Object storage pricing: S3 vs Blob vs Cloud Storage vs OCI

Rohit Athaide · 2026年11月4日 公開

Short answer: object storage has four charges, which are stored data, requests, retrieval and data out. A low rate per gigabyte is only the first of them, and the cheapest-looking storage can carry the most expensive extras. To compare clouds, estimate all four for your pattern of use.

Key points

• Stored data is billed per gigabyte-month, by storage class.
• Requests are billed per thousand or ten thousand reads, writes and listings.
• Cold classes add retrieval fees and minimum storage periods.
• Data leaving the cloud is egress and is billed separately.
• Small files and frequent reads push request charges up.

What are the four charges?

Which services are equivalent?

They are equivalent as general object stores with hot and cold classes. Class names and minimum storage periods differ, so match a cold class to a cold class by access pattern. See classes and lifecycle rules.

What does a worked example look like?

Made-up round numbers to show the method, not real prices. 1,000 GB of files, 5 million reads and 100,000 writes a month, and 200 GB read by the public.

The cold class wins here, because the data is read little. Change the reads to 100 million a month and the cold class loses, since its request and retrieval charges grow. The same files can sit in either class depending on how often they are used.

What makes the bill grow?

• Many small files. Request charges can exceed storage charges when you write millions of tiny objects.
• Cold classes read often. Cheap storage with a retrieval fee costs more than a warm class if you read it every week.
• Early deletion. Cold classes bill a minimum period even if you delete sooner.
• Reading from another region or the internet. Egress is its own line. See data transfer costs and using a CDN.
• Versioning and old copies. Every version is stored and billed.

How do I estimate it?

1. Count stored gigabytes and how they split between hot and cold.
2. Count requests per month, split into reads and writes.
3. Estimate gigabytes read back and gigabytes sent out.
4. Price each part on each cloud in the same region group.

The object storage compare page shows each cloud’s price lines and units, with links to vendor pages.

Which mistakes are common?

• Putting log files of tiny objects in a cold class.
• Forgetting versioning and keeping every old copy.
• Ignoring the minimum storage period when moving data between classes.
• Serving public files straight from storage instead of through a CDN.

Decision checklist

• How often is each dataset read? Weekly or more means a warm class.
• Are objects small? Combine them if you can.
• Is there a lifecycle rule to move old data down and delete expired data?
• Is public traffic served through a CDN?

Next steps

For long retention of machine data, see object storage for sensor archives. For queries over stored files, read data lake cost basics and partitioning and Parquet. The data platform estimator prices storage, queries and transforms together.