Cloud Cost Comparison

Compare cloud storage costs across AWS, Azure, Google Cloud, Oracle and Alibaba

Cloud storage pricing looks like it should be a single number per gigabyte, and that is exactly why storage bills surprise people. This tool models a storage workload the way a bill is actually assembled — capacity, requests, retrieval and egress — and shows what the same workload costs on five providers side by side. Describe the workload once, and you get a monthly and annual figure per provider, a line-item breakdown, and the gap between the cheapest and the most expensive option.

What the tool covers

Six workload types are modelled, each on its own tab, and every one is compared across AWS, Microsoft Azure, Google Cloud, Oracle Cloud and Alibaba Cloud.

Four regions are offered — US East, US West, EU West and Asia Pacific — and each maps to the equivalent region on every provider, so a US East comparison lines up us-east-1, eastus, us-east1, us-ashburn-1 and Alibaba's us-east-1 rather than comparing a US price against a European one.

Comparing object storage like for like

The trap in storage comparison is that the four providers do not name the same thing the same way, and their default tier is not the same product. A like-for-like comparison means fixing five things before you look at any price:

• Access tier. Hot, cool, archive and deep archive exist everywhere under different names. Comparing one provider's hot tier against another's cool tier tells you nothing. The tool normalises this to four classes — standard, infrequent access, glacier and deep archive — and applies the matching tier on every provider.
• Redundancy. Single-zone, regionally redundant and geo-redundant are three different products at three different prices. Pick the one that matches your durability requirement, not the one that makes the number look best.
• Request volume. Reads and writes are priced separately and writes usually cost more per operation. You need both figures.
• Egress. How many gigabytes actually leave the provider's network each month, as opposed to how much you store.
• Retrieval. On cold tiers, reading your own data carries a charge on top of egress.

The object storage form asks for all five, which is the point. A comparison that only asks for capacity is not comparing bills, it is comparing one line of a bill.

Why price per GB is the least important number

Take a modest workload: 1,000 GB stored on a standard tier, 100,000 GET requests and 10,000 PUT requests a month, 100 GB leaving the network. If standard storage lists at $0.02 per GB-month and egress at $0.09 per GB, the bill assembles like this:

Egress is already 31% of that bill while you are only storing a terabyte. Now change one input and nothing else: make it a public download bucket serving 2,000 GB a month instead of 100 GB. Egress becomes 2,000 × $0.09 = $180, against an unchanged $20 of storage. The bill is now nine times the storage line, and the per-GB storage rate you spent an afternoon comparing decides 10% of what you pay.

This is the reason the tool insists on a transfer figure. A provider with a slightly higher storage rate and materially cheaper egress wins outright on a serving workload and loses on an archive workload, and you cannot tell which case you are in from a per-GB table.

The costs people forget

• Egress and bandwidth. Charged per GB leaving the provider's network, and the rate varies by a wide margin between providers. Transfers to the internet, to another region, and to another cloud are usually priced differently. Traffic between two services in the same region and zone is often free — but the same traffic across zones is not.
• Requests and operations. Priced per thousand operations, with writes and listings dearer than reads. Trivial for a bucket of large media files; significant for anything that reads millions of small objects, and genuinely painful for a workload that lists a large bucket on a schedule.
• Retrieval fees on cold tiers. Separate from egress and separate from the request charge. Archive tiers are cheap to fill and expensive to read, and a single unplanned full restore can cost more than a year of the storage it saved.
• Minimum storage durations. Cold tiers bill a minimum retention period per object — delete or overwrite an object before that period elapses and you are still charged for the remainder. A lifecycle rule that shuffles short-lived objects into an archive tier can raise the bill rather than lower it.
• Minimum object size. Some tiers round every object up to a minimum billable size. A bucket of many tiny objects can be billed for several times the bytes it actually holds.
• Snapshots, versioning and incomplete uploads. Old versions, snapshot chains and abandoned multipart uploads all occupy billable capacity and none of them appear in the folder sizes you are looking at.

Cold tiers: read the retrieval line before the storage line

The object storage form reveals a retrieval field as soon as you select infrequent access, glacier or deep archive, because on those classes retrieval is a real cost centre rather than a rounding error. The decision rule is straightforward: estimate how much of the data you will read back in a typical month, then check whether the retrieval charge on that volume exceeds what you save by leaving the tier. For data you genuinely never touch — compliance archives, finished project media, log retention — the archive tiers are dramatic savings. For data someone might want back this quarter, the arithmetic frequently reverses, and it reverses hardest at the exact moment you need the data most.

Model both. Run the comparison once on standard with your real retrieval volume, then again on the cold tier with the same volume, and compare totals rather than per-GB rates.

Block storage is a different question

The Storage tab has a second mode for block volumes — EBS, Azure Managed Disks, Google Persistent Disks. Block storage attaches to one instance and is priced on provisioned capacity, not on what you have written, so a 500 GB volume that is 10% full costs the same as one that is full. That single fact accounts for a large share of avoidable block storage spend.

The form offers general-purpose SSD, high-performance SSD with provisioned IOPS, throughput-optimised HDD, cold HDD and legacy magnetic. It asks for provisioned IOPS only on the high-performance type and throughput only on general-purpose SSD, because those are the volume types where the figure is billable rather than inherited. Snapshot storage is a separate input, and it is worth filling in honestly: snapshot chains accumulate quietly and are frequently the largest storage line nobody budgeted for.

Reading the results

Submitting a workload returns a row per provider with monthly cost, annual cost, and a breakdown splitting the total into storage, data transfer, requests and other charges. The cheapest and most expensive options are identified, along with the difference between them in dollars and as a percentage, plus an average across providers. That breakdown is the part to read carefully — two providers can land within a few dollars of each other on the total while arriving there completely differently, and which one wins will flip the moment your traffic pattern shifts. Widen the egress input by a factor of two and re-run; if the ranking changes, your decision is a bandwidth decision, not a storage decision.

How the figures are produced

Be clear about what this is. The comparison is calculated server-side from a maintained table of representative list rates per provider, tier and service — it is a modelling tool, not a live feed from each provider's price list API. The rates are single representative figures rather than per-region variants, so changing region maps the comparison to the equivalent region on each provider without re-pricing every line item. Nothing here accounts for enterprise discount agreements, committed-use or reserved pricing on storage, free-tier allowances, promotional credits, or the taxes that appear on the invoice.

Use it the way it is built to be used: to find the shape of the answer, to see which cost component dominates your workload, and to narrow five providers to two. Then confirm the finalists against each provider's own calculator with your actual account terms before anything gets committed. A comparison tool that told you it was authoritative to the cent would be lying to you.

Questions worth asking before you migrate

• What does it cost to leave? Egress is charged on the way out, so moving a large dataset to a cheaper provider carries a one-off bill from the provider you are leaving. Model that transfer as a month of very high egress and check the payback period.
• Is the traffic actually egress? Data served through a CDN is billed differently from data served straight from a bucket, and often much more cheaply. Get the split right before comparing.
• How much of the stored data is live? Most estates have a long tail that is never read. Splitting the workload into a hot portion and an archive portion and modelling each separately usually beats picking one tier for everything.
• Are you comparing durability like for like? A single-zone price against a geo-redundant price is not a comparison, it is two different products.
• What happens at 3x the volume? Re-run the comparison at the capacity and traffic you expect in a year. The provider that wins today may not be the one that wins at scale, and storage migrations are not cheap to repeat.