What if the storage bill you pay every month has almost nothing to do with how much data you keep? That is the strange reality of cloud object storage. The headline price per gigabyte looks tiny, often a cent or two. Then the invoice arrives with retrieval charges, egress fees, minimum storage durations, and request costs stacked on top. Choosing between providers is less about the sticker price and more about understanding how each one charges you after the data lands.
This comparison walks through the major providers, the storage tiers they offer, and the traps that catch teams in their first year. It focuses on what the model actually is, where providers genuinely differ, and how to pick one without regretting it later.
What object storage actually is, and why it is not a drive
Object storage keeps data as discrete objects, each with a key, the data itself, and metadata. You do not mount it as a drive and you do not edit a file in place. There are no real folders, only prefixes that look like folders. If your application expects file locking or random partial writes, object storage will fight you.
The model dates back to March 2006, when Amazon launched S3 as the first widely available cloud object storage service. It solved a specific problem: storing enormous quantities of unstructured data without a filesystem’s overhead. That is why it became the default home for backups, media libraries, logs, data lake files, and static website assets.
The trade-off is simple. You give up filesystem semantics and get near-limitless scale, high durability targets, and pay-per-use pricing. Most teams accept that trade without thinking twice. The trouble starts when they assume the pricing is equally simple.
The big three, side by side

Amazon S3, Azure Blob Storage, and Google Cloud Storage dominate the market. They share the same basic shape but differ in tier names, minimum durations, and retrieval behavior. Here is the rough picture, with the caveat that pricing and tier details change often and should be checked against current documentation before you commit.
| Feature | Amazon S3 | Azure Blob Storage | Google Cloud Storage |
|---|---|---|---|
| Hot tier | S3 Standard | Hot | Standard |
| Warm tier | S3 Standard-IA | Cool | Nearline |
| Cold tier | S3 Glacier Instant Retrieval | Cold | Coldline |
| Archive tier | S3 Glacier Deep Archive | Archive | Archive |
| Typical minimum duration | 30 to 180 days by class | 30 to 180 days by tier | 30, 90, or 365 days by class |
| Retrieval speed (archive) | Minutes to hours | Hours | Hours |
| Consistency | Strong read-after-write since December 2020 | Strong | Strong |
The consistency row matters more than people realize. Before December 2020, S3 offered eventual consistency for overwrites and deletes, which meant a newly written object might not appear immediately. S3 moved to strong read-after-write consistency for all new object writes and overwrites that month. Azure and Google had already moved in the same direction. If you find an old tutorial warning about “S3 eventual consistency,” it is out of date.
Durability claims are another place where the marketing outruns the evidence. All three publish an eleven-nines figure, meaning 99.999999999 percent durability. That number is a design target, not an independently audited guarantee. It is a reasonable signal, but treat it as a vendor claim rather than a certificate.
Where the money actually goes
The per-gigabyte storage rate is the least interesting number on the invoice. Four other costs do the damage.
Egress fees are the big one. Moving data out of a cloud provider costs money beyond a free monthly allowance, and the rate varies by provider and region. Teams that build a service reading data constantly from storage can find egress dwarfing storage costs. Some competitors, including Cloudflare R2, advertise zero egress fees specifically to attack this pain point.
Retrieval fees apply to colder tiers. Archive storage is cheap to hold and expensive to read. If you archive data and then need it urgently, you pay both a retrieval fee and a waiting period. S3 Glacier Deep Archive, introduced in 2019, is built for data you expect to touch almost never.
Minimum storage durations punish short-lived data. Google Cloud Storage, for example, applies minimum durations of 30, 90, or 365 days depending on the class. Delete or move an object early and you are billed as if it stayed. A team that dumps temporary files into a cold tier and cleans up weekly will pay for months of storage it never used.
Request costs seem trivial until you are serving millions of small objects. Per-request pricing adds up fast for workloads with many tiny reads.
The practical lesson: model your access pattern before choosing a tier, not after. If you know roughly how often data is read and how long it lives, the right tier usually picks itself.
Products that support the S3 API
One of the quieter shifts in this market is that S3’s API became a de facto standard. Many third-party providers, including Cloudflare R2, Backblaze B2, and Wasabi, expose S3-compatible APIs. That means tools written for S3 often work against them with a configuration change.
This is genuinely useful for avoiding lock-in and for multi-cloud setups. But “S3-compatible” is not a standardized guarantee. Feature coverage varies by provider. A tool that relies on a specific S3 feature, such as object lock, inventory reports, or a particular storage class, may fail silently or behave differently elsewhere. Test the specific features you depend on rather than trusting the compatibility label.
The same caution applies to migration tooling. Moving data between providers is straightforward in principle and tedious in practice, especially when metadata, permissions, and lifecycle rules need to come along.
Practical tips that save real money
Experience from teams running object storage at scale produces a few consistent lessons.
Set lifecycle rules on day one. Objects that nobody sets an expiry on live forever. A rule that moves data to a colder tier after 30 days and deletes it after a year costs nothing to configure and often cuts storage spend noticeably.
Watch object size. Many small objects cost more in requests than in storage. Bundling small files into larger archives, where the workload allows it, reduces request overhead.
Check the free egress allowance. Every provider includes some free transfer out per month. Workloads that stay under it avoid the most common bill shock.
Do not archive what you might need tomorrow. Archive tiers are for compliance copies, old logs, and media masters. Putting active data there creates retrieval delays that no one budgets for.
Tag everything. Cost allocation by team or project is nearly impossible after the fact if objects were never tagged. It takes minutes to enforce at write time and saves hours of forensic accounting later.
The same discipline shows up in adjacent tool categories. Teams comparing storage costs often end up doing the same exercise for the software around it, whether that is a social media management tools pricing comparison for teams or an AI translation tools pricing comparison of what you actually pay. The pattern is identical: the advertised rate hides the real cost.
Compliance, residency, and the regional question
Data residency is a genuine differentiator, not a marketing checkbox. Providers offer region-specific storage, and some publish broader commitments about where data is stored and processed. Azure, for instance, has published an EU Data Boundary commitment covering storage and processing within the EU for covered services.
The details differ by provider, region, and service, and they change. If residency is a hard requirement, verify against the provider’s current compliance documentation rather than a blog post from three years ago. The same goes for certifications. A provider may hold a certification globally but not in the specific region you chose.
How to choose without overthinking it
For most teams, the decision comes down to a short list of questions.
- Where does the data live, and does that region satisfy residency rules?
- How often is the data read, and does that pattern fit a cold tier?
- How much data leaves the provider each month, and what does that cost?
- Which specific features does the workload need, and does the candidate support them?
- What is the realistic total bill at current volume, including requests and retrieval?
If two providers come out close, the tiebreaker is usually operational familiarity. A team that already runs on one cloud gains little from switching to save a few percent on storage. Migration costs, retraining, and the risk of subtle behavioral differences usually outweigh the savings.
The broader tooling question deserves the same scrutiny. A content audit tools pricing comparison of what you really pay follows the same logic as storage: list price is a starting point, not an answer. So does an AI presentation tools pricing comparison of what you actually pay. The vendors that win long term are rarely the cheapest on the rate card.
Conclusion
Cloud object storage is one of the least glamorous and most consequential infrastructure choices a team makes. The model itself is settled: objects, keys, prefixes, no filesystem semantics, near-limitless scale. What separates providers is not the core capability but the cost structure around it, especially egress, retrieval, and minimum durations.
The big three remain the safe default for most workloads, with tier names that differ but behavior that rhymes. The S3-compatible challengers are worth a serious look for egress-heavy workloads, provided you test the specific features you need. Whatever you pick, set lifecycle rules early, model your access pattern honestly, and read the bill before you sign. The cheapest headline rate is rarely the cheapest outcome.
Frequently asked questions
What is the difference between object storage and block storage?
Object storage keeps whole objects addressed by a key, with metadata attached, and is designed for massive scale and unstructured data. Block storage presents raw volumes that a filesystem formats and mounts, which suits databases and virtual machines. Object storage cannot be edited in place; block storage can.
Is S3 cheaper than Azure Blob or Google Cloud Storage?
Not reliably. Base storage rates are close enough that egress, retrieval, and request costs usually decide the winner. A workload that reads data constantly may cost far more on one provider than another despite a similar per-gigabyte rate. Model the full bill, not the storage line.
What does eleven nines of durability actually mean?
It means the provider designs for a probability of losing an object of roughly one in a hundred billion per year. It is a published design target, not an independently audited result. In practice, the bigger risks are accidental deletion and misconfigured permissions, which object lock and versioning address.
Can I move data between object storage providers easily?
Moving the bytes is usually straightforward; moving everything around them is not. Metadata, permissions, lifecycle rules, and application logic that depends on provider-specific features all need attention. S3-compatible APIs reduce the friction but do not eliminate it, because compatibility coverage varies.
Why do egress fees matter so much?
Because they scale with usage rather than storage volume. A service that reads data frequently pays egress on every transfer out, and those charges can exceed storage costs many times over. Providers with zero egress fees target exactly this problem, which is why the fee deserves its own line in any comparison.
Should archived data go straight to the cheapest tier?
Only if you are confident it will not be needed soon. Archive tiers carry retrieval fees and waiting times measured in hours, and cold classes often impose minimum storage durations of 90 days or a year. Deleting early still gets billed. Match the tier to the access pattern, not to the lowest rate on the page.