Insights/Infrastructure
// infrastructure

Kubernetes is overkill for most startups

28 July 2026·6 min read

Kubernetes is an excellent tool. It is also the most over-prescribed technology in the startup world. Here is when you actually need it — and when you do not.

gradio infra --audit --team-size 4

→ Checking service count…

→ Measuring orchestration overhead…

⚠ 4 developers + Kubernetes = 1 FTE on ops

⚠ 200+ lines of YAML for a single deploy

✓ Docker Compose: 15 min setup, same result

✓ PaaS (Render/Fly.io): 5 min, zero ops

We are not anti-Kubernetes. It is a powerful orchestration platform that solves real problems at scale. The issue is that startups adopt it long before they have those problems, and the complexity kills their velocity.

What Kubernetes gives you

Kubernetes provides: declarative infrastructure, automatic scaling, self-healing, rolling deployments, service discovery, and load balancing. These are all valuable. The question is whether you need them now, and whether the cost is justified.

What Kubernetes costs you

Kubernetes requires: YAML manifests (hundreds of lines), an understanding of pods, services, ingress, configmaps, secrets, volumes, namespaces, RBAC, and network policies. A CI/CD pipeline that builds container images, pushes to a registry, and deploys via kubectl or Helm. Monitoring and logging configured for distributed containers. And someone on the team who understands Kubernetes deeply enough to debug it when it breaks.

For a 4-person team shipping a single application, this is weeks of setup and ongoing maintenance. It is a full-time job for one person — a person who is not building your product.

// You need Kubernetes when…
  • ✓10+ services scaling independently
  • ✓Dedicated platform/DevOps team exists
  • ✓Deployment frequency requires automation
  • ✓Cloud costs justify bin-packing savings

When you actually need Kubernetes

You need Kubernetes when: you have 10+ services that need to scale independently, you have a dedicated platform/DevOps team, your deployment frequency requires automated orchestration, or your cloud provider costs are high enough that bin-packing saves meaningful money.

Until then, simpler alternatives work: a single VPS with Docker Compose for early stage, a managed PaaS (Render, Railway, Fly.io) for growth stage, or AWS ECS / Google Cloud Run for mid-stage. These give you 90% of the benefits with 10% of the complexity.

The migration path

Start simple. When you outgrow it, migrate. The migration from Docker Compose to Kubernetes is well-documented and straightforward. The migration from "we set up Kubernetes too early and now everything is tangled" to a clean Kubernetes setup is not. Simplicity first, complexity when earned.

Simplicity first, complexity when earned. The migration from Docker Compose to Kubernetes is straightforward. The migration from premature Kubernetes to clean Kubernetes is not.

Keep reading

Want this level of analysis for your system?

We apply the same rigour to real client engagements. Architecture reviews, security audits, and stack recommendations — all fixed-price, all actionable.