-as-production-use-hits-82-in-2025-cncf-annual-cloud-native-survey/
• AWS EC2 Price List API - https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/current/us-east-1/index.json