Lucface
← The log
2026-04-30

I run 4 production sites on Vercel and Neon free tiers

Total monthly infra cost for 4 production sites: about $0. Here's where the free tier ends.

#vercel#neon#free-tier#infra-cost

The setup

Four production sites run on Vercel + Neon HTTP serverless on the bluebarn team:

  • struvo (the product)
  • sd-ai-studio (community brand)
  • twentyfive (CRM)
  • lucface (this site)

Total monthly infra cost across all four: about $0.

Where the free tier holds

Vercel Hobby:

  • 100 GB/mo bandwidth (we use ~3 GB across all 4)
  • 100K function invocations (we use ~12K)
  • Unlimited static deploys
  • 1 concurrent build (we serialize, never blocks)

Neon Free:

  • 0.5 GB storage per project (we use 80 MB on the largest)
  • 191 compute hours/mo per project, scaled to zero on idle
  • 10 branches per project (we use 2-3 per project)

Compute hours scaled to zero is the big unlock. Neon spins down on inactivity. We pay for nothing during the 20+ hours/day no one's hitting these sites.

Where the free tier breaks

Three things would force a paid upgrade:

  1. Compute time. If a Struvo customer kicks off a heavy aggregation that runs > 191 hours/month, we hit the wall.
  2. Storage. When the LLM Wiki + transcripts + photos pile up, we'll cross 0.5 GB.
  3. Bandwidth. A viral YouTube video pointing to a Lucface post could chew through 100 GB if the OG image isn't aggressively cached.

Each of these has a fix that's free until the day it isn't.

The migration plan

When we cross the line, the upgrade math is:

  • Vercel Pro: $20/seat/mo. Worth it the moment we have a customer who needs uptime guarantees.
  • Neon Scale: starts at $19/mo. Worth it the moment we have customer data we can't lose.

Both are sub-$50/mo for the first paid tier. That's the kind of cost we'd happily absorb on the first paying customer.

What you can steal

If you're shipping side projects:

  1. Default to Vercel + Neon HTTP. Cold starts are fine for most workloads.
  2. Track bandwidth, function invocations, and compute hours weekly. The dashboards make this easy.
  3. Don't pre-pay for Pro until a real workload demands it.

The discipline isn't about saving money. It's about not adding infrastructure cost before it earns its slot. Same as the V1 product cuts — only ship what's required.