Infrastructure as an API call
The defining shift is not the technology — a cloud instance is essentially a VPS — but the model around it. Classic hosting rents you a fixed thing per month. Cloud platforms let a program create a server in seconds, resize it, clone it into ten, and delete it, paying only for the minutes it existed. Infrastructure stops being a purchase and becomes a variable you can script.
That elasticity is why the model won: capacity can follow demand instead of being guessed in advance. But the meter cuts both ways. A fixed-price server that is too small gets slow; a cloud account that is misconfigured gets expensive — surprise bills from forgotten instances, egress traffic or runaway autoscaling are the cloud's version of downtime. Cost management is a real skill on these platforms, not an afterthought.
- Hyperscalers (AWS, Google Cloud, Azure) — hundreds of services, global regions, enterprise depth, and pricing complexity to match.
- Developer clouds (DigitalOcean, Hetzner Cloud, Linode and peers) — a deliberately smaller catalog of instances, storage and managed databases with flat, predictable pricing.
- The managed layer — platforms that operate cloud servers for you, blending this model with managed hosting.
