AWS (Amazon Web Services)

The wiki’s first hyperscaler — the high end of the spectrum the other pages defined against. Where hetzner-cloud sells one cheap box and render/railway/fly-io/ vercel sell a deploy workflow, AWS sells everything, à la carte (200+ services): you assemble the box, the network, the database, the queue, the CDN, each separately metered. It anchors the “raw IaaS” end well past hetzner-cloud in breadth and complexity (and, usually, per-resource cost).

The free tier — a credit funnel (revamped 2025)

AWS moved its free tier toward a credit model (oracle-cloud-free‘s closest analog, sharpened):

• Up to $200 in credits for new accounts — $100 on signup + up to $100 earned by exploring services.
• Free Plan: access to 90+ services for up to 6 months; the account auto-closes at 6 months unless converted to a paid plan.
• 30+ always-free services within monthly usage limits, on free and paid plans (the classic Lambda/DynamoDB always-free allowances persist, though this page states them only as “within monthly usage limits”).

So the same “free has a shape” mechanic the synthesis tracks applies — but AWS’s shape is a time-boxed credit burn-down (6-month expiry) rather than oracle-cloud-free‘s permanent Always-Free allotment or render‘s spin-down. It is an acquisition funnel into PAYG, like the others, just with the steepest post-funnel complexity.

Where it sits

AWS is the baseline the cost-model spectrum implicitly references: Tier-2 PaaS pitches (“no overprovisioning,” “no idle markup,” git-push deploys) are largely pitches against assembling raw AWS yourself. Its breadth is also why managed PaaS exists — render is Heroku-lineage precisely to hide this. Tracked as the anchor, not yet with neutral price-performance data (the standing sourcing caveat: vendor surface, not third-party benchmark).

A distinct compute shape — aws-lambda-microvms

Beyond the à-la-carte breadth, AWS now ships aws-lambda-microvms — Firecracker-based isolated VM sandboxes with a serverless model and, distinctively, state-preserving suspend/resume (up to 8h; suspended = no charge). It’s a new point on the wiki’s cost-model spectrum (per-second compute + warm scale-to-zero) and is pitched first at agent sandboxes for running LLM-generated code — the cross-spoke seam to agentic-tooling-wiki.

Related

hetzner-cloud · oracle-cloud-free · digitalocean · cloudflare · vercel · aws-lambda-microvms · synthesis