Everyone asks which hosting is cheapest and gets a number back. Two dollars. Seven dollars. Twenty dollars.
That number is the least useful thing about hosting.
The real cost shows up on the day something breaks, and you find out what you can't do. No shell. No logs. No way to roll back.
I've seen a site stay down for fourteen months on hosting that cost about $4/month. It was, technically, very cheap.
The three tiers, with the useless numbers first
| Monthly | What it is | |
|---|---|---|
| Shared cPanel | $2-10 | Disk, MySQL, PHP, a control panel |
| Render | Free, or ~$7 | Managed deploys, logs, rollback |
| AWS (App Runner + RDS) | ~$20-40 | Real infrastructure, real complexity |
Those numbers are correct and they explain almost nothing.
Shared cPanel: genuinely the right answer, sometimes
For a WordPress install, a static site or a brochure site, shared hosting at a few dollars a month is the correct choice. Don't let anyone shame you out of it.
What you're giving up is access:
- Shell access is often disabled entirely - no SSH, no running a command
- No real application logs, just a generic error page
- No deploy pipeline. You upload files.
- No rollback
- Some panels block you from uploading a zip containing JavaScript
That last one sounds trivial until it's 11pm and you're trying to deploy a build.
What that actually costs when it goes wrong
A legacy PHP app on shared hosting starts returning a truncated response. Fast first byte, so the uptime monitor reports everything fine - then the connection just stops partway through the page. Every page that loads one particular include dies at the same byte count.
On a server with shell access that's a twenty-minute job: tail the error log, find the fatal, fix it.
With shell disabled and no log access, you're guessing. And because the monitor said "up", nobody escalated it.
Fourteen months.
The hosting didn't fail. The access failed. That's the cost nobody puts on the pricing page.
Render: the $7 that buys the most
The free tier spins down after 15 minutes of no traffic, with roughly a 50-second cold start. You get 750 instance hours a month - enough for exactly one always-on service. The free Postgres deletes itself after 30 days.
Starter is about $7/month and the sleeping stops.
What that $7 buys is precisely what cPanel won't give you: logs, a deploy pipeline, rollback, environment variables, a health check. For a small API that's the best value on this list, and it isn't close.
AWS: and the free tier change nobody updated
A realistic small deployment is App Runner for the service and RDS Postgres for the data:
- RDS
db.t4g.micro- about $12-15/month - App Runner - $5-25/month depending on traffic
- Realistically $20-40/month at low traffic
But the number everyone quotes is out of date:
Accounts created before 15 July 2025 keep the legacy 12-month free tier.
Accounts created after that get 6 months plus up to $200 in credits.
Almost every AWS tutorial still says "free for twelve months". For a new account it isn't. Plan for six.
https://aws.amazon.com/free
And the honest point: AWS isn't expensive at this size - it's complicated. IAM users, security groups, a region choice, ECR, a deploy role, DNS validation. The cost is measured in hours, not dollars, and those hours are all front-loaded.
The costs that never appear on a pricing page
| cPanel | Render | AWS | |
|---|---|---|---|
| Time to first deploy | Minutes | Minutes | A day |
| Time to debug at 2am | Hours, or never | Minutes | Minutes |
| Rollback | None | One click | Redeploy previous image |
| Scaling | Buy a bigger plan | Change a dropdown | Automatic |
| Lock-in | Low | Low | High |
What I'd actually pick
WordPress, static sites, brochure sites -> shared cPanel. Correct answer, don't overthink it.
A side project or prototype -> Render free, with Neon for the database.
A real app with real users -> Render Starter at $7. Best money-per-problem-solved on this list.
Compliance, scale, or a team -> AWS, and budget the hours rather than the dollars.
The one rule
Choose the cheapest tier that still lets you see what went wrong.
That single rule would have saved fourteen months.
Figures checked November 2026 against the vendors' own pages - they change often, so check the date before planning around any of it.
Top comments (0)