Elasticity is the product
You are paying to add and remove capacity in minutes rather than weeks, and to pay for what you used. If your load is flat and predictable, a fixed server is usually cheaper.
Where the savings actually come from
Rarely raw compute. Usually from not staffing the things you no longer run:
- Hardware replacement
- Power and cooling
- Network redundancy
- Out-of-hours hands
The trap
Elastic pricing rewards architectures that scale down. A cloud deployment that never scales down is simply a more expensive server with better tooling.