"What will hosting cost me?" is the first question every founder asks and the one with the most misleading answers online. Cloud bills scale with users, but not smoothly - they move in steps, and knowing where the steps are is the whole game.
Here's a stage-by-stage budget for a typical SaaS: a web app with user accounts, a database, and an AI feature, using real plan prices we track.
Stage 1: zero to 100 users - $0/mo
Genuinely free, not trial-free:
- Frontend: Vercel Hobby or Netlify free tier. Fine for thousands of visits.
- Backend: Render free web service. The catch: it spins down after 15 minutes idle, so the first request after a quiet spell takes ~30 seconds. Annoying, free.
- Database + auth: Supabase free tier - 500MB Postgres, 50K monthly active users of auth. This is the piece that saves you weeks.
- AI feature: Claude Haiku 4.5 at $1/M input tokens. A hundred users' worth of AI summaries costs about $2-3/mo, your only real bill.
Total: roughly $3/mo. The free stack is legitimate now; use it shamelessly.
Stage 2: 100 to 1,000 users - $35 to $60/mo
Two upgrades become non-optional around here: the backend can't spin down anymore, and the database needs room.
| Piece | Plan | Cost |
|---|---|---|
| Backend | Render Starter (always-on) | $7/mo |
| Database + auth | Supabase Pro | $25/mo |
| Frontend | Vercel Hobby (still fine) | $0 |
| AI usage | Haiku 4.5, ~5M tokens/mo | $5-15/mo |
Total: $37-47/mo. Cheaper variant: skip Render and put the backend on a Hetzner VPS for $4.50/mo - if you're comfortable managing a server. That's the recurring theme: every stage has a "pay with money" and a "pay with time" option.
Stage 3: 1,000 to 10,000 users - $100 to $250/mo
Real traffic. The backend needs more RAM, the database needs headroom, and you should be paying for the frontend tier that includes support and analytics.
| Piece | Plan | Cost |
|---|---|---|
| Backend | Render Standard, or a DigitalOcean 8GB Droplet | $25-48/mo |
| Database | Supabase Pro + compute add-on, or DO Managed Postgres | $25-60/mo |
| Frontend | Vercel Pro | $20/mo |
| AI usage | ~30-50M tokens/mo, mid-tier model | $30-90/mo |
Total: roughly $100-220/mo. For context, that's around 2 cents per user per month. If your product charges even $5/mo, infrastructure is under 1% of revenue - which is why "how do I cut my hosting bill" is usually the wrong question at this stage. The right question is whether the AI line item is on the correct model tier, because that's the line that balloons.
The two mistakes that actually cost money
- Paying for scale you don't have. Buying the $250/mo stack at 50 users because a blog post said you'd need it eventually. You can upgrade every piece above in an afternoon; buy the step you're on.
- Using a frontier model for a mid-tier job. The gap between Claude Opus and Claude Haiku is 5x on input and 5x on output. Most SaaS AI features (summaries, autocomplete, categorization) run fine on the cheap tier. This one line item swings more than every hosting decision combined.
Prices as tracked on June 16, 2026. Live numbers for every plan mentioned are on the Compute and LLM API marketplaces.
Want this table generated for your specific product? The AI stack planner asks six questions and builds the budget for your stage, not a hypothetical one.