The self-hosting comeback: why developers are leaving PaaS (and when they shouldn't)

A $5 VPS in 2026 is a serious machine: multiple cores, several gigabytes of RAM, NVMe storage. Meanwhile, a decade of platform pricing changes — free tiers shrinking, usage-based bills spiking with traffic — has pushed a visible cohort of developers back toward running their own infrastructure. The difference this time is tooling.

The new self-hosted stack is actually good

The old objection to self-hosting was never the server cost — it was the operational misery. That's what changed:

• Coolify turns any VPS into a push-to-deploy platform: git integration, automatic SSL, preview environments, one-click databases. It's the closest thing yet to "Heroku, but it's your box."
• Gitea gives you GitHub-shaped code hosting — repos, PRs, CI with Actions-compatible syntax — in a single lightweight binary that runs happily on a Raspberry Pi.
• Uptime Kuma replaces a paid uptime service with one Docker container and a genuinely pleasant UI.
• Add PostgreSQL, Grafana, and Prometheus, and you have a stack that would have required a platform team ten years ago.

Docker Compose matured, ARM servers got cheap, and a large community now maintains deployment recipes for all of it. The activation energy genuinely dropped.

The honest math

Self-hosting wins clearly when:

• Your workload is steady. A predictable app on a fixed-size server is the best case — platform pricing is built around bursty workloads and charges a premium for elasticity you may not need.
• Egress or storage dominates your bill. Serving media or large artifacts through usage-priced platforms is where invoices go exponential.
• You're cost-sensitive and time-rich. Side projects, internal tools, homelabs: the hours are the hobby.

The platforms win clearly when:

• An outage at 3 a.m. is your problem alone. Vercel and Fly.io employ people to wake up for you. On your VPS, unattended-upgrades and a dead disk are your pager.
• Preview deployments drive your team's workflow. Reproducing per-PR environments, DDoS protection, and global edge delivery yourself is real engineering, not configuration.
• Compliance is in the room. SOC 2 on self-hosted infrastructure means you're the one being audited.

The quiet middle path is underrated: a managed platform for the user-facing app, self-hosted services for everything internal — monitoring, git, analytics, job runners. Most of the savings, a fraction of the risk.

If you're going to do it

Three rules from the people who do this well. Automate from day one — if your server can't be rebuilt from a script and a backup, you have a time bomb, not infrastructure. Test the restore, not the backup — a backup you've never restored is a hope. Keep the blast radius small — one compromised WordPress plugin on the same box as your database is how homelab stories end up on the front page of Hacker News.

Self-hosting in 2026 is a legitimate engineering choice, not a statement of ideology. Run the numbers for your actual workload, be honest about whose pager rings, and let that decide.