Key Takeaways
- The cloud is a layered platform, not a location.
- IaaS, PaaS, serverless and managed services each have different trade-offs.
- Costs are usually driven by egress, storage and inefficient design — not compute.
- Multi-cloud is expensive and only worth it for a small subset of use cases.
- Security in the cloud is a shared responsibility with the provider.
The Service Model Ladder
| Layer | You Manage | You Do Not Manage |
|---|---|---|
| IaaS | OS, runtime, app, data | Hardware, hypervisor |
| PaaS | App, data | Runtime, OS, hardware |
| Serverless | App logic only | Everything else |
| SaaS | Configuration | Everything else |
The higher you climb, the less operational surface you own — and the more of your bill is a productivity trade.
How To Choose A Layer
- Predictable heavy compute — IaaS with reservations.
- Standard web app with a database — PaaS.
- Event-driven, spiky traffic — serverless.
- Anything the vendor already does well — managed service.
Most modern architectures mix all four.
The Big Three Providers
- AWS — the widest catalogue, the deepest tooling, the most complexity.
- Azure — strong enterprise integration, best for organisations on Microsoft.
- Google Cloud — best-in-class data and AI stack, cleaner developer experience.
For a side-by-side comparison see the AWS vs Azure vs Google Cloud article.
Common Mistakes
- Lifting-and-shifting on-prem architectures without redesign.
- Choosing exotic managed services that lock in with no exit.
- Ignoring egress costs.
- Under-investing in identity and access management.
- Treating the cloud as a magic scale button.
Final Summary
Cloud computing is powerful, but the value comes from thoughtful design and disciplined operations. The vendors compete on services; you compete on how you compose them.
