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?