The Startup That Wasted 6 Months on K8s

Kubernetes is Overkill for 80% of Startups Architecture

I consulted for a SaaS startup in 2024. 8 engineers. 3 microservices. 2,000 daily active users. They had spent 6 months setting up and maintaining a self-managed EKS cluster. Their lead engineer was spending 30% of his time on Kubernetes issues — nodes not draining, pods in CrashLoopBackOff, Ingress controllers misbehaving. Their competitors were shipping features. They were fighting infrastructure.

The Hidden Costs of Kubernetes for Small Teams

  • EKS control plane: $73/month before a single workload runs
  • Minimum viable cluster: 3 nodes (t3.medium) = ~$100/month extra
  • Engineering overhead: Cluster upgrades every 3–4 months, each taking 1–2 days
  • Cognitive load: Every new engineer needs K8s onboarding — YAML, kubectl, concepts
  • Debugging complexity: A simple "why isn't my app starting?" can take hours in K8s vs minutes with Docker

What They Should Have Used (And Eventually Did)

We migrated them to: ECS Fargate for container orchestration (serverless — no node management), RDS for their database, CloudFront + S3 for static assets, GitHub Actions for CI/CD. Cost went from ₹1.8L/month to ₹65K/month. Engineering overhead on infrastructure went to near zero. Feature velocity tripled.

The K8s Readiness Checklist

Kubernetes makes sense when you can check most of these:

  • ✅ 10+ services that need independent scaling
  • ✅ Dedicated DevOps/Platform team (at least 1 FTE just for K8s)
  • ✅ Traffic patterns that require HPA (horizontal pod autoscaling)
  • ✅ Multi-tenancy requirements (namespaces + RBAC)
  • ✅ Need for complex deployment strategies (canary, blue-green at service level)
  • ✅ 50+ developers who benefit from a standardized platform
⚠️ The uncomfortable truth: Kubernetes is amazing engineering. But engineering excellence and business effectiveness are not always the same thing. Choose the tool that moves your business forward — not the tool that makes your resume look impressive.