Approaches Overview
Seven ways to run your services in 2026 — and how to pick.
Almost every operational setup is a variation on one of these. Pick the smallest one that meets your actual needs.
The shortlist
| Approach | Effort | Cost | Scales to | Pitch |
|---|---|---|---|---|
| Single VPS | Low | $5–50/mo | Tens of users | One box. SSH in. Ship. |
| Self-host fleet (no k8s) | Medium | $50–500/mo | Thousands | A few VPSes wired up with Ansible / Kamal / SSH. |
| Self-host PaaS | Low–Med | $20–200/mo | Thousands | Coolify / Dokku — git push to your own server. |
| Kubernetes | High | $$$$ | Anything | The standard for multi-team / multi-service platforms. |
| Managed cloud | Low–Med | $$$ | Anything | Push the ops to AWS / GCP / Azure / Vercel. |
| Homelab | Medium | $0–$$ (hardware) | Hobbyist | Mini-PCs, NAS, Proxmox, Pi-hole. |
| Serverless-only | Low | Pay-per-use | Anything | Workers / Lambda / Functions, no servers. |
How they actually compare in 2026
- Most products under ~$10M ARR thrive on either Self-host PaaS (Coolify / Kamal) or Managed cloud (Render / Fly / Railway / Vercel + a managed Postgres). These are the boring-but-correct answers.
- Kubernetes has matured but lost some default status. Use it when you have multiple teams shipping to one platform, or when you genuinely need its primitives. Otherwise the ops tax doesn't pay off.
- Cloudflare Workers + Pages + Queues + R2 + D1 is now a complete stack for many apps. Cheap, global, no servers to babysit.
- Self-host fleet still wins on cost and clarity for ops-fluent teams. Kamal 2 made it dramatically easier than 2022-era setups.
- Homelab is bigger than ever — Proxmox + Tailscale + a Beelink mini-PC will run more than most teams need.
- Single VPS is underrated for solo founders. A single $20 Hetzner box on Coolify runs a multi-user SaaS just fine until you out-grow it.
Decision factors
- Team size. One person → single VPS or PaaS. Three+ devs → fleet or managed. Many teams → Kubernetes or full managed cloud.
- Stateful workloads. A real Postgres / Redis / Elasticsearch makes a single-VPS or PaaS approach harder; managed cloud or k8s with operators wins.
- Compliance. SOC 2 / HIPAA / FedRAMP push you toward managed cloud or a properly-built k8s + GitOps setup.
- Geography. Global low-latency → Cloudflare / Vercel-edge / Fly. Regional → anywhere.
- Budget. Self-host is cheap up front but expensive in time. Managed cloud is the opposite.
- Failure tolerance. "We can be down for an hour while I fix it" → simpler. "Customers stop paying if we're down 30 seconds" → managed + multi-region.
When to change
You'll know it's time to evolve when:
- Single VPS → fleet: you can't deploy without a maintenance window.
- Fleet → k8s / managed: you're inventing your own scheduler / secret store / service mesh.
- Managed cloud → self-host: the bill is crushing you and you have ops capacity.
- Anything → serverless: your traffic is bursty and you're paying for idle capacity.
What's worth dabbling with
- Kamal 2 — even if you don't switch, the deploy model is clarifying.
- Tailscale + a single VPS — the pleasant-est ops experience available in 2026.
- k3s on a $20 box — feel Kubernetes without the cluster cost.
- Coolify on Hetzner — five minutes from "blank server" to "shipping apps."
- Cloudflare Tunnel — expose any local service to the internet without opening ports.