The 40% S3 lifecycle saving is real on one line of the bill and nowhere else

The 40 percent saving routinely quoted for S3 lifecycle policies survives contact with a calculator: a dev.to walkthrough puts 30 percent of a 100 TB us-east-1 bucket into Standard-IA and 40 percent into Glacier Instant Retrieval, and the storage line drops from $2,304.00 to $1,254.40, a 45.6 percent cut ^[2]. It matters because that figure covers one line item on one bucket shape, and the same piece says plainly that the saving leaks back out in the quarter after the rule ships, in the line items nobody re-opens ^[1][3].

The baseline is not in dispute. AWS bills 1 TB as 1,024 GB, so 100 TB is 102,400 GB ^[5]; all-Standard, that is 51,200 GB at $0.023 plus 51,200 GB at $0.022, or $2,304.00 a month ^[6]. The author says every rate comes from the Price List Bulk API us-east-1 offer file dated 2026-08-07, or is derived from it ^[4]. So the move saves $1,049.60 a month ^[7], $12,595.20 a year ^[8], and takes the blended rate across the whole bucket from $0.0225 to $0.01225 per GB-month ^[9]. Note what it costs to get there: 70 percent of your bytes leave Standard, leaving 30 percent hot ^[10]. The write-up concedes the 45.6 percent assumes roughly two-thirds of bytes are older than 30 days, the usual shape for logs, backups and ML snapshots ^[11].

Now the leakage. Requests and retrievals bill on top of storage, and the piece names them as exactly where the saving disappears ^[12]. Billing starts the instant the rule is satisfied, not when the copy completes, so you pay the destination rate from day one while AWS is still moving bytes in the background ^[13]. Glacier Flexible and Deep Archive add 40 KB of overhead per object, 8 KB at S3 Standard rates and 32 KB at the destination Glacier rate ^[15] - so 20 percent of that overhead is billed at your most expensive tier ^[16], and ten million objects means 381.47 GB of metadata you are paying to store ^[17]. Transitions also only go down the ladder, with the User Guide saying the Deep Archive transition "can go only one way" ^[14]. The advice is to pick the class by read frequency and then check the minimum duration, because per-GB storage and per-GB retrieval are the two numbers that drive cost ^[18].

One honest gap worth copying: the us-east-1 offer file exposes Deep Archive transition, retrieval and checksum SKUs but no per-GB timed-storage SKU, so the author refuses to quote a rate and tells you to pull it from the pricing console before budgeting, offering Intelligent-Tiering's Deep Archive Access tier at $0.00099/GB-mo as the nearest verifiable neighbour ^[19]. That neighbour is about 23 times cheaper than Standard's first band ^[20], which is precisely why an unverified rate is a dangerous thing to build a forecast on. One Zone-IA is described as 20 percent cheaper at 99.5 percent availability in a single AZ, suitable for thumbnails, transcodes or a second backup copy, never the only copy ^[21].

Watch the object-size filter in the sample rule: logs/ with ObjectSizeGreaterThan 131072, tiering at 30 days to Standard-IA and 120 days to Glacier IR ^[24]. That threshold is 128 KB ^[25], and it exists because small objects cost more to move than to keep. Sixty days after your rule ships, re-open retrieval GB, request counts and per-object overhead, not the storage line you already congratulated yourself on.