Let me show you how it works by way of an example scenario and a diagram:

In this example, we have three ESXi hosts in a vSphere HA cluster. All the hosts are identical and have 6 GHz of CPU and 9 GB of memory.

There are five powered-on VMs that are also identical (in reality, getting identical VM requirements are rare; this is for academic purposes only). Each VM requires 2 GHz of CPU and 2 GB of memory, and a reservation is also set for the entire allocated CPU and memory.

So you have now gathered how many resources you have and how much you need.

Based on this, the current CPU failover capacity is ((18 GHz-10 GHz)/18GHz = 44%, and the current memory failover capacity is ((27 GB-10 GB)/27 GB) = 62%.

Now, because your cluster was configured with 25 percent failover capacity, 19 percent of CPU resources are still available to power on additional VMs. Also, you still have 37 of percent memory resources for additional VMs.