Back to Blog
Does Your Startup Need Kubernetes? Usually Not — Here's the 4-Question Test

Does Your Startup Need Kubernetes? Usually Not — Here's the 4-Question Test

Dennis Reinkober2 min read
TL;DR

If you have one product, one team, and under ~50k users, you don't need Kubernetes. A Hetzner VM with Docker Compose and a deploy script handles that stage for €20–€100/month and near-zero ops overhead. Kubernetes earns its complexity when you have multiple teams, spiky scaling, or isolation requirements — not before.

Kubernetes is great technology. It's also the most common piece of premature infrastructure we remove from startup stacks. The pattern is always the same: someone set it up "so we're ready to scale," and now a two-person team spends a day a week feeding the cluster.

The 4-question test

Answer honestly. Two or more "yes" answers → Kubernetes is worth discussing. Zero or one → it isn't.

  1. Do multiple teams deploy independently and step on each other without isolation?
  2. Is your load genuinely spiky — 10x differences between quiet and peak hours that you must autoscale through?
  3. Does compliance or your enterprise customers require workload isolation, network policies, or multi-tenancy guarantees?
  4. Do you already run 10+ services that need orchestrated rollouts, service discovery, and self-healing?

Most pre-Series-A startups answer no to all four.

What to run instead

Boring, proven, cheap:

  • One VM (or two behind a load balancer) with Docker Compose. Handles tens of thousands of users for most web products. We've written about the economics in Hetzner vs. Vercel.
  • A modular monolith instead of microservices, so there's less to orchestrate in the first place — the reasoning is in our monolith post.
  • A managed platform (Cloud Run, App Runner) when you'd rather pay per request than manage a box.

What Kubernetes actually costs a small team

The cluster fee is the cheap part. The real bill: upgrades every few months, YAML sprawl, ingress debugging, secrets management, and the hiring constraint — every engineer you add now needs K8s literacy. Realistic overhead for a small team: 4–8 hours a week. That's a fifth of an engineer maintaining infrastructure your product doesn't need yet.

When it is the right call

Multiple product teams, real multi-tenancy, enterprise compliance requirements, or steady 10x traffic swings — at that point Kubernetes is the standard answer for good reasons, and migrating a well-built Compose setup onto it takes weeks, not months. Deferring the decision costs you almost nothing.

Not sure which side of the line you're on? We audit startup infrastructure and give you a straight answer — see our Cloud & DevOps services.

Similar Posts