DEV Community

Alex Chen
Alex Chen

Posted on

I Replaced My $564/Year Hosting Bill with a $0 Free-Tier Stack (Vercel + Neon + Cloudflare)

Last month my hosting bill hit $47. For a side project with 200 users. That's insane.

So I spent a weekend rebuilding my entire stack on free tiers only. Three months later: $0/month, 99.98% uptime, faster page loads than before.

Here's the exact setup — and why I think paying for AWS as an indie dev in 2026 is mostly a bad habit.

The $47/month mistake

My old stack:

Service Cost/month What it did
AWS EC2 t3.small $15.20 ran a 400MB Node app
RDS db.t3.micro $13.50 Postgres with 800MB of data
CloudFront + S3 $9.80 static assets
Route53 + misc $8.50 DNS, data transfer surprises
Total $47.00 for 200 users 🤡

The EC2 instance averaged 3.2% CPU utilization. I was paying for a server that was asleep 97% of the time.

The $0 stack that replaced it

Layer Service Free tier limit My usage
Hosting Vercel 100GB bandwidth, 125K serverless invocations/day 4.1GB / ~3K invocations
Database Neon 0.5GB storage, 190 compute hours 340MB, autosleeps
CDN + DNS Cloudflare Unlimited bandwidth, free DDoS protection everything proxied
Auth Supabase Auth 50K MAU 200 MAU
Cron jobs Cloudflare Workers 100K requests/day 12 invocations/day
Total $0.00

The actual migration code

Moving Postgres from RDS to Neon took 20 minutes:

# Dump from RDS
pg_dump -h mydb.abc123.us-east-1.rds.amazonaws.com \
  -U postgres -Fc myapp > backup.dump

# Restore to Neon (connection string from dashboard)
pg_restore -h ep-cool-frog-123456.us-east-2.aws.neon.tech \
  -U alex -d myapp --no-owner backup.dump
Enter fullscreen mode Exit fullscreen mode

The Neon serverless driver made cold starts a non-issue:

import { neon } from '@neondatabase/serverless';

const sql = neon(process.env.DATABASE_URL!);

export async function getUsers() {
  // HTTP-based, no connection pooling needed on serverless
  return await sql`SELECT * FROM users LIMIT 50`;
}
Enter fullscreen mode Exit fullscreen mode

Cloudflare Worker replaced my $8/month cron service:

export default {
  async scheduled(event: ScheduledEvent, env: Env) {
    await fetch('https://myapp.vercel.app/api/cleanup', {
      headers: { Authorization: `Bearer ${env.CRON_SECRET}` },
    });
  },
};
Enter fullscreen mode Exit fullscreen mode 
# wrangler.toml — runs daily, costs $0
[triggers]
crons = ["0 3 * * *"]
Enter fullscreen mode Exit fullscreen mode

The results after 90 days

  • Cost: $564/year → $0/year
  • p95 latency: 340ms → 180ms (Cloudflare's edge is closer to my users than us-east-1 was)
  • Uptime: 99.98% (one 4-minute Vercel incident in May)
  • Deploy time: 8 minutes (git push + CI) → 40 seconds

The controversial part

I'll say it: most indie devs paying for AWS/GCP are doing it out of habit, not necessity. The free tiers of Vercel + Neon + Cloudflare + Supabase comfortably handle a side project up to ~50K monthly users. By the time you outgrow them, you'll have revenue to pay for infrastructure — or you'll know exactly which bottleneck to pay for, instead of renting a sleeping EC2 instance.

Yes, vendor lock-in is real. Yes, cold starts exist (Neon wakes in ~300ms, Vercel functions in ~50ms). Neither mattered for my app. They probably don't matter for yours either.

I'm now using the money I saved on better tooling — including a free AI coding setup (details here) that replaced my paid Copilot subscription.

What's in your free-tier stack? Anyone running production traffic on $0/month — where did it break first?

Top comments (0)