Provisioned capacity vs what you actually use
Render's tiers buy a permanently running instance — cost tracks capacity. Railway meters actual vCPU/memory usage instead (with a $5 minimum); Vercel and Netlify bill request-shaped execution against included credits. Spiky or idle-leaning workloads are the classic case where that difference decides the bill.
Evidence: [1] Render Pricing [2] Railway pricing [3] Vercel Pricing [4] Netlify pricing