On: Why did one day of AI cost more than a month of servers?

This operation should not have been retried even once. The overall operation failed due to an unsatisfied precondition that a second attempt could not have rectified (the non-existent target resource). The status of the job should have transitioned to permanent failure, awaiting manual intervention.

I think we are talking at cross-purposes - a retry policy of retries=0 is still a retry policy