Pay-as-you-go

Pay-as-you-go bills you for what you actually consume: compute seconds, gigabytes transferred, requests served, storage held. There's no flat commitment, so a quiet month costs little and a busy one costs more. The hyperscalers run almost entirely on this model, and so do the developer PaaS platforms like Vercel, Netlify, and Railway.

The appeal is that you don't pay for idle capacity, which suits spiky or unpredictable workloads well. The danger is that the meter never fully stops, and success arrives as a bill. A traffic spike, a runaway loop, or a misconfigured job can turn a $20 month into a $2,000 one, and usage-based clouds are notorious for surprise invoices in exactly this way.

Flat-rate hosting is the opposite trade: predictable cost, capped resources. A DigitalOcean Droplet or a Hetzner VPS charges the same whether it's idle or slammed. For steady workloads that predictability is often worth more than the theoretical savings of paying only for what you use.