[lacunary]

it has to cost some amount in reserved capacity too. for every n suspended machines there is some small fraction of a machine's cpu/ram capacity that must be kept in reserve, like in a fractional lending system.

[fooker]

You can 'suspend' to SSD/aka hibernate.

Can be pretty fast.

[account42]

I think gp means that when a customer wants to connect to the VM there needs to be hardware (CPU and RAM) available to run it. While this can be less than the total number of (suspended) VMs it has to have some buffer of "unused" hardware to account for usage spikes that still needs to be paid for.

[littlestymaar]

I think gp's argument was that you must have enough capacity to run peak load, which will be way above average load.