That's false. The only thing which slows things down is that Autoscaler is provisioning compute resources in a way which is dog slow.

If you add Karpenter (https://karpenter.sh) to the cluster it can scale based on the rules set out in the configuration and sends out requests directly to the underlying cloud orchestrator. Which in practice means it spins up new compute resources near instantly.

I see that they added support for Azure aswell recently (been a while since I used it)