Extensive studies show that on average, a production server utilizes only about 10~20% of its available resources (CPU, memory, disk).
The primary reason for this is simple: if you don't allow for headroom for future growth or daily or periodic traffic bursts, the system will suffer performance degradation.
However, if you can collect 5 or 10 of these server instances on a single piece of hardware, chances are that not all of them will require extra resources at the same time.
So you still need the exact same amount of actual physical resources, but overhead can be managed and provisioned much more efficiently.
Don't underestimate the advantages of the ability to instantly migrate a complete server from dead hardware to working hardware, either.