In tip.science we currently have 1 autoscaling service that handles all the scaling.

For the electrode-selector for example the created EC2 do not have a GPU, and the node label gpu does not make sense. Nevertheless this is currently set via ENV NODES_MONITORING_NODE_LABELS which is global.
It would make sense to have labels that get set only for specific EC2 type (e.g. t3.medium shall not have gpu=true but g4dn yes)

Otherwise multiple autoscaling services shall be used.