Cloud Storage Types Explained: Object, Block and File Storage (2026)
Cloud storage is not one thing. AWS, Azure, and GCP each offer object, block, file, and archive tiers with meaningfully different pricing models, minimum durations, retrieval speeds, and billing units. Picking the wrong tier for a workload can multiply your bill or delay a recovery by 15 hours.
This guide covers the four storage categories, a full pricing breakdown across all three major providers, eight billing gotchas with dollar-amount examples, and an interactive cost estimator you can use now.
The Four Cloud Storage Types
Cloud storage types differ by how data is addressed, how many systems can access it simultaneously, and what performance profile they deliver. Object storage addresses files as key-value pairs over HTTP, scales without provisioning, and suits static assets and data lakes. Block storage presents raw disk volumes to a single VM, delivers sub-millisecond I/O, and suits databases. File storage mounts a shared network filesystem to multiple VMs concurrently and suits legacy applications. Archive storage stores infrequently read data at the lowest per-GB cost, with retrieval times ranging from milliseconds (GCP) to 48 hours (AWS Deep Archive).
Govern the data behind your AI. The AI Data Governance & Quality Assessment: a checklist to keep your data trustworthy.
Your purchase helps keep our hubs free to read.
Cloud storage is billed as a measured service (NIST SP 800-145). You pay for what you provision or consume, metered per GB or GiB per month, with separate charges for data egress (outbound data transfer leaving the provider's network) and API calls.
Quick Reference: Type vs. Use Case
| Type | API | Access Pattern | Typical Latency | Relative Cost |
|---|---|---|---|---|
| Object | HTTP / S3 | Read / write anytime | ms to s | Low |
| Block | OS disk | Frequent random I/O | <1 ms | Medium |
| File | NFS / SMB | Shared concurrent | ms | Medium |
| Archive | Object API | Write-heavy, rare reads | ms to 48 hr | Very low |
AWS S3: Eight Storage Classes
Now that the four storage types are clear, here is how each major cloud provider implements them in practice, starting with AWS, which ships the broadest set of storage class options.
Amazon S3 stores over 500 trillion objects and handles more than 200 million requests per second. The 11 nines (99.999999999%) durability figure applies to every standard storage class. S3's eight classes span from millisecond-access Standard to $0.00099/GB Deep Archive, covering almost every cost and performance point.
| Storage Class | Price / GB / mo | Min Duration | Min Object Size | Retrieval Fee |
|---|---|---|---|---|
| Standard | $0.023 | None | None | None |
| Intelligent-Tiering | $0.023 (frequent tier) | None | 128 KB | None |
| Standard-IA | $0.0125 | 30 days | 128 KB | $0.01 / GB |
| One Zone-IA | $0.01 | 30 days | 128 KB | $0.01 / GB |
| Express One Zone | ~$0.16 / GB + $0.000400 / 1k requests | None | None | None |
| Glacier Instant Retrieval | $0.004 | 90 days | 128 KB | $0.03 / GB |
| Glacier Flexible Retrieval | $0.0036 | 90 days | 40 KB | Free (bulk) |
| Glacier Deep Archive | $0.00099 | 180 days | 40 KB | $0.02 / GB |
Pricing sourced from AWS S3 pricing page (us-east-1 region). Verify current rates at aws.amazon.com/s3/pricing before budgeting.
Intelligent-Tiering: The Set-and-Forget Option
Intelligent-Tiering monitors each object's access patterns and automatically moves it between a frequent-access tier and an infrequent-access tier. There is no retrieval charge when S3 moves objects, and no operational overhead after setup. A small monthly monitoring fee applies per object.
128 KB minimum applies. Objects smaller than 128 KB are billed at Standard rates and never moved to a lower-cost tier. Run aws s3api list-objects-v2 --query 'Contents[?Size<`131072`]' before migrating to confirm object size distribution.
Express One Zone: Single-Digit Milliseconds
S3 Express One Zone uses a Directory bucket in a single Availability Zone and delivers single-digit millisecond access, approximately 10x faster than Standard. It is designed for latency-sensitive ML training, analytics, and high-frequency trading data. Express One Zone uses a different pricing model (per-request plus storage), so compare total cost against Standard for your specific request volume.
Four S3 Bucket Types
S3 now has four distinct bucket types, each serving a different access pattern:
- General Purpose: Standard for most workloads. Stores objects across AZs for durability.
- Directory: Powers Express One Zone. Single AZ, built for high-throughput random access.
- Table: Delivers 3x faster metadata queries. Suited to analytics workloads over structured data.
- Vector: Designed for AI and ML embedding retrieval. Reduces data transfer for nearest-neighbor queries.
Data transfer: The first 100 GB of outbound data per month is free in aggregate across all S3 regions. After that, standard egress rates apply (typically $0.09/GB to the internet from us-east-1). Model your egress volume before committing to an architecture that reads large volumes of data outside AWS.
See the full class comparison at aws.amazon.com/s3/storage-classes for the latest IOPS (input/output operations per second) limits and throughput caps by class.
Azure Blob Storage: Five Tiers and Rehydration Risk
Azure Blob Storage uses access tiers rather than separate storage classes. Tiers can be set at the account level or changed on individual blobs. The critical difference from S3: Archive is an offline tier that requires rehydration before you can read data, taking up to 15 hours with the standard priority method.
| Tier | Price / GB / mo | Min Duration | Notes |
|---|---|---|---|
| Premium | SSD-backed (varies by type) | None | Low-latency, block blobs only |
| Hot | $0.018 | None | Default tier for active data |
| Cool | $0.01 | 30 days | Infrequent access, lower read throughput |
| Cold | $0.0045 | 90 days | Rare access, higher retrieval fee |
| Archive | $0.00099 | 180 days | Offline, rehydrate before reading (up to 15 hr) |
East US region pricing. Verify at azure.microsoft.com/pricing/details/storage/blobs.
Redundancy Options: Six Choices
Azure Blob offers six redundancy options ranging from single-datacenter copies to geo-redundant storage with secondary-region read access:
| Option | Copies | Scope | Notes |
|---|---|---|---|
| LRS | 3 | Same datacenter | Lowest cost, no AZ protection |
| ZRS | 3 | 3 AZs, same region | Survives AZ failure |
| GRS | 6 | LRS + async secondary region | Recovery after regional outage |
| RA-GRS | 6 | GRS + secondary read access | Secondary reads have up to 30 min lag |
| GZRS | 6 | ZRS primary + LRS secondary | Best of zone + geo protection |
| RA-GZRS | 6 | GZRS + secondary read | Maximum resilience |
Archive rehydration takes up to 15 hours. To read data from Azure Archive tier, you must first copy the blob to Hot or Cool and wait for rehydration. You cannot read archive blobs in-place. Build this delay into your RTOs (Recovery Time Objectives: the maximum acceptable downtime if you need to restore from this storage) before archiving SLA-bound data.
Early Deletion Penalties: A Real Dollar Example
Azure charges early deletion fees when you remove blobs before their tier's minimum duration expires. The fee equals the per-GB rate multiplied by the remaining days of the minimum period.
Example: Archive 1 TB, then delete it at day 90 (half of the 180-day minimum). Azure charges approximately $2.97 in early deletion fees for the remaining 90 days (1,000 GB × $0.00099/month × 3 remaining months). Model lifecycle costs before choosing Archive, especially for data with uncertain retention periods.
The same logic applies to Cool (30-day minimum) and Cold (90-day minimum). If data moves frequently between tiers before minimums are met, the penalties compound. Use Azure lifecycle management policies with daysAfterModificationGreaterThan conditions to enforce minimum ages automatically.
RA-GRS billing gotcha: Switching from RA-GRS to GRS mid-month still bills at the RA-GRS rate for 30 additional days. Plan redundancy changes at the start of a billing cycle.
For block storage options, see Azure Managed Disks (SSD, Ultra Disk, and shared disk options). Managed Disks bill in decimal GB, unlike GCP Persistent Disk which bills in GiB.
GCP Cloud Storage: The Archive Difference Nobody Mentions
GCP Cloud Storage has four standard classes plus the Rapid bucket type for streaming workloads. The headline capability that sets it apart from AWS and Azure: GCP Archive retrieves data in milliseconds, using the same API as Standard storage. There is no rehydration wait, no special retrieval job, just a per-GiB retrieval fee when you read the data.
| Class | Price / GiB / mo | Min Duration | Retrieval Fee | Retrieval Speed |
|---|---|---|---|---|
| Standard | $0.02 | None | None | Milliseconds |
| Nearline | $0.01 | 30 days | $0.01 / GiB | Milliseconds |
| Coldline | $0.004 | 90 days | $0.02 / GiB | Milliseconds |
| Archive | $0.0012 | 365 days | $0.05 / GiB | Milliseconds |
| Rapid (zonal) | $0.000150685 / GiB-hr | None | None | Milliseconds |
US multi-region pricing. GCP Cloud Storage bills per GiB, same as Persistent Disk. Both use binary gibibytes (1 GiB = 2^30 bytes = 1.0737 GB). Verify at cloud.google.com/storage/pricing.
GCP Archive uses the same API as Standard. No rehydration, no special request type. Read any Archive object directly. You pay a $0.05/GiB retrieval fee, but access is always available immediately. For 1 TiB (1,024 GiB) of reads, that is $51.20 in retrieval fees. Factor this into total cost before assuming Archive is cheaper than Coldline for your access pattern.
Watch the 365-Day Minimum
GCP Archive has a 365-day minimum storage duration, compared to 180 days for AWS Glacier Deep Archive and Azure Archive. If your data retention policy requires destruction at 12 months, that matches the minimum exactly. If your policy is shorter, you may face early deletion fees. Confirm retention policies align before choosing GCP Archive for compliance data.
GCP Persistent Disk: Block Storage Pricing
GCP offers five block storage types for VM-attached storage:
| Disk Type | Price / GiB / mo | Best For |
|---|---|---|
| Standard HDD (pd-standard) | $0.04 | Sequential reads, cold backups |
| Balanced SSD (pd-balanced) | $0.10 | General-purpose VMs |
| SSD (pd-ssd) | $0.17 | Databases, IOPS-sensitive workloads |
| Hyperdisk Balanced | $0.125 | Configurable IOPS and throughput |
| Local SSD | $0.08 | Scratch data, ephemeral cache only |
Local SSD data is lost when the VM stops or is deleted. Local SSD is physically attached to the host machine. If the VM instance stops, is preempted, or is deleted, all data on the Local SSD is gone. Use it only for scratch data, temporary processing results, or caches you can reconstruct. Never store production data or anything not replicated elsewhere on Local SSD.
Binary GiB billing on Persistent Disk. GCP Persistent Disk bills in binary gibibytes (1 GiB = 2^30 bytes = 1.0737 GB). A 1 TB (1,000 GB) disk is provisioned as approximately 931 GiB. But if you provision 1,024 GiB to get a full tebibyte, you pay about 7% more than a decimal-GB comparison would suggest. Model disk sizes in binary units when budgeting against AWS EBS or Azure Managed Disks, which bill in decimal GB.
Rapid Buckets: Streaming Workloads
Rapid is a zonal bucket type designed for high-I/O streaming workloads. It bills per GiB-hour ($0.000150685 / GiB-hr) rather than per month, making it cost-effective for data that needs fast access for a short period, such as ML training batches or video processing queues. Rapid buckets exist in a single zone and have no minimum storage duration or retrieval fee.
See full Persistent Disk specifications, IOPS limits, and regional availability at cloud.google.com/compute/docs/disks.
S3 vs. Azure Blob vs. GCS Compared
The table below distills the key decision points across all three platforms. Where a provider has a clear advantage, it is called out in the cell.
Billing Gotcha Register
These eight gotchas have caused real cost surprises for cloud teams. Each has a mitigation or detection method.
Find Your Storage Type
Answer four questions to get a storage-type recommendation for your workload.
Cloud Storage Cost Estimator
Enter your storage amount and select a tier category to see side-by-side monthly cost estimates across AWS, Azure, and GCP. Estimates cover storage only and exclude egress, API calls, and retrieval fees.
Frequently Asked Questions
Object storage keeps files as flat blobs addressed by a unique key, with rich metadata and HTTP access. It scales to exabytes and works well for backups, media, and logs. Block storage presents raw volumes to an OS, which then applies its own filesystem. It has much lower latency and is suited to databases and VM boot disks. File storage adds POSIX semantics and shared multi-VM access via NFS or SMB.
AWS Glacier Instant Retrieval returns data in milliseconds. Glacier Flexible Retrieval offers expedited (1–5 minutes), standard (3–5 hours), or bulk (up to 12 hours) options. Glacier Deep Archive takes 12–48 hours for standard retrieval and does not offer expedited. GCP Archive, by contrast, retrieves in milliseconds at any time. It uses the same API as Standard storage with no rehydration wait.
Azure charges an early deletion penalty equal to the per-GB rate multiplied by the remaining days in the minimum period. For Archive tier with a 180-day minimum, deleting 1 TB at day 90 costs approximately $2.97 in early deletion fees (1,000 GB × $0.00099/month × 3 remaining months). The same logic applies to Cool (30-day min) and Cold (90-day min). Use Azure lifecycle management policies with daysAfterModificationGreaterThan conditions to enforce minimum ages before tiering down.
GCP bills both Cloud Storage objects and Persistent Disk in binary gibibytes (GiB), where 1 GiB equals 2^30 bytes, or 1.0737 GB. This means a 1 TB (1,000 GB) disk is billed as approximately 931 GiB (if you provision 1,024 GiB to get a true tebibyte, you pay about 7% more than a decimal-GB comparison would suggest). AWS EBS and Azure Managed Disks bill in decimal GB. Always verify unit definitions on the vendor's pricing page before budgeting large workloads.
S3 Intelligent-Tiering monitors each object's access patterns and automatically moves it between frequent-access and infrequent-access tiers, with no retrieval charge for the moves. Use it when access patterns are unpredictable or vary significantly over time. It has a minimum object size of 128 KB. Objects smaller than that are billed at Standard rates and never moved. A small per-object monthly monitoring fee applies. For uniformly active data, Standard is simpler and the monitoring fee may not be justified.
AWS S3 Glacier Deep Archive and Azure Archive both cost $0.00099/GB per month, making them equally cheapest by storage price. GCP Archive costs $0.0012/GiB per month, slightly more. However, GCP Archive retrieves in milliseconds and requires no rehydration, while AWS Deep Archive takes 12–48 hours and Azure Archive takes up to 15 hours. If your workload ever needs rapid access to archived data, GCP Archive's operational advantage is significant, even at a slightly higher storage cost. The retrieval fee ($0.05/GiB on GCP vs. $0.02/GB on AWS) should be factored into total cost of ownership for any archive tier you expect to read from.