If you autoscale based on CPU consumption, doesn’t the macro level scaling achieve the same thing? Is the worry scaling small scale services where marginal scal

If you autoscale based on CPU consumption, doesn’t the macro level scaling achieve the same thing? Is the worry scaling small scale services where marginal scaling is a higher multiple, e.g. waste from unused capacity?