Amazon S3 cost, made transparent: see what you pay, what sits unopened, and what each change saves

Why the S3 bill is hard to read

S3 is metered in six ways: storage per GB-month by storage class, requests per thousand, retrievals per GB on the colder classes, data transfer out to the internet, management features you turned on, and lifecycle transitions, which are billed as requests. Each shows up as its own usage type, and the console bill sums them per Region rather than per bucket.

Some of the storage you pay for is also invisible in the normal bucket view. A versioned bucket keeps every overwritten and deleted object as a noncurrent version until a lifecycle rule removes it. An upload that was interrupted leaves its parts in the bucket as an unfinished multipart upload, billed at the Standard rate, and the console object list does not show them.

• The per-GB rate depends on the storage class, and a bucket can hold objects in several classes at once.
• Colder classes carry minimums: 30 or 90 or 180 days of billing however soon you delete, and a 128 KB minimum billable size on the infrequent and archive instant classes.
• Moving a million objects to a cheaper class costs a million transition requests, which is why the first month of a lifecycle rule can look worse than the month before.
• The first 100 GB of transfer out each month is free across the account, and everything above it is billed per GB.

What AWS gives you to decode it

AWS does publish everything needed to answer the four questions. The pieces live in different consoles and none of them is per file.

Cost Explorer breaks the bill down by usage type, so you can see storage against requests against transfer for the last months. It only becomes per bucket if you set up cost allocation tags in advance. CloudWatch publishes a daily storage metric per bucket and per storage class, which is how you learn which buckets hold the bytes and in which class. S3 Storage Lens adds an account-wide dashboard, with the free tier covering object counts and sizes and the paid advanced metrics covering more. S3 Inventory writes a paid listing of every object. Server access logs and CloudTrail data events record reads, but only from the day you turn them on, and only if you keep and query the files.

Cost Explorer: understanding your S3 usage report ↗CloudWatch storage metrics per bucket and storage class ↗S3 Storage Lens ↗

The four questions a transparent view has to answer

Put the pieces together and the questions a finance lead or an operations manager actually asks are simple. The work is in joining the sources and in being honest about which figures are measured and which are estimated.

• What do we pay now, and for what: bytes stored per storage class at the account's real billed rates, requests, retrievals and transfer, from the bill rather than from a calculator.
• Which buckets carry it: each bucket's storage cost at list price, its object count and class mix, and whether it already has lifecycle rules or Intelligent-Tiering.
• How much of the data is unread: a figure with its evidence level stated. Billed transfer out puts a hard ceiling on how much was read from outside the Region, so when nothing inside the Region can read a bucket for free, the unread share is measured. When something can, the figure is an estimate and the gap should be named.
• What each change saves, net: the monthly reduction in dollars and as a share of the cost, minus what the change adds, such as transition requests, the Intelligent-Tiering monitoring fee or a retrieval charge.

The three changes that cut most bills

Once the picture is clear, the fixes are usually the same three lifecycle changes, and all of them are standard S3 features you keep after any tool is gone.

• Expire old versions: a rule that removes noncurrent versions after a set number of days, keeping one newer version, and clears expired delete markers. Skipped on buckets with Object Lock.
• Abort unfinished uploads: a rule that deletes incomplete multipart uploads after a few days, so interrupted transfers stop being billed.
• Move cold data to Intelligent-Tiering: objects of 128 KB and more that nobody opens for 30 days drop to a tier that costs about 45 percent less, and after 90 days to one that costs about 80 percent less and still opens instantly. Archive tiers that need a restore wait are a separate opt-in per bucket.