Cloud Storage Costs

The Three Bills Inside Every Cloud Storage Invoice

Here is how most people pick a cloud storage provider: they go to the pricing page, find the number next to "Standard Storage," and multiply it by their estimated terabyte count. S3 is $0.023/GB. R2 is $0.015/GB. B2 is $0.006/GB. Decision made in 30 seconds.

Here is the problem: that number is the smallest of the three charges on a real invoice. Cloud storage bills are not one line item — they are three, and the one that gets the pricing-page headline is not the one that determines your total cost.

I built the cloud storage cost comparison calculator because I watched a friend's startup get a $9,000 S3 bill for a workload they'd estimated at $400. The storage estimate was correct. They had modeled one-third of the invoice. The other two-thirds — egress and API operations — were 22× what they'd budgeted. They are not unusual. Every team I've talked to that runs a production workload on cloud object storage has some version of this story: the first bill was a surprise, the second bill was what they'd budgeted for the first, and by the third month they'd re-architected around the pricing model.

This article is about understanding all three dimensions before you get the first bill. I've modeled the numbers across six providers for four common workload patterns, and the spread between cheapest and most expensive within each pattern is rarely less than 5× and, in one case, exceeds 30×.

The Vendor Field: What Six Providers Actually Charge

The cloud object storage market in 2026 has split into two camps. The hyperscalers (AWS, GCP, Azure) charge for all three dimensions with pricing that hasn't fundamentally changed in five years. The challengers (Cloudflare R2, Backblaze B2, Wasabi) have each attacked a different dimension of the hyperscaler pricing model — R2 eliminated egress, B2 slashed the base storage rate, Wasabi eliminated both egress and API charges. The result is a market where the cheapest vendor for one workload is the most expensive for another.

Here is the raw pricing across all three dimensions, mid-2026, for the hot/standard tier in US East:

Sources: Provider public pricing pages, June 2026. All prices are US East region for the standard/hot storage tier. Egress pricing shown is for the first tier (typically 0–10 TB); hyperscaler egress rates decline at higher volumes. API operations are Class A (write) and Class B (read) equivalents. Enterprise negotiated discounts and committed-use contracts can reduce these rates — the numbers shown are the list prices a new customer pays.

Look at the extremes in this table. The cheapest storage (B2 at $0.006) is 3.8× cheaper than the most expensive (S3 at $0.023). The cheapest egress (R2 and Wasabi at $0.00) is — literally — infinitely cheaper than GCP at $0.12/GB. And API operations, which most people ignore entirely, range from Wasabi's zero to a line item that can match the storage cost for write-heavy workloads.

These differences are not marginal. They flip the vendor ranking depending on what your application actually does with the data. A backup workload (write once, read almost never) crowns B2 and Wasabi. A CDN origin workload (write rarely, read constantly with high egress) crowns R2. A data lake query workload (balanced reads and writes, moderate egress) can crown any of the six depending on the exact ratios. There is no "cheapest cloud storage." There is only "cheapest cloud storage for your egress-to-storage ratio."

The Egress Trap: S3 vs R2 at Real Scale

The single most expensive mistake in cloud storage is comparing $/GB storage rates while ignoring egress. It is the mistake the pricing pages are designed to encourage — put the low storage number in 48-point font, and bury the egress rate in a collapsible table three scrolls down.

Let's make the egress impact explicit. Take a workload that stores 10 TB and serves 50 TB of egress per month — a small CDN origin, a software distribution server, a video hosting backend. This is not a large workload. It's a mid-size SaaS product or a popular open-source project's download server.

* B2 egress: 3× stored volume free = 30 TB free, 20 TB billable × $0.01/GB = $200. API ops estimated at moderate read/write volume. All prices exclude the first-byte-free tier minimums where applicable. Egress calculated at first-tier pricing; hyperscaler egress rates decline at higher volumes but the order-of-magnitude gap persists.

The storage cost — the number on the pricing page — ranges from $61 (B2) to $236 (S3), a 3.9× spread. The egress cost ranges from $0 (R2, Wasabi) to $6,144 (GCP). That is not a spread — it is the difference between a line item and a bankruptcy. If your workload is egress-heavy, the $/GB storage number is noise. The egress rate is the signal.

This is why "S3 vs R2 cost comparison 1TB" is the wrong question. At 1 TB stored and 1 TB egress, S3 costs $113/month and R2 costs $15/month — a 7.5× gap, noticeable but not catastrophic. At 1 TB stored and 100 TB egress, S3 costs $9,023/month and R2 costs $15/month — a 600× gap. The storage number didn't change. The egress ratio did. Use our cloud storage cost calculator to model your exact numbers — the interactive bars make the egress-to-storage tradeoff visible in a way a static table can't.

Four Workload Profiles, Costed

Abstract pricing tables are useful for understanding the mechanics. But nobody runs "an average workload." You run a specific application with specific access patterns. Here are four common ones, costed across all six vendors with all three dimensions included. All scenarios use standard/hot tier storage in US East, no lifecycle tiering (we'll handle that separately), and moderate API operation estimates.

Profile A: The CDN Origin (Egress-Heavy)

10 TB stored · 100 TB egress/month · 50K writes · 500K reads

Software downloads, video streaming origin, container image registry, public dataset hosting. The defining characteristic: every byte stored is downloaded many times over. Egress dominates the bill so completely that storage and API costs are rounding errors.

* B2 egress: 100 TB total, 30 TB free (3× storage), 70 TB billable × $0.01 = $700. First-tier hyperscaler egress rates used; actual S3/GCP/Azure egress declines at higher volumes but the gap vs. zero-egress vendors remains >50×.

Profile B: Backup & Archive (Storage-Heavy, Low Egress)

100 TB stored · 2 TB egress/month · 1M writes (initial) · 50K reads (restores)

Database backups, VM snapshots, log archives, compliance retention. Data goes in, sits for months or years, and is retrieved rarely — typically in an emergency. Storage cost dominates. Egress is negligible. API ops matter only during the initial upload.

* B2 egress: 2 TB is well within the 3× storage free egress allowance (300 TB free on 100 TB stored).

Profile C: The Data Lake (Balanced, Ops-Heavy)

50 TB stored · 15 TB egress/month · 2M writes · 10M reads

Analytics data lake, log aggregation backend, feature store for ML training. Large volumes of data are written, queried, and occasionally exported to external systems. API operations are a meaningful cost driver because every Athena/Presto/Spark query generates thousands of LIST and GET requests.

* B2 egress: 15 TB is within 3× storage (150 TB free). API ops at moderate volume are a minor cost across all vendors at this scale.

Profile D: Log & Telemetry Storage (Write-Heavy, Grow-Over-Time)

5 TB new/month, 30 TB retained · 2 TB egress · 5M writes · 500K reads

Application logs, metrics, traces, security event data. This workload is characterized by continuous ingestion: high write operation counts with data that ages out or tiers down over time. Write API costs can match or exceed storage costs on some providers.

* B2 egress: 2 TB within free allowance. Write ops at scale add $25/mo on providers that charge for them — noticeable but not dominant.

For this workload, the zero-API-fee structure of Wasabi makes it competitive with B2 despite slightly higher storage pricing. At 5M writes/month, hyperscaler API charges add $25 — not trivial but not the decision variable. The decision variable remains egress: if you ever need to export your logs to an external analytics platform, the hyperscaler egress charges will dwarf everything else. If your logs stay within the same cloud as your analytics, the hyperscalers become more competitive — not on price, but on the zero-egress-between-services advantage. This is the lock-in mechanism working as designed.

The Cold Storage Calculus: When Does Cheaper Storage Actually Save Money?

Every hyperscaler offers cold storage tiers that cost a fraction of standard storage. S3 Glacier Deep Archive is $0.00099/GB — 23× cheaper than S3 Standard. GCP Archive is $0.0012/GB. Azure Archive is $0.00099/GB. These numbers look like obvious wins for data that isn't accessed frequently. But cold storage is not cheaper storage — it's a bet that you won't need the data, and if you lose the bet, it's more expensive than standard.

The cold storage pricing model works like this: you pay less per GB to store the data, but you pay a retrieval fee ($0.01–0.05/GB) every time you access it, and you commit to a minimum retention period (30–180 days). If you delete data before the minimum, you pay for the full period anyway. If you retrieve data frequently, the retrieval fees overwhelm the storage savings.

The Breakeven Math

Let's model 100 TB of data, comparing S3 Standard ($0.023/GB) against S3 Glacier Deep Archive ($0.00099/GB) with standard retrieval ($0.02/GB, 12-hour retrieval time) and a 180-day minimum retention:

The breakeven is surprisingly high: you have to access more than 90% of your cold-stored data every month before standard storage becomes cheaper — and even then, only marginally. The real reason not to use cold storage isn't cost. It's latency and operational complexity. Glacier Deep Archive retrieval takes 12 hours. If your application needs sub-second access to any object at any time, cold storage is not an option regardless of the price. If your data is a compliance archive that will be accessed once every two years in response to a subpoena, cold storage is a 23× cost reduction with no downside.

The Fine Print: Minimum Retention, Fair Use, and the Gotchas

The pricing tables above assume you're reading the headline numbers. But cloud storage pricing has a sub-basement of fine print that changes the total cost of ownership for specific workloads. Here are the ones that have burned people I know.

Minimum Retention: The 90-Day Trap

AWS Glacier and Glacier Deep Archive have a 90-day and 180-day minimum storage duration, respectively. If you upload an object to Glacier Deep Archive and delete it 30 days later, you are charged for 180 days of storage. At $0.00099/GB, that's $0.00594 for one GB deleted early — 6× the listed monthly rate. For 100 TB deleted after 30 days, that's a $594 penalty instead of the $99 you budgeted. Azure Archive has similar minimums (180 days). GCP Archive requires 365 days. Wasabi has a 90-day minimum on all storage — even hot — which means short-lived data costs 3× the advertised monthly rate.

If your workload involves data that lives for less than the minimum retention period — log files that rotate weekly, temporary build artifacts, staging data — cold tiers and Wasabi are more expensive than they appear. Model the effective rate as listed_rate × max(minimum_days, actual_days) ÷ actual_days. The cloud storage cost calculator surfaces these minimums in the results for every vendor.

Wasabi's Fair Use Policy: The Egress Asterisk

Wasabi's zero-egress pricing is the most attractive number in cloud storage. But it comes with a fair-use policy: your monthly egress should not exceed your stored volume. If you store 10 TB and egress 30 TB, Wasabi will flag your account. The policy is not automatic enforcement — it's a terms-of-service boundary — but it means Wasabi is not a CDN origin. For the CDN origin workload in Profile A (10 TB stored, 100 TB egress), Wasabi's $72/mo number is not available to you. R2's $156/mo is the real zero-egress option at that scale.

R2's Single Tier: No Lifecycle, No Archive

R2 charges $0.015/GB for everything. There is no cold tier, no archive tier, no infrequent access discount. For data that will be accessed regularly, this is fine — the zero-egress more than compensates. For 500 TB of compliance archives that will sit untouched for seven years, R2 costs $90,000/year while S3 Glacier Deep Archive costs $5,940/year. R2 is not the answer for archival storage. The absence of tiering is the price of simplicity.

B2's Egress Allowance: Free Up to 3×, Then Cheap

B2's egress model is the most nuanced of the six. You get free egress up to 3× your stored volume per month — if you store 10 TB, you get 30 TB of free egress. Beyond that, egress is $0.01/GB, which is 9× cheaper than S3's standard egress rate. For most workloads, the 3× allowance covers all egress. For the CDN origin workload (10× egress-to-storage ratio), B2's effective egress cost is $0.007/GB when you average the free and paid traffic — still the cheapest non-zero-egress option. The B2 model is closest to "egress at cost" in a market where most egress pricing is pure margin.

How to Pick a Cloud Storage Provider: The Decision Framework

After modeling every vendor against every workload, the decision tree collapses to three questions. Answer them in order, and the vendor picks itself.

Question 1: What is your egress-to-storage ratio?

• Egress > 2× storage (CDN origin, streaming, public datasets): R2. Zero egress at $0.015/GB is unbeatable here. Wasabi's fair-use policy excludes you.
• Egress 0.5–2× storage (balanced workloads): R2 or B2, depending on storage volume. At low storage volumes, the zero-egress convenience of R2 is worth the $0.009/GB storage premium over B2. At high storage volumes (50+ TB), B2's $0.006/GB base rate wins.
• Egress < 0.5× storage (backups, archives, data lakes): B2 or Wasabi. B2 is cheapest on storage. Wasabi is simplest (zero ops fees, zero egress, single predictable bill). If your data ages predictably, hyperscaler cold tiers enter the conversation.

Question 2: Does your data age predictably?

• Yes (logs age out after 30 days, backups after 90 days, compliance after 7 years): Hyperscaler lifecycle policies become valuable. S3 Intelligent Tiering automatically moves data between tiers based on access patterns — at a small per-object monitoring fee. If your data ages, you can capture cold-tier savings without manual lifecycle management. B2 and R2 don't offer this.
• No (random access patterns, unpredictable hot/cold distribution): R2 or B2. Paying a flat rate for everything is simpler and often cheaper than trying to predict access patterns and getting hit with retrieval fees when you're wrong.

Question 3: Where does your compute live?

• AWS: If your application runs on EC2/Lambda and your data is queried by Athena/EMR, the S3 premium buys zero-egress access between AWS services. The premium is real, but so is the architecture cost of moving your compute to match your storage. Model both sides. Our AWS Egress Cost Calculator will show you the cross-cloud egress penalty.
• Cloudflare: If you're already on Cloudflare for CDN/DNS/WAF, R2's zero-egress between Cloudflare services makes it the obvious choice — data in R2 flows to Cloudflare's CDN with no bandwidth charge.
• Cloud-agnostic (Kubernetes, self-managed query engines): B2 or R2. You have no lock-in penalty to amortize. Pick the cheapest vendor for your workload profile.

If the answer to all three questions points to different vendors, model it. Our cloud storage cost comparison calculator takes your actual storage volume, egress, and operation counts and ranks all six vendors side by side. The tool exists because the answer changes at every egress-to-storage ratio — and the only way to pick correctly is to run your numbers, not someone else's.

One more thing: don't optimize prematurely. At 1 TB, the monthly cost difference between the cheapest vendor (B2 at $6) and the most expensive (S3 at $23) is $17/month — $204/year. That's not worth the engineering time to migrate. Pick whatever is easiest to integrate. The cross-vendor spread only becomes material above 10 TB, and it only becomes a hiring-or-architecture decision above 50 TB. Optimize when the gap pays for the migration, and not before. The GPU training cost analysis has a similar dynamic — the provider premium only matters once you're spending thousands, not hundreds.

Frequently Asked Questions

Methodology & Disclosure

Pricing data is sourced from public provider pricing pages accessed June 2026 for US East regions (us-east-1 for AWS, us-east4 for GCP, East US for Azure, and default US regions for R2, B2, and Wasabi). All prices are list rates for pay-as-you-go customers with no committed-use discounts, volume discounts, or enterprise agreements. Hyperscaler egress rates decline at higher volume tiers (e.g., S3 egress drops from $0.09/GB to $0.085/GB after 10 TB, GCP egress drops to $0.11/GB after 1 TB, etc.) — the first-tier rates used here represent the ceiling for moderate-scale workloads. Actual bills for large-scale deployments will be lower than the hyperscaler totals shown.

API operation estimates are simplified — real-world API costs depend on object size distribution (many small objects drive higher operation counts per GB) and specific SDK behavior (multipart uploads, paginated LIST operations, retry logic). The scenarios in this analysis assume moderate object sizes (1–10 MB) with typical SDK overhead. If your workload consists of billions of small objects, API operation costs will be higher than shown — potentially by an order of magnitude. The cloud storage cost calculator allows you to adjust operation counts independently to model this.

Cold storage breakeven calculations assume standard retrieval speeds; expedited retrieval (AWS) and high-priority retrieval (GCP, Azure) cost 3–5× more and are not included except where noted. Minimum retention penalty calculations use the provider's published minimum duration for each tier as of June 2026.

Disclosure: jslet is an independent research project. We are not sponsored by any cloud provider, storage vendor, or infrastructure company. The cloud storage cost comparison calculator on this site was built because the author has been on both sides of the "surprise egress bill" conversation and wanted a tool that forces the full three-dimensional comparison before the invoice arrives. No affiliate links. No referral codes. No sponsored recommendations. If a vendor's pricing changes, tell us and we'll update the data.

References & Further Reading

1. AWS (2026). "S3 Pricing — Storage Classes, Data Transfer, and Request Pricing." US East (N. Virginia) region pricing for S3 Standard, Intelligent-Tiering, Glacier, and Glacier Deep Archive. aws.amazon.com
2. Cloudflare (2026). "R2 Pricing — Zero Egress Object Storage." Flat-rate storage with no egress fees, Class A/B operation definitions. developers.cloudflare.com
3. Backblaze (2026). "B2 Cloud Storage Pricing." Storage, egress (3× free allowance), and transaction pricing. backblaze.com
4. Wasabi (2026). "Hot Cloud Storage Pricing." Flat-rate storage with no egress or API fees, fair use policy, and 90-day minimum. wasabi.com
5. Google Cloud (2026). "Cloud Storage Pricing — Storage Classes and Network Egress." Standard, Nearline, Coldline, and Archive tier pricing with Class A/B operation definitions. cloud.google.com
6. Microsoft Azure (2026). "Blob Storage Pricing — Hot, Cool, Cold, and Archive Tiers." Block blob pricing with minimum retention periods and retrieval fees. azure.microsoft.com
7. AWS (2026). "Data Transfer Pricing — Internet Egress and Inter-Region Transfer." Tiered egress pricing structure and AWS-to-internet data transfer costs. aws.amazon.com
8. Cloudflare Blog (2024). "Why Cloud Storage Pricing Needs a Reset — The Bandwidth Tax and the Zero-Egress Movement." Analysis of egress pricing economics and the structural shift in the object storage market. blog.cloudflare.com

📜 Copyright & Attribution

"Cloud Storage Costs: The Egress Trap — Why Your $/GB Number Is Lying to You (2026)" — jslet Research, June 2026.
https://www.jslet.com/cloud-storage-cost-real

📡 Enjoyed this? Storage is $0.02/GB. Your bill is never $0.02/GB. RSS delivers one real-cost breakdown per week. No vendor pitch. No tracking. Just the line items that matter. RSS Feed → | More options →