Object, Block and File Storage: Which One You Need

Three kinds of cloud storage with very different prices and properties. Using the wrong one is expensive and awkward.

17 March 2026 3 min read

Cloud storage comes in three shapes. They are priced very differently, they behave very differently, and choosing the wrong one produces systems that are both slow and expensive.

The three shapes

	Object	Block	File
Accessed by	HTTP API	Attached as a disk	Network mount
Shared access	Many readers and writers	Usually one instance	Many instances
Partial updates	No — replace the object	Yes	Yes
Capacity	Effectively unlimited	Fixed, resizable	Elastic
Relative cost	Lowest	Moderate	Highest
Latency	Tens of milliseconds	Sub-millisecond	Low milliseconds
Use for	Media, backups, static assets, data lakes	Databases, operating systems	Shared application data, legacy apps

Examples: S3, Cloud Storage, Blob Storage and Spaces are object. EBS, Persistent Disk and managed disks are block. EFS, Filestore and Azure Files are file.

Object storage: use it more than you think

It is the cheapest and most durable option, and a large share of what applications store on local disk belongs here instead.

User uploads, images and documents.
Backups and database dumps.
Static site assets, served through a CDN.
Logs and analytics data.
Machine learning datasets and model artefacts.
Anything written once and read many times.

Moving uploads from instance disk to object storage is often the single change that makes an application horizontally scalable, because it removes local state.

Storage classes

Object storage is tiered by access frequency. Cheaper storage costs more to read.

Class	Storage cost	Retrieval	For
Standard	Highest	Free, instant	Active data
Infrequent access	Lower	Charged per GB	Monthly access
Archive	Very low	Charged, minutes to hours	Compliance, old backups
Deep archive	Lowest	Charged, up to 12 hours	Long-term retention

Use lifecycle rules to move data automatically. Watch minimum storage durations — deleting an archived object early often incurs a charge for the remaining minimum period, which can make aggressive tiering counterproductive for short-lived data.

Block storage

Behaves like a physical disk. Attach it to one instance, format it, mount it. Databases and operating systems need it, because they require low-latency random access and partial writes.

Performance often scales with volume size — see cloud disk types.
Snapshots are the backup mechanism, and they accumulate cost silently.
Volumes can usually grow but not shrink, so start conservatively.
A volume left attached to nothing still bills. This is the most common source of storage waste.

File storage: useful and expensive

A network file system several instances can mount simultaneously. It solves shared-state problems for applications that expect a filesystem, and it costs several times more per gigabyte than object storage.

Legitimate uses: legacy applications that cannot be changed, shared configuration or content directories, and workloads genuinely needing POSIX semantics across machines. If the application could use object storage instead, it should.

Choosing

Does it need to look like a disk to a database or operating system? Block.
Do several machines need simultaneous filesystem access to the same data? File.
Is it read and written as whole items over the network? Object.
Unsure, and the application will accept it? Object — cheapest and most durable.

Common questions

Can I run a database on object storage?

Traditional databases, no — they need block semantics. Some analytics engines query data directly in object storage, which is a different architecture with different trade-offs.

Which is most durable?

Object storage, typically with eleven nines of durability through automatic replication. Block storage is durable but replicated within a smaller scope, which is why snapshots still matter.

Is object storage slow?

Higher latency per request than block, and very high aggregate throughput with parallel requests. For its intended uses the latency is irrelevant.

How do I choose a storage class?

By access pattern, not by age. Data read weekly belongs in standard even if it is old; data never read after a month belongs in archive even if it is new.

Where to go next

See keeping storage buckets private and cloud disk types.

AWS Google Cloud Microsoft Azure