Here is the complete infrastructure bill for running a planetary simulation with a 3D viewer, AI intelligence layer, daily automated advancement, and static hosting:

$0.00/month.

No cloud provider. No Kubernetes. No database server. No Redis. No message queue. No CDN subscription.

Here’s the stack:

Component Service Cost
Compute (simulation) GitHub Actions (cron) Free
Database JSON file in git Free
API GitHub raw content Free
Static hosting GitHub Pages Free
AI inference Client-side (101KB weights) Free
CI/CD GitHub Actions Free
Monitoring Git log + commit messages Free

Compute: A GitHub Action runs once per day. It executes a Python script, advances the simulation one step, and commits the result. Total runtime: ~5 seconds.

Database: The state file is committed to the repo. Reading is a raw GitHub fetch. Writing is a git commit. History is git log. There is no schema migration because JSON is schemaless.

API: Clients fetch the raw file from GitHub. This is a CDN-backed, globally distributed read API with caching headers. You didn’t build it. You didn’t deploy it. It’s just there.

AI: The model weights are a 101KB JSON file served as a static asset. Inference runs in the browser. The user’s CPU is the compute.

The catch: Write throughput is limited to git commit speed. There’s no real-time push. Multi-writer conflicts require merge strategies. You can’t run SQL queries against a JSON file.

When it’s appropriate: Systems with low write frequency and high read frequency. Time-stepped simulations. Personal dashboards. Status pages. Any system where “updated once a day” is fast enough.

The point isn’t that this replaces cloud infrastructure. The point is that most projects don’t need cloud infrastructure and assume they do because nobody showed them the alternative.

Your side project doesn’t need a database server. Your simulation doesn’t need AWS. It needs a JSON file and a cron job.