1.  [](/)
2.  [Curated Lists](/curated-lists)
3.  Best S3-Compatible Object Storage Providers

[Back to All Lists](/curated-lists)

# Best S3-Compatible Object Storage Providers in 2026

We compared 17 S3-compatible object storage providers across pricing, egress fees, S3 API completeness, durability, compliance, and the gotchas nobody tells you until after migration. Every price re-verified against the provider's own pricing page in July 2026, including Wasabi's 90-day minimum retention on pay-as-you-go. STRG was added on November 18, 2026.

![Ethan Steininger](/people/ethan.webp)

By [Ethan Steininger](/blog/authors/ethan) · [LinkedIn](https://www.linkedin.com/in/ethansteininger/)

Published April 10, 2026 · Updated November 18, 2026

Last tested: July 28, 2026

17 tools evaluated

Whichever provider you pick, Mixpeek connects to it: Cloudflare R2, Backblaze B2, Wasabi, Tigris, STRG, Amazon S3, Google Cloud Storage and Azure Blob. Connect a bucket and the images, video, audio and documents in it become searchable.

[Make the files in your bucket searchable](https://mixpeek.com/docs/integrations/object-storage/overview)

[One-click templatesStart from a templateEach template deploys a working search app on your own files.Browse templates](/templates)

## Key Takeaways

*   Cloudflare R2 ranks first, best for read-heavy workloads (CDN origins, RAG, API serving) where egress costs would otherwise dominate: and where versioning/object lock aren't required. The only major provider with truly unlimited zero egress: no caps, no reasonable-use fine print, no CDN partner lock-in. Every byte downloaded is free.
*   Backblaze B2 is second, best for cost-conscious teams storing large volumes of data (backups, media, ML datasets) with CDN delivery via Cloudflare. The lowest mainstream storage price at $6.95/TB/mo combined with free CDN egress through the Bandwidth Alliance: and the only provider that publishes verifiable drive failure data for durability transparency.
*   The highest-ranked free option is MinIO (Self-Hosted) at #7. Free (AGPL-3.0 open source); AIStor enterprise license with support available.
*   We ranked 17 options on total cost of ownership (30%), s3 api compatibility (25%), durability & reliability (20%), feature completeness (15%) and migration & portability (10%). Prices and claims were checked against each vendor's public pages in July 2026.

Skip the comparison? Mixpeek runs s3-compatible object storage providers on your own data: extraction, indexing, and search in one platform.

[Get started](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=curated_list_quick_answer&utm_campaign=mvs_onramp)

## How We Evaluated

Evaluated by the Mixpeek engineering team, who build and operate multimodal retrieval infrastructure in production. Last tested July 2026; rankings re-checked when the market shifts, with pricing and claims verified against each vendor's public documentation.

### Total Cost of Ownership

30%

Not just storage $/GB: we modeled egress, API requests, minimum retention penalties, and escape costs (what it costs to leave). A $0.005/GB provider with hidden retention fees can cost more than a $0.015/GB provider with zero egress.

### S3 API Compatibility

25%

Tested against real workloads: multipart uploads, presigned URLs, lifecycle policies, event notifications, object lock, versioning, and batch operations. 'S3 compatible' is a spectrum: we tested where each provider falls on it.

### Durability & Reliability

20%

Published durability guarantees, SLA commitments, erasure coding details, and whether the provider actually shows their math. Some providers claim 11 nines without publishing verifiable data.

### Feature Completeness

15%

Versioning, object lock, event notifications, replication, lifecycle policies, static hosting, CDN integration, and compute ecosystem.

### Migration & Portability

10%

How easy is it to move data in and out? Egress fees, migration tooling (rclone, Super Slurper, Sippy), and vendor lock-in risk.

## Quick answer

The short version, before the detail:

*   [Cloudflare R2](#item-1)best for read-heavy workloads (cdn origins, rag, api serving) where egress costs would otherwise dominate: and where versioning/object lock aren't requiredThe only major provider with truly unlimited zero egress: no caps, no reasonable-use fine print, no CDN partner lock-in. Every byte downloaded is free.
*   [Backblaze B2](#item-2)best for cost-conscious teams storing large volumes of data (backups, media, ml datasets) with cdn delivery via cloudflareThe lowest mainstream storage price at $6.95/TB/mo combined with free CDN egress through the Bandwidth Alliance: and the only provider that publishes verifiable drive failure data for durability transparency.
*   [Tigris](#item-3)best for teams that need r2-style zero egress but can't compromise on versioning or compliance (object lock/worm)The only zero-egress provider that also supports versioning, object lock, and multiple storage tiers, combining R2's cost advantage with S3's compliance features.
*   [AWS S3](#item-4)best for teams already deep in the aws ecosystem that need lambda triggers, sagemaker pipelines, or athena analytics on their storageThe reference S3 implementation with the deepest ecosystem of any cloud provider: 6 storage tiers, native vector search, and seamless integration with 200+ AWS services.
*   [Hetzner Object Storage](#item-5)best for eu-based teams that need affordable s3 storage with gdpr-friendly data residency and full versioning/lock supportThe cheapest fully-featured S3 provider (versioning + object lock + lifecycle) with EU-only data residency guaranteed by infrastructure location, not just policy.
*   [Wasabi](#item-6)best for cold or write-once storage where objects live longer than 90 days: backups, archives, compliance data. not for high-churn datasetsOne flat $7.99/TB/mo with no egress and no API fees: but the 90-day minimum retention per object is the critical trade-off that defines which workloads fit.
*   [MinIO (Self-Hosted)](#item-7)best for air-gapped, on-prem, or regulated environments where data cannot leave your infrastructure. also ideal for local dev/test against s3 apisThe most complete open-source S3 implementation with event notifications, versioning, object lock, and erasure coding: the only option that runs entirely on your own infrastructure.
*   [Google Cloud Storage](#item-8)best for teams already on google cloud using vertex ai or bigquery: but be aware of the highest egress costs of any major providerDeepest Google Cloud integration with Autoclass automatic tiering: the only provider where storage tier management is fully hands-off based on actual access patterns.

## Overview

For most workloads in 2026 the answer is Cloudflare R2 at $15/TB/month if you serve data out, because zero egress fees dominate every other line item once downloads exceed roughly a third of what you store. Choose Backblaze B2 at $6.95/TB/month if you store far more than you read, Wasabi at $7.99/TB/month flat if you want one predictable number and can live with a 90-day minimum retention, Hetzner if your compute is already in Europe, and AWS S3 if you need the deepest tiering and the largest ecosystem rather than the lowest price. Self-host MinIO only when a compliance boundary requires it, since you take on the durability engineering yourself. That ranking follows from one finding: sticker price per terabyte is the least important number on the page. Egress, minimum retention windows, per-request charges and the cost of leaving routinely swing total cost by more than the storage rate itself, so a $6.95/TB provider with a 3x egress allowance can beat a $4/TB provider once you actually read the data. S3 compatibility is likewise a spectrum rather than a checkbox: we tested each provider against multipart uploads, presigned URLs, lifecycle policies, event notifications, object lock, versioning and batch operations, and several nail the core API while missing versioning or capping object size. Every price below was re-verified against the provider's own pricing page in July 2026, and where a figure could not be machine-verified we left the prior number rather than restate a guess. If the goal is vector search directly over these buckets, not just storing training data, see [Can you run a vector database on S3?](/guides/vector-database-on-s3-object-storage) for how storage-first vector search works. Object storage alone returns whole objects, so whether you can search what sits in a bucket without downloading it first is a question about [compute pushdown](/guides/compute-pushdown-search-object-storage) rather than about the bucket.

## Best S3-Compatible Object Storage Providers: comparison at a glance

#

Tool

Best for

Pricing

Key differentiator

Main limit

1

[Cloudflare R2](#item-1)

Read-heavy workloads (CDN origins, RAG, API serving) where egress costs would otherwise dominate: and where versioning/object lock aren't required

$15/TB/mo storage; $0 egress; $4.50/1M Class A ops; $0.36/1M Class B ops; 10 GB free forever

The only major provider with truly unlimited zero egress: no caps, no reasonable-use fine print, no CDN partner lock-in. Every byte downloaded is free.

No versioning: accidental deletes are permanent

2

[Backblaze B2](#item-2)

Cost-conscious teams storing large volumes of data (backups, media, ML datasets) with CDN delivery via Cloudflare

$6.95/TB/mo storage; egress free up to 3x average monthly storage, then $0.01/GB; unlimited free egress via CDN partners (Fastly, Cloudflare, bunny.net); Class A/B/C API calls free, Class D $0.004/10K; first 10 GB free

The lowest mainstream storage price at $6.95/TB/mo combined with free CDN egress through the Bandwidth Alliance: and the only provider that publishes verifiable drive failure data for durability transparency.

Only 3 regions: US-West, US-East, EU-Central

3

[Tigris](#item-3)

Teams that need R2-style zero egress but can't compromise on versioning or compliance (object lock/WORM)

$20/TB/mo (Standard); $10/TB/mo (Infrequent Access); $4/TB/mo (Archive and Archive Instant Retrieval); egress free (Tigris pricing page, checked 2026-11-17)

The only zero-egress provider that also supports versioning, object lock, and multiple storage tiers, combining R2's cost advantage with S3's compliance features.

$20/TB/mo storage: more expensive than R2 or B2

4

[AWS S3](#item-4)

Teams already deep in the AWS ecosystem that need Lambda triggers, SageMaker pipelines, or Athena analytics on their storage

$23/TB/mo (Standard, first 50 TB); $0.09/GB egress to internet; 6 tiers down to $3.60/TB/mo (Glacier Deep Archive); $5/1M PUTs

The reference S3 implementation with the deepest ecosystem of any cloud provider: 6 storage tiers, native vector search, and seamless integration with 200+ AWS services.

$23/TB/mo: 4x more expensive than B2 for raw storage

5

[Hetzner Object Storage](#item-5)

EU-based teams that need affordable S3 storage with GDPR-friendly data residency and full versioning/lock support

EUR 6.26/TB/mo storage and EUR 1.00/TB egress, both read off Hetzner's own pricing widget (excl. VAT); base plan bundles the first 1 TB stored and 1 TB out; ingress and internal transfer free; full S3 API

The cheapest fully-featured S3 provider (versioning + object lock + lifecycle) with EU-only data residency guaranteed by infrastructure location, not just policy.

EU-only: no North American or APAC regions

6

[Wasabi](#item-6)

Cold or write-once storage where objects live longer than 90 days: backups, archives, compliance data. Not for high-churn datasets

$7.99/TB/mo flat (pay-as-you-go); no egress or API fees; 90-day minimum retention per object on pay-as-you-go, 30-day on Reserved Capacity

One flat $7.99/TB/mo with no egress and no API fees: but the 90-day minimum retention per object is the critical trade-off that defines which workloads fit.

90-day minimum retention PER OBJECT: delete early, pay anyway

7

[MinIO (Self-Hosted)](#item-7)

Air-gapped, on-prem, or regulated environments where data cannot leave your infrastructure. Also ideal for local dev/test against S3 APIs

Free (AGPL-3.0 open source); AIStor enterprise license with support available

The most complete open-source S3 implementation with event notifications, versioning, object lock, and erasure coding: the only option that runs entirely on your own infrastructure.

You manage everything: hardware, upgrades, monitoring, backups, durability

8

[Google Cloud Storage](#item-8)

Teams already on Google Cloud using Vertex AI or BigQuery: but be aware of the highest egress costs of any major provider

$20/TB/mo (Standard); $0.12/GB egress (highest of big 3); Autoclass tiering available

Deepest Google Cloud integration with Autoclass automatic tiering: the only provider where storage tier management is fully hands-off based on actual access patterns.

Highest egress fees: $0.12/GB, moving 100 TB out costs $12,000

9

[Oracle Cloud Object Storage](#item-9)

Teams that need substantial free egress (up to 10 TB/mo) but can't commit to Cloudflare's ecosystem for R2

$25.50/TB/mo (Standard); $8.50/TB egress ($0.0085/GB) after 10 TB free; $0.40/10K requests

10 TB/month of free egress: the most generous free transfer tier of any provider, making it cost-competitive for moderate egress workloads without locking into Cloudflare's ecosystem.

$25.50/TB/mo standard storage: second most expensive after S3

10

[DigitalOcean Spaces](#item-10)

Small teams and startups storing web assets, documents, and small files: not suitable for large objects like video or model weights

$5/mo base (250 GB + 1 TB transfer); $20/TB/mo additional storage; $10/TB egress

The simplest all-in-one package for small teams: S3 storage plus CDN plus developer-friendly docs in a single $5/mo plan. No other provider bundles CDN at this price.

Max object size is 5 GB: not 5 TB like nearly every other provider

11

[Linode (Akamai) Object Storage](#item-11)

Teams on Linode/Akamai infrastructure that want co-located S3 storage with CDN integration

$5/mo base (250 GB + 1 TB transfer); $20/TB/mo additional storage; $10/TB additional egress

Backed by Akamai's global CDN infrastructure: the strongest edge delivery network of any independent storage provider.

5 GB max object size: same limitation as DigitalOcean Spaces

12

[OVHcloud Object Storage](#item-12)

EU teams wanting zero-egress storage with multiple tiers and strong data sovereignty guarantees

$7/TB/mo (Standard); $4/TB/mo (IA); $1.50/TB/mo (Archive); zero egress; 30-day minimum retention

Zero-egress EU-sovereign storage with three pricing tiers: the only European provider combining free egress with archive-class pricing at $1.50/TB/mo.

30-day minimum retention on ALL tiers: shorter than Wasabi but still a gotcha

13

[Storj](#item-13)

Large-scale cold storage and backup where cost and encryption matter more than access speed

$7/TB/mo storage; $7/TB egress; $5/mo account minimum; 25 GB free on a 30-day trial

Decentralized architecture across 30,000+ independent nodes with mandatory client-side encryption: no single entity (including Storj) can access your data.

Higher tail latency than centralized providers (P99 can spike)

14

[Azure Blob Storage](#item-14)

Teams on Azure that need tight integration with Synapse, Azure ML, or Data Lake Storage Gen2

$18/TB/mo (Hot); $10/TB/mo (Cool); $1/TB/mo (Archive); $0.087/GB egress

Deepest integration with Microsoft's analytics and ML stack: Synapse, Azure ML, Data Lake Gen2, and Cognitive Services all work natively with Blob Storage.

S3 compatibility endpoint is in preview: gaps in multipart upload and presigned URLs

15

[STRG](#item-15)

Teams that serve or re-read their data often and want one predictable per-TB bill with egress included

$6.80/TB/mo storage with all egress and standard API requests included; first 10 GB free; no minimum storage period or retrieval fees (strg.com, checked 2026-11-18)

Egress and standard API requests are part of the per-TB price, with no minimum storage period.

Only two regions (US-East and EU-Central)

16

[IDrive e2](#item-16)

Teams that want Wasabi-level pricing without the 90-day retention trap: ideal for iterative workloads with data churn

$6/TB/mo storage ($6/mo minimum); egress free up to 3x active storage, then $0.01/GB; no minimum retention; no API fees

The cheapest S3 storage with zero minimum retention: the only sub-$5/TB provider where you can delete objects the same day without penalty.

Reasonable-use egress policy limits heavy download workloads

17

[Rabata](#item-17)

Teams that want predictable pricing without per-request fees: especially workloads with high API call volume where GET/PUT fees add up

$10/TB/mo storage; $10/TB egress; $0 API fees; Backup tier $49/10TB/mo

The only S3-compatible provider with truly zero API fees: no per-request charges on any operation, making it ideal for high-frequency read/write workloads.

No free egress tier: all egress charged at $0.01/GB

## The Mental Model That Makes Multimodal Search Click

![](https://i.ytimg.com/vi/6mQK21ivkSs/hqdefault.jpg)

### Jump to

[#1 Cloudflare R2](#item-1)[#2 Backblaze B2](#item-2)[#3 Tigris](#item-3)[#4 AWS S3](#item-4)[#5 Hetzner Object Storage](#item-5)[#6 Wasabi](#item-6)[#7 MinIO (Self-Hosted)](#item-7)[#8 Google Cloud Storage](#item-8)[#9 Oracle Cloud Object Storage](#item-9)[#10 DigitalOcean Spaces](#item-10)[#11 Linode (Akamai) Object Storage](#item-11)[#12 OVHcloud Object Storage](#item-12)[#13 Storj](#item-13)[#14 Azure Blob Storage](#item-14)[#15 STRG](#item-15)[#16 IDrive e2](#item-16)[#17 Rabata](#item-17)

1

### Cloudflare R2

Zero egress fees: no asterisks, no reasonable-use policy, no CDN partner requirements. R2 eliminates the single biggest cost trap in object storage. Fully S3-compatible, globally distributed across 330+ edge PoPs, and deeply integrated with Cloudflare Workers for serverless compute. The missing features (no versioning, no object lock) are real trade-offs, but for most workloads the zero-egress economics are transformative.

What Sets It Apart

The only major provider with truly unlimited zero egress: no caps, no reasonable-use fine print, no CDN partner lock-in. Every byte downloaded is free.

Use with MVS

Use Cloudflare R2 as the object-storage backend for read-heavy MVS namespaces where agents repeatedly retrieve vectors, payloads, and source media without egress fees dominating the bill.

#### Strengths

*   +Truly zero egress fees: saves $900+/mo on a 10 TB read-heavy workload vs S3
*   +Full S3 API compatibility: works with boto3, rclone, every standard tool
*   +330+ edge PoPs with automatic global distribution
*   +Workers integration for serverless processing at the edge
*   +Generous free tier: 10 GB storage + 10M reads/mo forever

#### Limitations

*   \-No versioning: accidental deletes are permanent
*   \-No object lock or WORM compliance: not suitable for regulated data
*   \-No S3 Select or batch operations
*   \-$15/TB/mo storage is 2.5x more than B2 or Wasabi

#### Real-World Use Cases

*   •CDN origin storage where content is served globally without egress penalties eating into margins
*   •API backends that serve user-uploaded media: every download is free regardless of volume
*   •RAG document stores where retrieval queries pull chunks continuously without accumulating transfer fees
*   •Static site and JAMstack deployments with Workers-based server-side rendering at the edge

#### Choose This When

When egress cost is your primary concern: content delivery, API serving, or any workload that reads far more than it writes. R2 also wins when you want edge compute via Workers.

#### Skip This If

When you need versioning, object lock, or WORM compliance for regulatory requirements. Also avoid if your workload is storage-heavy but read-light, since $15/TB/mo storage is expensive compared to B2 or Wasabi.

#### Integration Example

import boto3

r2 = boto3.client(
    "s3",
    endpoint\_url="https://ACCOUNT\_ID.r2.cloudflarestorage.com",
    aws\_access\_key\_id="R2\_ACCESS\_KEY",
    aws\_secret\_access\_key="R2\_SECRET\_KEY",
)

# Upload assets: zero egress when users download
r2.upload\_file("/tmp/video.mp4", "media-bucket", "videos/promo.mp4")

# Generate presigned URL for direct client download
url = r2.generate\_presigned\_url(
    "get\_object",
    Params={"Bucket": "media-bucket", "Key": "videos/promo.mp4"},
    ExpiresIn=3600,
)
print(f"Download (zero egress): {url\[:60\]}...")

$15/TB/mo storage; $0 egress; $4.50/1M Class A ops; $0.36/1M Class B ops; 10 GB free forever

Best for: Read-heavy workloads (CDN origins, RAG, API serving) where egress costs would otherwise dominate: and where versioning/object lock aren't required

[Visit Website](https://cloudflare.com/r2)

2

### Backblaze B2

The best value in object storage. B2 costs 1/4 of AWS S3 with full S3 compatibility, 10 TB max object size (largest on this list), and free egress to Cloudflare and Fastly via the Bandwidth Alliance. Backblaze also publishes drive failure data openly: rare transparency in an industry where most providers won't show their durability math.

What Sets It Apart

The lowest mainstream storage price at $6.95/TB/mo combined with free CDN egress through the Bandwidth Alliance: and the only provider that publishes verifiable drive failure data for durability transparency.

Use with MVS

Use Backblaze B2 for storage-heavy MVS namespaces that hold hundreds of millions of embeddings, media references, and metadata while keeping raw storage near B2's low per-TB pricing.

#### Strengths

*   +Cheapest mainstream storage at $6.95/TB/mo: 1/3 the price of S3
*   +Free egress to Cloudflare, Fastly, and other CDN partners
*   +10 TB max object size: largest of any provider
*   +Full S3 API compatibility with both native and S3-compatible APIs
*   +Publicly published drive failure statistics: verifiable durability

#### Limitations

*   \-Only 3 regions: US-West, US-East, EU-Central
*   \-No serverless compute integration
*   \-Single-tier storage: no archive or IA pricing options
*   \-100 bucket limit per account

#### Real-World Use Cases

*   •Backup and disaster recovery for enterprise data with free CDN-routed restores via Cloudflare
*   •Media asset storage for video platforms delivering content through Cloudflare CDN at zero egress
*   •Large-scale dataset hosting where $6.95/TB/mo keeps storage costs predictable at petabyte scale
*   •NAS replacement for organizations migrating from on-prem storage to cloud with minimal cost increase

#### Choose This When

When you need the cheapest reliable storage at scale and deliver content via Cloudflare or Fastly CDN. Also ideal for very large objects (up to 10 TB per object).

#### Skip This If

When you need multi-region availability, storage tiering (archive/IA classes), or more than 100 buckets per account.

#### Integration Example

import boto3

b2 = boto3.client(
    "s3",
    endpoint\_url="https://s3.us-west-004.backblazeb2.com",
    aws\_access\_key\_id="B2\_KEY\_ID",
    aws\_secret\_access\_key="B2\_APP\_KEY",
)

# Upload a 10 TB file (largest max object size of any provider)
b2.upload\_file("/tmp/large\_backup.tar", "backups", "daily/2026-04-01.tar")

# List recent backups
for obj in b2.list\_objects\_v2(Bucket="backups", Prefix="daily/")\["Contents"\]:
    print(f"{obj\['Key'\]}: {obj\['Size'\] / 1e9,.1f} GB")

$6.95/TB/mo storage; egress free up to 3x average monthly storage, then $0.01/GB; unlimited free egress via CDN partners (Fastly, Cloudflare, bunny.net); Class A/B/C API calls free, Class D $0.004/10K; first 10 GB free

Best for: Cost-conscious teams storing large volumes of data (backups, media, ML datasets) with CDN delivery via Cloudflare

[Visit Website](https://backblaze.com/cloud-storage)

3

### Tigris

Tigris fixes R2's biggest gaps, zero egress plus versioning and object lock, with automatic edge caching that moves data closer to where it's accessed. Built on FoundationDB for strong consistency. Newer service with less production track record than R2 or B2, but technically the most complete zero-egress option.

What Sets It Apart

The only zero-egress provider that also supports versioning, object lock, and multiple storage tiers, combining R2's cost advantage with S3's compliance features.

Use with MVS

Use Tigris behind MVS for globally distributed namespaces where vector search should stay close to users and object-lock/versioning requirements rule out simpler zero-egress stores.

#### Strengths

*   +Zero egress fees like R2, but with versioning and object lock
*   +Automatic edge caching: data follows access patterns globally
*   +4 storage tiers including archive
*   +Strong consistency guarantees via FoundationDB

#### Limitations

*   \-$20/TB/mo storage: more expensive than R2 or B2
*   \-Newer service with smaller community and less track record
*   \-Some rate limits not yet published
*   \-No static website hosting

#### Real-World Use Cases

*   •Compliance-sensitive workloads that need zero-egress economics AND object lock / WORM guarantees
*   •Multi-region applications where data automatically caches at the edge closest to users
*   •Tiered storage strategies using Standard, IA, and Archive classes within a single bucket
*   •Financial or healthcare data storage requiring strong consistency with audit-friendly versioning

#### Choose This When

When you need zero egress fees but cannot compromise on versioning, object lock, or WORM compliance. Also ideal for globally distributed workloads that benefit from automatic edge caching.

#### Skip This If

When you need a battle-tested provider with a long track record: Tigris is newer and has a smaller community. Also avoid if $20/TB/mo storage is too expensive for your volume.

#### Integration Example

import boto3

tigris = boto3.client(
    "s3",
    endpoint\_url="https://fly.storage.tigris.dev",
    aws\_access\_key\_id="TIGRIS\_KEY",
    aws\_secret\_access\_key="TIGRIS\_SECRET",
)

# Enable versioning: unlike R2, Tigris supports this
tigris.put\_bucket\_versioning(
    Bucket="compliant-data",
    VersioningConfiguration={"Status": "Enabled"},
)

# Upload with object lock for WORM compliance
tigris.put\_object(
    Bucket="compliant-data",
    Key="records/2026/q1.parquet",
    Body=open("/tmp/q1.parquet", "rb"),
    ObjectLockMode="GOVERNANCE",
    ObjectLockRetainUntilDate="2029-04-01T00:00:00Z",
)

$20/TB/mo (Standard); $10/TB/mo (Infrequent Access); $4/TB/mo (Archive and Archive Instant Retrieval); egress free (Tigris pricing page, checked 2026-11-17)

Best for: Teams that need R2-style zero egress but can't compromise on versioning or compliance (object lock/WORM)

[Visit Website](https://tigrisdata.com)

4

### AWS S3

The reference implementation and the deepest ecosystem in cloud computing. S3 is the most expensive option on this list, but it offers capabilities nobody else has: 6 storage tiers, S3 Vectors for native vector search, Intelligent Tiering, EventBridge integration, and tight coupling with Lambda, SageMaker, Athena, and the entire AWS stack. You pay a premium, but you get unmatched feature depth.

What Sets It Apart

The reference S3 implementation with the deepest ecosystem of any cloud provider: 6 storage tiers, native vector search, and seamless integration with 200+ AWS services.

Use with MVS

Use AWS S3 as the MVS storage layer when your vectors, source files, and enrichment jobs already live inside AWS and need IAM, EventBridge, SageMaker, or Bedrock integration.

#### Strengths

*   +Most complete S3 implementation: it IS the standard
*   +6 storage tiers with Intelligent Tiering for automatic optimization
*   +Deepest ecosystem: Lambda, SageMaker, Athena, EventBridge, Glacier
*   +S3 Vectors: native vector search built into S3
*   +33 regions with cross-region replication

#### Limitations

*   \-$23/TB/mo: 4x more expensive than B2 for raw storage
*   \-Egress at $0.09/GB dominates costs on read-heavy workloads ($689/mo for 10 TB stored + 5 TB egress)
*   \-Moving 100 TB out costs $9,000 in egress: financial lock-in
*   \-IAM complexity tax: policies, VPC endpoints, encryption configs

#### Real-World Use Cases

*   •Event-driven architectures with S3 notifications triggering Lambda, Step Functions, or EventBridge workflows
*   •Data lake storage with Athena SQL queries and Glue ETL pipelines running directly on S3
*   •Intelligent Tiering for unpredictable access patterns: automatically moves objects between hot and cold tiers
*   •Cross-region disaster recovery with automatic replication and 33-region availability

#### Choose This When

When you are already invested in AWS and need features that only S3 offers: Intelligent Tiering, EventBridge, Lambda@Edge, or S3 Vectors. The ecosystem premium pays for itself when you use adjacent services.

#### Skip This If

When cost matters more than ecosystem: S3 is 4x more expensive than B2 for storage and egress creates financial lock-in at scale. Moving 100 TB out costs $9,000.

#### Integration Example

import boto3

s3 = boto3.client("s3")

# Enable Intelligent Tiering for automatic cost optimization
s3.put\_bucket\_intelligent\_tiering\_configuration(
    Bucket="data-lake",
    Id="auto-tier",
    IntelligentTieringConfiguration={
        "Id": "auto-tier",
        "Status": "Enabled",
        "Tierings": \[
            {"AccessTier": "ARCHIVE\_ACCESS", "Days": 90},
            {"AccessTier": "DEEP\_ARCHIVE\_ACCESS", "Days": 180},
        \],
    },
)

# Upload with server-side encryption
s3.put\_object(
    Bucket="data-lake",
    Key="datasets/users.parquet",
    Body=open("/tmp/users.parquet", "rb"),
    ServerSideEncryption="aws:kms",
)

$23/TB/mo (Standard, first 50 TB); $0.09/GB egress to internet; 6 tiers down to $3.60/TB/mo (Glacier Deep Archive); $5/1M PUTs

Best for: Teams already deep in the AWS ecosystem that need Lambda triggers, SageMaker pipelines, or Athena analytics on their storage

[Visit Website](https://aws.amazon.com/s3)

5

### Hetzner Object Storage

The quiet winner for EU-based teams. Hetzner offers S3-compatible storage at EUR 6.26/TB/mo with full versioning, object lock, and EUR 1.00/TB egress (with 1 TB free internal transfer). EU-only regions (Germany, Finland) make it a strong fit for GDPR-sensitive workloads. No-nonsense provider with transparent pricing.

What Sets It Apart

The cheapest fully-featured S3 provider (versioning + object lock + lifecycle) with EU-only data residency guaranteed by infrastructure location, not just policy.

Use with MVS

Use Hetzner Object Storage with MVS for EU-only vector namespaces where data residency, object lock, and low storage cost matter more than hyperscaler-adjacent services.

#### Strengths

*   +EUR 6.26/TB/mo with full S3 features
*   +Versioning, object lock, and lifecycle policies included
*   +EU-only regions ideal for GDPR compliance
*   +Transparent, no-surprise pricing

#### Limitations

*   \-EU-only: no North American or APAC regions
*   \-Smaller community than AWS or Cloudflare ecosystems
*   \-No CDN integration or edge caching built in
*   \-Less mature tooling and documentation than hyperscalers

#### Real-World Use Cases

*   •GDPR-compliant data storage for EU SaaS applications that cannot store customer data outside the EU
*   •Budget backup storage for European businesses replacing expensive local NAS systems
*   •Static asset hosting for EU-focused web applications with low egress costs
*   •Versioned document storage for legal or financial firms needing audit trails with object lock

#### Choose This When

When you are EU-based, need GDPR-compliant data residency, and want full S3 features (versioning, object lock) at the lowest price available.

#### Skip This If

When you need storage outside Europe, a global CDN, or the rich ecosystem of adjacent services that hyperscalers provide.

#### Integration Example

import boto3

hetzner = boto3.client(
    "s3",
    endpoint\_url="https://fsn1.your-objectstorage.com",
    aws\_access\_key\_id="HETZNER\_KEY",
    aws\_secret\_access\_key="HETZNER\_SECRET",
)

# Enable versioning and object lock for compliance
hetzner.put\_bucket\_versioning(
    Bucket="eu-compliance",
    VersioningConfiguration={"Status": "Enabled"},
)

# Upload GDPR-sensitive data: stays in EU by design
hetzner.upload\_file(
    "/tmp/customer\_data.parquet",
    "eu-compliance",
    "customers/2026/q1.parquet",
)

EUR 6.26/TB/mo storage and EUR 1.00/TB egress, both read off Hetzner's own pricing widget (excl. VAT); base plan bundles the first 1 TB stored and 1 TB out; ingress and internal transfer free; full S3 API

Best for: EU-based teams that need affordable S3 storage with GDPR-friendly data residency and full versioning/lock support

[Visit Website](https://hetzner.com/storage/object-storage)

6

### Wasabi

Flat-rate storage at $7.99/TB/mo with no egress fees and no API fees. But the 90-day minimum retention policy is a deal-breaker for many workloads: delete or overwrite an object before 90 days and you still pay for the full 90 days. On a dataset with churn, this can double your effective cost.

What Sets It Apart

One flat $7.99/TB/mo with no egress and no API fees: but the 90-day minimum retention per object is the critical trade-off that defines which workloads fit.

Use with MVS

Use Wasabi for MVS namespaces with write-once or low-churn embeddings, such as archived media catalogs, compliance corpora, and historical RAG collections that will outlive the 90-day minimum.

#### Strengths

*   +One predictable number at $7.99/TB/mo: no egress, no API fees
*   +No egress fees and no API request fees
*   +Full S3 API compatibility
*   +16 regions across NA, EU, and APAC

#### Limitations

*   \-90-day minimum retention PER OBJECT: delete early, pay anyway
*   \-'Free' egress capped by reasonable-use policy (egress cannot exceed stored volume)
*   \-No lifecycle policies for automatic tiering
*   \-No event notifications

#### Real-World Use Cases

*   •Long-term backup storage where files are written once and retained for months or years
*   •Compliance archives for financial, legal, or healthcare data with mandated retention periods exceeding 90 days
*   •Cold storage for large media libraries that are infrequently accessed but must remain available
*   •NAS-to-cloud migration for organizations that want to replace aging on-prem hardware with cheap cloud storage

#### Choose This When

When you store write-once data that will live longer than 90 days (backups, archives, compliance records) and want the simplest possible pricing with no egress or API fees.

#### Skip This If

When your data has any churn: frequent updates, overwrites, or deletions within 90 days will trigger minimum retention charges. Also avoid if you need event notifications or lifecycle policies.

#### Integration Example

import boto3

wasabi = boto3.client(
    "s3",
    endpoint\_url="https://s3.wasabisys.com",
    aws\_access\_key\_id="WASABI\_KEY",
    aws\_secret\_access\_key="WASABI\_SECRET",
)

# Upload backup: remember: 90-day minimum retention applies
wasabi.upload\_file("/tmp/db\_backup.sql.gz", "backups", "daily/2026-04-01.sql.gz")

# WARNING: deleting before 90 days still incurs storage charges
# wasabi.delete\_object(Bucket="backups", Key="daily/2026-04-01.sql.gz")
# ^ You pay for remaining days even after deletion

$7.99/TB/mo flat (pay-as-you-go); no egress or API fees; 90-day minimum retention per object on pay-as-you-go, 30-day on Reserved Capacity

Best for: Cold or write-once storage where objects live longer than 90 days: backups, archives, compliance data. Not for high-churn datasets

[Visit Website](https://wasabi.com)

7

### MinIO (Self-Hosted)

The only serious self-hosted S3 implementation. MinIO runs on your hardware, your cloud, or your Kubernetes cluster with the most complete S3 API coverage of any open-source project. AGPL-3.0 licensed. You own everything: uptime, durability, backups, and the operational burden that comes with it.

What Sets It Apart

The most complete open-source S3 implementation with event notifications, versioning, object lock, and erasure coding: the only option that runs entirely on your own infrastructure.

Use with MVS

Use MinIO as the storage backend for self-hosted MVS deployments in air-gapped, regulated, or private-cloud environments where vectors and source objects cannot leave your infrastructure.

#### Strengths

*   +Most complete open-source S3 implementation
*   +Full control over data residency and security
*   +Hardware-bound performance: saturates whatever you give it (21+ TiB/s benchmarked)
*   +Rich event notifications: Webhooks, Kafka, NATS, Redis, PostgreSQL

#### Limitations

*   \-You manage everything: hardware, upgrades, monitoring, backups, durability
*   \-AGPL-3.0 license may restrict commercial use
*   \-Distributed mode is complex to operate at scale
*   \-Cost advantage disappears at small scale due to hardware amortization

#### Real-World Use Cases

*   •On-premises object storage for defense, government, or healthcare organizations with air-gap requirements
*   •Local development and CI/CD environments that need real S3 API behavior without cloud dependencies
*   •Kubernetes-native storage for containerized applications that need S3-compatible persistence
*   •High-throughput data pipelines on dedicated hardware where cloud egress costs are prohibitive

#### Choose This When

When data must stay on your infrastructure (air-gap, compliance, sovereignty), you have the ops team to manage it, and you need full S3 feature coverage including event notifications.

#### Skip This If

When you do not have dedicated infrastructure or Kubernetes expertise. Also avoid at small scale where hardware amortization makes cloud providers cheaper.

#### Integration Example

import boto3

minio = boto3.client(
    "s3",
    endpoint\_url="http://minio.internal:9000",
    aws\_access\_key\_id="minioadmin",
    aws\_secret\_access\_key="minioadmin",
)

# Create versioned bucket with event notifications
minio.create\_bucket(Bucket="data-pipeline")
minio.put\_bucket\_versioning(
    Bucket="data-pipeline",
    VersioningConfiguration={"Status": "Enabled"},
)

# Upload data: triggers webhook notification
minio.upload\_file("/tmp/batch.parquet", "data-pipeline", "inbox/batch\_001.parquet")

Free (AGPL-3.0 open source); AIStor enterprise license with support available

Best for: Air-gapped, on-prem, or regulated environments where data cannot leave your infrastructure. Also ideal for local dev/test against S3 APIs

[Visit Website](https://min.io)

8

### Google Cloud Storage

Strong ML ecosystem integration via Vertex AI and BigQuery, with Autoclass for automatic storage tier management. But GCS has the highest egress fees of any provider on this list at $0.12/GB: 33% more than AWS. The S3 compatibility layer is an interop API, not native, and has gaps around multipart presigned URLs.

What Sets It Apart

Deepest Google Cloud integration with Autoclass automatic tiering: the only provider where storage tier management is fully hands-off based on actual access patterns.

Use with MVS

Use GCS with MVS when your embedding generation, analytics, and source objects are already on Google Cloud and you want vector storage to follow the same Vertex AI and BigQuery data plane.

#### Strengths

*   +Tight Vertex AI, BigQuery, and Dataflow integration
*   +Autoclass handles lifecycle tiering automatically
*   +Global strong consistency
*   +40 regions with multi-region and dual-region options

#### Limitations

*   \-Highest egress fees: $0.12/GB, moving 100 TB out costs $12,000
*   \-S3 compatibility is an interop layer, not native: has gaps
*   \-365-day minimum retention on Archive tier
*   \-$20/TB/mo standard storage: expensive for large datasets

#### Real-World Use Cases

*   •Vertex AI training and serving pipelines that read data from GCS natively
*   •BigQuery analytics on structured and semi-structured data stored as Parquet or JSON in GCS
*   •Autoclass-managed storage for unpredictable access patterns where manual tiering is impractical
*   •Dual-region storage for disaster recovery within Google Cloud

#### Choose This When

When your infrastructure is on Google Cloud and you need native integration with Vertex AI, BigQuery, or Dataflow. Autoclass is also valuable for workloads with unpredictable access patterns.

#### Skip This If

When egress costs matter: GCS has the highest egress fees of any major provider at $0.12/GB. Also avoid if you need native S3 compatibility (GCS interop API has gaps).

#### Integration Example

from google.cloud import storage

client = storage.Client()
bucket = client.bucket("analytics-data")

# Upload with Autoclass: Google manages tiering automatically
blob = bucket.blob("events/2026/04/01.parquet")
blob.upload\_from\_filename("/tmp/events.parquet")

# Generate signed URL (native GCS: S3 interop has gaps here)
url = blob.generate\_signed\_url(
    version="v4",
    expiration=3600,
    method="GET",
)
print(f"Signed URL: {url\[:60\]}...")

$20/TB/mo (Standard); $0.12/GB egress (highest of big 3); Autoclass tiering available

Best for: Teams already on Google Cloud using Vertex AI or BigQuery: but be aware of the highest egress costs of any major provider

[Visit Website](https://cloud.google.com/storage)

9

### Oracle Cloud Object Storage

The most generous free egress tier of any provider: 10 TB/month. Oracle Cloud Object Storage is often overlooked, but for egress-heavy workloads that can't use R2, 10 TB of free monthly egress is a significant advantage. 46 regions, full S3 compatibility, and 3 storage tiers.

What Sets It Apart

10 TB/month of free egress: the most generous free transfer tier of any provider, making it cost-competitive for moderate egress workloads without locking into Cloudflare's ecosystem.

Use with MVS

Use Oracle Cloud Object Storage with MVS for Oracle-heavy enterprise namespaces that need broad region coverage, database-adjacent backups, and a generous free egress tier.

#### Strengths

*   +10 TB/mo free egress: most generous free tier of any provider
*   +Full S3 API compatibility
*   +46 regions: more than any other provider
*   +3 storage tiers with automatic tiering available

#### Limitations

*   \-$25.50/TB/mo standard storage: second most expensive after S3
*   \-Less polished console and documentation than AWS or GCS
*   \-Smaller community and ecosystem
*   \-Less tooling and SDK support than hyperscaler competitors

#### Real-World Use Cases

*   •Media distribution platforms that serve up to 10 TB/mo of content without egress fees
*   •Hybrid cloud architectures needing broad region coverage across 46 global locations
*   •Oracle Database workloads that need co-located object storage for backups and exports
*   •Government and enterprise applications leveraging Oracle's compliance certifications

#### Choose This When

When your monthly egress is under 10 TB and you want it free without switching to Cloudflare R2, or when you need Oracle Cloud integration for database workloads.

#### Skip This If

When your egress exceeds 10 TB/mo (R2 becomes cheaper), when you need polished documentation and tooling, or when community ecosystem matters.

#### Integration Example

import boto3

oci\_s3 = boto3.client(
    "s3",
    endpoint\_url="https://NAMESPACE.compat.objectstorage.us-ashburn-1.oraclecloud.com",
    aws\_access\_key\_id="OCI\_ACCESS\_KEY",
    aws\_secret\_access\_key="OCI\_SECRET\_KEY",
)

# Upload: first 10 TB/mo of downloads are free
oci\_s3.upload\_file("/tmp/media.mp4", "content-delivery", "videos/promo.mp4")

# List objects across tiers
response = oci\_s3.list\_objects\_v2(Bucket="content-delivery", Prefix="videos/")
for obj in response.get("Contents", \[\]):
    print(f"{obj\['Key'\]}: {obj\['Size'\] / 1e6,.1f} MB")

$25.50/TB/mo (Standard); $8.50/TB egress ($0.0085/GB) after 10 TB free; $0.40/10K requests

Best for: Teams that need substantial free egress (up to 10 TB/mo) but can't commit to Cloudflare's ecosystem for R2

[Visit Website](https://oracle.com/cloud/storage)

10

### DigitalOcean Spaces

Simple, developer-friendly S3-compatible storage with a built-in CDN: but max object size is 5 GB, not 5 TB. This is the gotcha that catches most teams. For small files (web assets, config, documents) Spaces works great. For video, backups, or ML checkpoints, you'll hit the wall fast.

What Sets It Apart

The simplest all-in-one package for small teams: S3 storage plus CDN plus developer-friendly docs in a single $5/mo plan. No other provider bundles CDN at this price.

Use with MVS

Use DigitalOcean Spaces for small MVS namespaces attached to DigitalOcean-hosted apps where source objects are small, setup speed matters, and the included CDN is useful.

#### Strengths

*   +Simple pricing: $5/mo gets 250 GB + 1 TB egress + built-in CDN
*   +Developer-friendly with good documentation
*   +Full S3 API compatibility for standard operations
*   +Built-in CDN included at no extra cost

#### Limitations

*   \-Max object size is 5 GB: not 5 TB like nearly every other provider
*   \-No object lock or WORM compliance
*   \-$5/mo minimum charge regardless of usage
*   \-Rate limited: 750 GET/s and 150 PUT/s per IP

#### Real-World Use Cases

*   •Startup web apps storing user-uploaded images, documents, and profile photos under 5 GB each
*   •Static asset hosting for websites with the built-in CDN for global delivery
*   •Configuration and artifact storage for CI/CD pipelines on DigitalOcean infrastructure
*   •Small SaaS applications that need simple, affordable S3-compatible storage with predictable costs

#### Choose This When

When you are a small team or startup that needs simple, cheap S3 storage with built-in CDN for web assets and small files.

#### Skip This If

When you need to store objects larger than 5 GB, require object lock or WORM compliance, or expect high request rates above 750 GET/s.

#### Integration Example

import boto3

spaces = boto3.client(
    "s3",
    endpoint\_url="https://nyc3.digitaloceanspaces.com",
    aws\_access\_key\_id="DO\_SPACES\_KEY",
    aws\_secret\_access\_key="DO\_SPACES\_SECRET",
)

# Upload web assets: CDN included automatically
spaces.upload\_file(
    "/tmp/logo.png",
    "my-app-assets",
    "images/logo.png",
    ExtraArgs={"ACL": "public-read", "ContentType": "image/png"},
)

# Access via CDN: https://my-app-assets.nyc3.cdn.digitaloceanspaces.com/images/logo.png

$5/mo base (250 GB + 1 TB transfer); $20/TB/mo additional storage; $10/TB egress

Best for: Small teams and startups storing web assets, documents, and small files: not suitable for large objects like video or model weights

[Visit Website](https://digitalocean.com/products/spaces)

11

### Linode (Akamai) Object Storage

S3-compatible object storage from Akamai (formerly Linode) with a $5/mo flat rate for 250 GB and strong integration with Linode compute instances. Backed by Akamai's global CDN network. The 5 GB max object size limits large-file use cases, but for small-to-medium workloads it is developer-friendly with good docs.

What Sets It Apart

Backed by Akamai's global CDN infrastructure: the strongest edge delivery network of any independent storage provider.

Use with MVS

Use Linode Object Storage with MVS for Akamai/Linode-hosted applications that need co-located vector storage for small-to-medium media and document search workloads.

#### Strengths

*   +Flat $5/mo for 250 GB storage + 1 TB outbound transfer
*   +Backed by Akamai's global CDN and edge network
*   +Full S3 API compatibility for standard operations
*   +11 regions across NA, EU, and APAC

#### Limitations

*   \-5 GB max object size: same limitation as DigitalOcean Spaces
*   \-No versioning or object lock
*   \-Rate-limited at 750 requests/second per bucket
*   \-$20/TB/mo additional storage: competitive but not cheapest

#### Real-World Use Cases

*   •CDN origin storage leveraging Akamai's edge network for global content delivery
*   •Application asset storage co-located with Linode compute instances for low-latency access
*   •Backup storage for Linode-hosted applications with straightforward pricing

#### Choose This When

When you run workloads on Linode/Akamai and want co-located storage with access to Akamai's CDN for content delivery.

#### Skip This If

When you need large object support (>5 GB), versioning, object lock, or storage pricing below $5/mo minimum.

#### Integration Example

import boto3

linode = boto3.client(
    "s3",
    endpoint\_url="https://us-east-1.linodeobjects.com",
    aws\_access\_key\_id="LINODE\_KEY",
    aws\_secret\_access\_key="LINODE\_SECRET",
)

# Upload assets: Akamai CDN available for delivery
linode.upload\_file("/tmp/app-bundle.js", "static-assets", "js/app.bundle.js")

# Generate presigned download URL
url = linode.generate\_presigned\_url(
    "get\_object",
    Params={"Bucket": "static-assets", "Key": "js/app.bundle.js"},
    ExpiresIn=86400,
)

$5/mo base (250 GB + 1 TB transfer); $20/TB/mo additional storage; $10/TB additional egress

Best for: Teams on Linode/Akamai infrastructure that want co-located S3 storage with CDN integration

[Visit Website](https://linode.com/products/object-storage)

12

### OVHcloud Object Storage

European cloud provider with S3-compatible storage and a strong GDPR compliance story. OVHcloud dropped egress fees in January 2026, making it a zero-egress option for EU teams: but with a 30-day minimum retention on all tiers that can catch you off guard. Three storage classes (Standard, IA, Archive) with pricing competitive with B2.

What Sets It Apart

Zero-egress EU-sovereign storage with three pricing tiers: the only European provider combining free egress with archive-class pricing at $1.50/TB/mo.

Use with MVS

Use OVHcloud Object Storage behind MVS for EU-sovereign vector namespaces that benefit from zero egress and lifecycle tiers without depending on Cloudflare or US hyperscalers.

#### Strengths

*   +Zero egress fees as of January 2026
*   +Three storage classes: Standard ($7/TB), IA ($4/TB), Archive ($1.5/TB)
*   +Strong EU data sovereignty: French-headquartered, EU-only infrastructure
*   +S3-compatible with versioning and lifecycle policies

#### Limitations

*   \-30-day minimum retention on ALL tiers: shorter than Wasabi but still a gotcha
*   \-Smaller S3 ecosystem support than AWS or Cloudflare
*   \-Console and API documentation quality trails hyperscalers
*   \-Limited regions compared to AWS or Oracle

#### Real-World Use Cases

*   •EU-sovereign data storage for organizations subject to French or EU data residency requirements
*   •Tiered archival using Standard, IA, and Archive classes for lifecycle-managed datasets
*   •Zero-egress content delivery for EU-focused applications without Cloudflare dependency
*   •Backup storage with competitive archive pricing at $1.50/TB/mo

#### Choose This When

When you need zero-egress storage with EU data sovereignty and want lifecycle tiering down to $1.50/TB/mo for archived data.

#### Skip This If

When your data has churn within 30 days (minimum retention applies), when you need a large S3 ecosystem, or when you need regions outside Europe.

#### Integration Example

import boto3

ovh = boto3.client(
    "s3",
    endpoint\_url="https://s3.gra.io.cloud.ovh.net",
    aws\_access\_key\_id="OVH\_ACCESS\_KEY",
    aws\_secret\_access\_key="OVH\_SECRET\_KEY",
)

# Upload to Standard tier: zero egress on downloads
ovh.upload\_file("/tmp/dataset.parquet", "eu-data", "datasets/v2.parquet")

# Set lifecycle rule to move to IA after 30 days
ovh.put\_bucket\_lifecycle\_configuration(
    Bucket="eu-data",
    LifecycleConfiguration={
        "Rules": \[{
            "ID": "archive-old",
            "Status": "Enabled",
            "Transitions": \[{"Days": 30, "StorageClass": "STANDARD\_IA"}\],
        }\]
    },
)

$7/TB/mo (Standard); $4/TB/mo (IA); $1.50/TB/mo (Archive); zero egress; 30-day minimum retention

Best for: EU teams wanting zero-egress storage with multiple tiers and strong data sovereignty guarantees

[Visit Website](https://ovhcloud.com/en/public-cloud/object-storage)

13

### Storj

Decentralized object storage across 30,000+ independent nodes in 100+ countries. Data is encrypted client-side, erasure-coded into 80+ pieces, and distributed globally. At $7/TB/mo storage and $7/TB egress, it is the cheapest centrally-managed storage alternative. The trade-off is higher tail latency and lower throughput than centralized providers.

What Sets It Apart

Decentralized architecture across 30,000+ independent nodes with mandatory client-side encryption: no single entity (including Storj) can access your data.

Use with MVS

Use Storj for archival or cold MVS namespaces where encrypted, distributed storage is valuable and higher tail latency is acceptable for rarely queried vectors.

#### Strengths

*   +Just $7/TB/mo storage: cheapest alongside IDrive e2
*   +End-to-end encryption: data encrypted before leaving your machine
*   +Decentralized across 30,000+ nodes in 100+ countries
*   +S3-compatible gateway with standard SDK support

#### Limitations

*   \-Higher tail latency than centralized providers (P99 can spike)
*   \-Throughput limited by distributed retrieval across many nodes
*   \-Some S3 features missing (no versioning, limited lifecycle)
*   \-Smaller enterprise adoption and fewer compliance certifications

#### Real-World Use Cases

*   •Encrypted backup storage for organizations that need client-side encryption without managing key infrastructure
*   •Archival of large datasets where $7/TB/mo pricing enables petabyte-scale storage on a budget
*   •Distributing open-source datasets with built-in geographic redundancy across 100+ countries

#### Choose This When

When you prioritize encryption, geographic redundancy, and cost over access speed: ideal for backups, archives, and cold storage at petabyte scale.

#### Skip This If

When you need low-latency access, high throughput, enterprise SLAs, or advanced S3 features like versioning and lifecycle policies.

#### Integration Example

import boto3

storj = boto3.client(
    "s3",
    endpoint\_url="https://gateway.storjshare.io",
    aws\_access\_key\_id="STORJ\_KEY",
    aws\_secret\_access\_key="STORJ\_SECRET",
)

# Upload: client-side encrypted, erasure-coded across 80+ nodes
storj.upload\_file("/tmp/archive.tar.gz", "cold-storage", "backups/2026-q1.tar.gz")

# Verify distributed storage
head = storj.head\_object(Bucket="cold-storage", Key="backups/2026-q1.tar.gz")
print(f"Stored: {head\['ContentLength'\] / 1e9:.1f} GB (encrypted, 80+ node redundancy)")

$7/TB/mo storage; $7/TB egress; $5/mo account minimum; 25 GB free on a 30-day trial

Best for: Large-scale cold storage and backup where cost and encryption matter more than access speed

[Visit Website](https://storj.io)

14

### Azure Blob Storage

Microsoft's object storage with deep Azure ecosystem integration: Synapse Analytics, Azure ML, Cognitive Services, and Data Lake Storage Gen2. The S3 compatibility endpoint is still in preview and has notable gaps. If you are on Azure, Blob Storage is the natural choice. If you are not, there is no reason to start here.

What Sets It Apart

Deepest integration with Microsoft's analytics and ML stack: Synapse, Azure ML, Data Lake Gen2, and Cognitive Services all work natively with Blob Storage.

Use with MVS

Use Azure Blob Storage with MVS for Azure-native vector workloads where source data, ML jobs, compliance controls, and analytics already live in the Microsoft cloud.

#### Strengths

*   +Deep Azure ecosystem: Synapse, Azure ML, Cognitive Services, Data Factory
*   +Data Lake Storage Gen2 for hierarchical namespace and analytics workloads
*   +Immutable storage (WORM) with legal hold and time-based retention
*   +Archive tier at $1/TB/mo: cheapest of the big three

#### Limitations

*   \-S3 compatibility endpoint is in preview: gaps in multipart upload and presigned URLs
*   \-Egress at $0.087/GB: expensive for read-heavy workloads
*   \-Azure IAM (RBAC + SAS tokens) is more complex than S3 IAM
*   \-$18/TB/mo standard hot storage

#### Real-World Use Cases

*   •Azure Synapse Analytics pipelines reading and writing data to Blob Storage natively
*   •Data Lake Storage Gen2 for hierarchical namespace analytics with Spark and Databricks
*   •Immutable compliance storage using legal hold and time-based retention policies
*   •Azure ML training workflows with native Blob Storage dataset integration

#### Choose This When

When your infrastructure runs on Azure and you need native integration with Synapse Analytics, Azure ML, or Data Lake Storage Gen2.

#### Skip This If

When you need full S3 compatibility (the S3 endpoint is still in preview), when egress costs are a concern, or when you are not already on Azure.

#### Integration Example

from azure.storage.blob import BlobServiceClient

client = BlobServiceClient.from\_connection\_string("AZURE\_CONN\_STRING")
container = client.get\_container\_client("ml-data")

# Upload training data
with open("/tmp/train.parquet", "rb") as f:
    container.upload\_blob("datasets/train.parquet", f, overwrite=True)

# Set immutable policy for compliance
from azure.storage.blob import ImmutabilityPolicy
container.get\_blob\_client("datasets/train.parquet").set\_immutability\_policy(
    ImmutabilityPolicy(expiry\_time="2029-01-01", policy\_mode="Unlocked")
)

$18/TB/mo (Hot); $10/TB/mo (Cool); $1/TB/mo (Archive); $0.087/GB egress

Best for: Teams on Azure that need tight integration with Synapse, Azure ML, or Data Lake Storage Gen2

[Visit Website](https://azure.microsoft.com/en-us/products/storage/blobs)

15

### STRG

S3-compatible object storage at a flat $6.80/TB/mo that includes all egress and standard API requests, with no retrieval fees and no minimum storage period. Two regions, US-East (Reston, Virginia) and EU-Central (Amsterdam), with optional cross-region replication, Object Lock in governance and compliance modes, and access keys you can scope to some buckets and a prefix. A newer provider, so it has a shorter track record than B2 or Wasabi. STRG lists Mixpeek on its recommended tools page; no payment was made for either listing.

What Sets It Apart

Egress and standard API requests are part of the per-TB price, with no minimum storage period.

Use with MVS

Use STRG behind MVS when vector payloads and source media are read back often, since egress is included in the storage price.

#### Strengths

*   +One price per TB with every download and standard API call included
*   +No minimum storage period or retrieval fees, so data can churn freely
*   +Object Lock with governance mode, compliance mode and legal holds
*   +Access keys scoped to specific buckets and an object prefix

#### Limitations

*   \-Only two regions (US-East and EU-Central)
*   \-Newer service with little public production track record
*   \-Durability is a design target (11 nines); contractual service levels are agreed separately
*   \-A card is required at signup, checked with a $1 hold

#### Real-World Use Cases

*   •Media libraries and product assets that are downloaded far more often than they are written
*   •AI training sets read repeatedly by training jobs, where per-GB egress would dominate the bill
*   •Backups with Object Lock retention for compliance, in a US or EU region

#### Choose This When

When downloads are a large share of your usage and you want the bill to depend on stored data alone.

#### Skip This If

When you need regions outside the US and EU, a provider with a long track record, or a contractual SLA without a separate agreement.

#### Integration Example

import boto3

strg = boto3.client(
    "s3",
    endpoint\_url="https://s3.us-east-1.strg.com",
    aws\_access\_key\_id="STRG\_ACCESS\_KEY",
    aws\_secret\_access\_key="STRG\_SECRET\_KEY",
)

# Every GET is included in the storage price
strg.download\_file("media", "videos/launch.mp4", "/tmp/launch.mp4")

$6.80/TB/mo storage with all egress and standard API requests included; first 10 GB free; no minimum storage period or retrieval fees (strg.com, checked 2026-11-18)

Best for: Teams that serve or re-read their data often and want one predictable per-TB bill with egress included

[Visit Website](https://strg.com)

16

### IDrive e2

Budget S3-compatible storage at $6/TB/mo with no minimum retention, fixing Wasabi's biggest gotcha. Free egress under a reasonable-use policy. Less known than B2 or Wasabi, but delivers solid S3 compatibility at rock-bottom prices for teams that prioritize cost above all else.

What Sets It Apart

The cheapest S3 storage with zero minimum retention: the only sub-$5/TB provider where you can delete objects the same day without penalty.

Use with MVS

Use IDrive e2 for experimental or staging MVS namespaces where embeddings and payloads churn frequently and avoiding minimum-retention penalties matters.

#### Strengths

*   +Just $6/TB/mo: tied with Storj for cheapest storage
*   +No minimum retention period: delete anytime without penalty
*   +Free egress under reasonable-use policy
*   +10 regions across US, EU, and APAC

#### Limitations

*   \-Reasonable-use egress policy limits heavy download workloads
*   \-Smaller brand with less community support and fewer integrations
*   \-Documentation less thorough than major providers
*   \-No event notifications, versioning, or object lock

#### Real-World Use Cases

*   •Iterative development storage where datasets are frequently replaced without early-deletion penalties
*   •Budget backup storage for small businesses that cannot afford $15-23/TB/mo
*   •Staging environments that mirror production storage at minimal cost

#### Choose This When

When cost is your primary driver and you need the flexibility to delete or replace data frequently without retention penalties.

#### Skip This If

When you need enterprise SLAs, rich documentation, event notifications, or advanced S3 features like versioning and object lock.

#### Integration Example

import boto3

idrive = boto3.client(
    "s3",
    endpoint\_url="https://e2.idrivee2.com",
    aws\_access\_key\_id="IDRIVE\_KEY",
    aws\_secret\_access\_key="IDRIVE\_SECRET",
)

# Upload and delete freely: no minimum retention
idrive.upload\_file("/tmp/experiment.parquet", "staging", "v42/data.parquet")

# Replace previous version without penalty
idrive.delete\_object(Bucket="staging", Key="v41/data.parquet")
idrive.upload\_file("/tmp/experiment\_v2.parquet", "staging", "v42/data.parquet")

$6/TB/mo storage ($6/mo minimum); egress free up to 3x active storage, then $0.01/GB; no minimum retention; no API fees

Best for: Teams that want Wasabi-level pricing without the 90-day retention trap: ideal for iterative workloads with data churn

[Visit Website](https://idrive.com/e2)

17

### Rabata

Simple, flat-rate S3-compatible storage at $10/TB/mo with zero API fees: no per-GET, per-PUT, or per-DELETE charges. EU and US regions with a straightforward pricing model: $0.01/GB for both storage and egress. The Backup tier offers bulk pricing at $49/10TB/mo for cold data. Newer provider with less track record but competitive on simplicity.

What Sets It Apart

The only S3-compatible provider with truly zero API fees: no per-request charges on any operation, making it ideal for high-frequency read/write workloads.

Use with MVS

Use Rabata for MVS namespaces with high object-operation volume where predictable request pricing is more important than zero egress.

#### Strengths

*   +Zero API fees: no per-request charges for GET, PUT, or DELETE
*   +Simple pricing: $0.01/GB for both storage and egress, no hidden dimensions
*   +EU (Finland, Netherlands) and US regions available
*   +Full S3 API compatibility with multipart uploads and presigned URLs

#### Limitations

*   \-No free egress tier: all egress charged at $0.01/GB
*   \-Backup tier billed in 10 TB increments: not suitable for small cold datasets
*   \-Newer provider with limited production track record
*   \-Object lock and versioning status not publicly documented

#### Real-World Use Cases

*   •High-request-volume applications where per-GET and per-PUT fees on other providers add up to significant cost
*   •EU-compliant data storage with both Nordic (Finland) and Western European (Netherlands) region options
*   •Bulk backup and archival using the $49/10TB/mo Backup tier for cold data at scale

#### Choose This When

When your workload makes millions of API calls and per-request fees are a meaningful cost line. Also when you want dead-simple pricing with no hidden dimensions.

#### Skip This If

When you need zero egress fees (use R2 or Tigris instead), require documented object lock/versioning support, or need a provider with a long production track record.

#### Integration Example

import boto3

rabata = boto3.client(
    "s3",
    endpoint\_url="https://s3.rabata.io",
    aws\_access\_key\_id="RABATA\_KEY",
    aws\_secret\_access\_key="RABATA\_SECRET",
)

# Upload with zero API fees: no per-PUT charge
rabata.upload\_file("/tmp/dataset.parquet", "data-lake", "2026/q1/dataset.parquet")

# High-frequency reads with no per-GET fees
for i in range(10000):
    rabata.get\_object(Bucket="data-lake", Key="2026/q1/dataset.parquet")

$10/TB/mo storage; $10/TB egress; $0 API fees; Backup tier $49/10TB/mo

Best for: Teams that want predictable pricing without per-request fees: especially workloads with high API call volume where GET/PUT fees add up

[Visit Website](https://rabata.io)

## Which one fits your use case?

Find your situation in the first column. The last column says when the pick is the wrong call.

Choose it when

Pick

Look elsewhere when

Egress cost is your primary concern: content delivery, API serving, or any workload that reads far more than it writes. R2 also wins when you want edge compute via Workers.

[Cloudflare R2](#item-1)

You need versioning, object lock, or WORM compliance for regulatory requirements. Also avoid if your workload is storage-heavy but read-light, since $15/TB/mo storage is expensive compared to B2 or Wasabi.

You need the cheapest reliable storage at scale and deliver content via Cloudflare or Fastly CDN. Also ideal for very large objects (up to 10 TB per object).

[Backblaze B2](#item-2)

You need multi-region availability, storage tiering (archive/IA classes), or more than 100 buckets per account.

You need zero egress fees but cannot compromise on versioning, object lock, or WORM compliance. Also ideal for globally distributed workloads that benefit from automatic edge caching.

[Tigris](#item-3)

You need a battle-tested provider with a long track record: Tigris is newer and has a smaller community. Also avoid if $20/TB/mo storage is too expensive for your volume.

You are already invested in AWS and need features that only S3 offers: Intelligent Tiering, EventBridge, Lambda@Edge, or S3 Vectors. The ecosystem premium pays for itself when you use adjacent services.

[AWS S3](#item-4)

Cost matters more than ecosystem: S3 is 4x more expensive than B2 for storage and egress creates financial lock-in at scale. Moving 100 TB out costs $9,000.

You are EU-based, need GDPR-compliant data residency, and want full S3 features (versioning, object lock) at the lowest price available.

[Hetzner Object Storage](#item-5)

You need storage outside Europe, a global CDN, or the rich ecosystem of adjacent services that hyperscalers provide.

You store write-once data that will live longer than 90 days (backups, archives, compliance records) and want the simplest possible pricing with no egress or API fees.

[Wasabi](#item-6)

Your data has any churn: frequent updates, overwrites, or deletions within 90 days will trigger minimum retention charges. Also avoid if you need event notifications or lifecycle policies.

Data must stay on your infrastructure (air-gap, compliance, sovereignty), you have the ops team to manage it, and you need full S3 feature coverage including event notifications.

[MinIO (Self-Hosted)](#item-7)

You do not have dedicated infrastructure or Kubernetes expertise. Also avoid at small scale where hardware amortization makes cloud providers cheaper.

Your infrastructure is on Google Cloud and you need native integration with Vertex AI, BigQuery, or Dataflow. Autoclass is also valuable for workloads with unpredictable access patterns.

[Google Cloud Storage](#item-8)

Egress costs matter: GCS has the highest egress fees of any major provider at $0.12/GB. Also avoid if you need native S3 compatibility (GCS interop API has gaps).

## Frequently Asked Questions

### Can I run a vector database directly on object storage?

Yes, and it is the cheapest architecture for storage-heavy retrieval: engines like Mixpeek MVS and turbopuffer keep vectors at rest on S3-class storage (around $0.02-0.33/GB-month) and cache hot data for query speed, instead of holding everything in RAM. The cost math and vendor comparison are in [How Much Does a Vector Database Cost?](/guides/vector-database-cost-comparison).

### What is the cheapest S3-compatible object storage?

On raw storage price IDrive e2 is the cheapest at $6/TB/mo, with Backblaze B2 at $6.95/TB/mo and OVHcloud and Storj both at $7/TB/mo close behind. Wasabi is $7.99/TB/mo flat with no egress or API fees, though its 90-day minimum retention per object can double the effective cost on high-churn data. Hetzner is EUR 6.26/TB/mo with versioning, object lock and EU data residency. Sticker price rarely decides the bill though: Cloudflare R2 is $15/TB/mo and charges nothing to read your data out, which beats every provider above it once you download a meaningful fraction of what you store.

### Which object storage has zero egress fees?

Cloudflare R2 offers truly unlimited zero egress with no asterisks. Tigris and Fastly Object Storage also offer zero egress within their networks. Wasabi claims 'free' egress but enforces a reasonable-use policy (monthly egress cannot exceed stored volume). OVHcloud dropped egress fees in January 2026 but has a 30-day minimum retention on all tiers. Always read the fine print.

### Is S3 compatibility the same across all providers?

No: 'S3 compatible' is a spectrum. AWS S3 is the reference implementation. MinIO has the most complete open-source implementation. Cloudflare R2 supports core operations but lacks versioning and object lock. Azure Blob's S3 endpoint is still in preview. DigitalOcean, Vultr, and Linode cap objects at 5 GB. Event notifications are effectively AWS/GCS/MinIO only. Test your specific workload before migrating.

### What does it cost to migrate off a cloud storage provider?

Egress fees make leaving expensive: moving 100 TB off AWS S3 costs $9,000, off GCS costs $12,000, off Azure costs $8,700. Moving off Cloudflare R2 costs $0. This 'escape cost' is the most important number nobody compares when choosing a provider. Use our interactive calculator at storage.mixpeek.com to model your specific scenario.

### What is the Wasabi 90-day minimum retention policy?

Wasabi charges a minimum of 90 days for every object stored. If you upload a file and delete it after 30 days, you still pay for the remaining 60 days. This applies per-object, not per-account. On datasets with high churn (frequent updates or deletions), this can double your effective storage cost. Wasabi is best for write-once data like backups and archives that will stay for at least 90 days.

### Can I search and query objects stored in S3-compatible storage?

Yes. AWS offers S3 Select for basic SQL queries and the new S3 Vectors for vector search. For multimodal search across images, video, documents, and audio in any S3-compatible bucket, Mixpeek connects to your storage and adds feature extraction, embedding, and semantic search without moving your data. See mixpeek.com for details.

## See how Mixpeek handles this

Purpose-built for s3-compatible object storage providers, not bolted on.

### Talk to a Mixpeek engineer: free

30 minutes. Bring your use case and we'll tell you exactly what would work and what wouldn't.

[Schedule a Free Call](https://calendly.com/mixpeek/30min?hide_gdpr_banner=1&background_color=ffffff&text_color=1a1a2e&primary_color=6366f1&utm_source=curated_list_modal)

## Explore Other Curated Lists

infrastructure

### Best Embedding Visualization Tools

How do you actually look at millions of embeddings? We compared the leading tools for visualizing and exploring embedding spaces on scale, interactivity, projection quality (UMAP/t-SNE/PCA), and how well they connect points back to the underlying content.

7 tools ranked[View List](/curated-lists/best-embedding-visualization-tools)

infrastructure

### Best RAG Frameworks & Platforms

A practical comparison of frameworks and platforms for building Retrieval-Augmented Generation applications. We tested each on multi-document QA, citation accuracy, and multimodal RAG scenarios.

10 tools ranked[View List](/curated-lists/best-rag-frameworks)

infrastructure

### Best Open Source ML Pipeline Frameworks

A comparison of open-source frameworks for building, orchestrating, and deploying machine learning pipelines. Evaluated on flexibility, scalability, and production readiness for real-world ML workflows.

10 tools ranked[View List](/curated-lists/best-open-source-ml-pipelines)