We used GCS (like S3 for GCP) for storing prometheus backups (via thanos) at $previous_job. That thing cost an arm and a leg for what it did (presumably, thanos was partially at fault but I didn't dig that much).
I moved to minio (in a GCP VM) and reduced our overall GCP bill by 70%. Yes, using cloud storage was 2/3 of the cost of our cloud infrastructure.
But overall, going too far in the self-hosted route has its costs. Hardware depreciation is one (and the author mentions UPS which seems huge in addition to being critical), cooling/powering, and of course the time for maintenance. If you are going this route you are doing this because you want to learn that stuff, not because you want to save the subscription. Otherwise, just use less services and keep price comparison lists updated.
The sweet spot is renting dedicated services. Most of the maintenance is someone else's job, and you still just get a single bill. Compared to the big clouds providers, cost reductions of 50% to 90% are common.
And devops costs often go down (I do contracting on this area, and those of my customers who opt for dedicated services invariably need less of my time)
I moved to minio (in a GCP VM) and reduced our overall GCP bill by 70%. Yes, using cloud storage was 2/3 of the cost of our cloud infrastructure.
But overall, going too far in the self-hosted route has its costs. Hardware depreciation is one (and the author mentions UPS which seems huge in addition to being critical), cooling/powering, and of course the time for maintenance. If you are going this route you are doing this because you want to learn that stuff, not because you want to save the subscription. Otherwise, just use less services and keep price comparison lists updated.