Storage Lifecycle Policy
Automated rules that move objects between storage classes as they age, matching cost to the declining probability that data will be read.
Storage cost is dominated not by the volume stored but by the class it is stored in, and the default class is the most expensive one.
Access probability falls sharply with age for almost all data. A lifecycle policy exploits that: hot storage for the first 30 days, infrequent-access for 90, archive for a year, deletion or deep archive thereafter. The differential between hot and archive classes is typically an order of magnitude.
The details that decide whether it saves money:
Retrieval cost and latency. Archive classes charge to read and may take hours to restore. A policy that archives data which is then read regularly costs more than leaving it hot.
Minimum storage duration. Cheaper classes bill a minimum period, so moving short-lived objects into them is a net loss.
Transition requests are charged, which matters for very large numbers of small objects — sometimes enough that consolidating small files first is the bigger saving.
The categories most often overlooked entirely: old snapshots and machine images, which accumulate silently and are rarely reviewed; unattached volumes left behind by terminated instances; incomplete multipart uploads, which are billed and invisible in the object listing; and log data retained at full fidelity long past any use.