Just adding on to this.

Using DBeaver as a client and using a singlenode cluster and a pool of idling VM, it was possible to get the autostart time of the cluster down to 35 seconds, + 17 seconds for the query time on top to show the first 200 rows of a 500,000 record object doesn't really compare to the other Datawarehouse SaaS products out there.

Even looked to make sure I was not using ML runtime.

Welcome others thoughts and opinions on what approach to take or just leave as not viable.