AWS S3 Pricing Explained for Beginners: The 4 Meters That Bill You

TL;DR
- S3 bills four meters — stored GB-months, request counts by tier, data transfer out, and retrieval/early-deletion extras. For a small site, storage is almost never the line item that hurts.
- Order of magnitude to keep in your head: S3 Standard is roughly $20 per TB-month; internet egress is roughly $90 per TB after the first 100 GB free each month. That ratio is why "cheap object storage" surprises people.
- The two fixes that cut most beginner bills are lifecycle rules (move cold bytes to Standard-IA or Glacier Instant Retrieval) and a CDN in front so CloudFront absorbs the repeat GETs instead of S3.
What S3 actually meters (four lines, one invoice)
S3 is not billed like a hard drive, and nothing is provisioned. You pay an hourly rate on the bytes sitting there, rolled up into GB-months; a per-call charge for the API requests your app makes; a per-GB charge whenever data leaves the region over the internet; and a set of smaller fees that only show up once you use the colder storage classes. Beginners get blindsided by meter three, not meter one.
Storage price depends on class and region. S3 Standard sits on the order of $20 per TB-month (about 2.3 cents per GB-month) in us-east-1; I'm using that as a working number for 2026-11, not a quote — ap-southeast-2 and eu-central-1 run noticeably higher. Standard-IA is roughly half the storage rate but adds a retrieval fee and a 30-day minimum. Glacier Instant Retrieval and Deep Archive go lower again but impose 90-day and 180-day minimums. Churn files through a cold class and you can pay more than Standard would have cost.
Requests are split into tiers that behave very differently. GET/SELECT calls are cheap per thousand; PUT/COPY/POST/LIST cost several times more; DELETE is free. Data transfer is the meter with the sharpest edge: outbound to the internet is billed per GB with volume tiers, inbound is free, and cross-region replication is billed on both sides.
Then there are the fees that tutorials skip: retrieval charges on IA and Glacier, early-deletion charges when you ignore the minimum duration, Intelligent-Tiering's per-object monitoring fee, and advanced Storage Lens metrics.
Three starter setups: what the monthly invoice looks like
Here's the arithmetic I actually run when an SMB asks "what will this cost?" Prices below are us-east-1 list rates rounded, before tax, and they exclude CloudFront. Treat every number as an order of magnitude and re-check the pricing page for your region.
| Setup | Stored | Requests/month | Internet egress | Dominant line | Monthly ballpark |
|---|---|---|---|---|---|
| A. Portfolio site + blog images | 5 GB | 50k GET, 2k PUT | 15 GB | nothing — inside allowance | under $1 |
| B. SMB marketing site, 200 GB media + nightly backups | 200 GB | 2M GET, 50k PUT | 600 GB | egress | $45–$70 |
| C. SaaS with user uploads via presigned URLs | 2 TB | 10M GET, 400k PUT | 3 TB | egress, then storage | $280–$360 |
Look at row B: storage is about $4.60 of a ~$50 bill. The egress line is ten times the storage line, which is the single most useful thing a beginner can internalize about S3.
The formula I run before opening a pricing calculator
Four multiplications get you within 20% of the bill. Everything past that is edge cases.
# us-east-1 working numbers, rounded. Verify against aws.amazon.com/s3/pricing
P_STD = 0.023 # $/GB-month, S3 Standard
P_GET = 0.0004 # $ per 1,000 GET requests
P_PUT = 0.005 # $ per 1,000 PUT/COPY/POST requests
P_EGRESS = 0.09 # $/GB out to internet, first 10 TB tier
FREE_OUT = 100 # GB/month free egress, all AWS customers
gb, gets, puts, out_gb = 200, 2_000_000, 50_000, 600
storage = gb * P_STD
requests = gets/1000*P_GET + puts/1000*P_PUT
egress = max(0, out_gb - FREE_OUT) * P_EGRESS
print(round(storage,2), round(requests,2), round(egress,2), round(storage+requests+egress,2))
# -> 4.6 1.05 45.0 50.65
Before you trust the inputs, measure what's already in the bucket. Guessing storage size is how estimates end up 5x off.
```bash
# total bytes across the bucket (JMESPath sum, no pagination issues for most buckets)
aws s3api list-objects-v2 --bucket my-bucket --query "sum(Contents[].Size)" --output text
# human-readable check, including object count
aws s3 ls s3://my-bucket --recursive --summarize --human-readable | tail -n 2
# after the fact: what actually drove the bill, grouped by usage type
aws ce get-cost-and-usage
--time-period Start=2026-11-01,End=2026-12-01
--granularity MONTHLY --metrics UsageQuantity UnblendedCost
--filter '{"Dimensions":{"Key":"SERVICE","Values":["Amazon Simple Storage Service"]}}'
--group-by Type=DIMENSION,Key=USAGE_TYPE
That last call is the one that ends arguments. Usage types like `USE1-Requests-Tier2` and `USE1-DataTransfer-Out-Bytes` map directly to the meters above.
## Sensitivity: what moves the bill by an order of magnitude
| Factor | Why it bites | Rough impact |
|---|---|---|
| Average object size under 128 KB on Standard-IA | IA bills a 128 KB minimum per object | 10k tiny files billed as if 1.28 GB, not 0.2 GB |
| No cache headers, no CDN | Every page view becomes a billed GET + egress | 10–50x request and egress volume |
| Versioning left on forever | Every overwrite keeps a full copy | storage creeps up ~5–15%/month on active buckets |
| Abandoned multipart uploads | Parts sit invisible in `aws s3 ls` | can silently pass your bucket size |
| Cross-region replication | You pay storage twice plus transfer | near doubles the storage line |
| Colder class with churn | Early deletion fee + retrieval fee | IA/Glacier can beat Standard only for stable data |
Multipart uploads deserve a callout because they don't appear in a normal listing. A single aborted 5 GB upload leaves parts billed as storage with no object to show for it. Lifecycle rule `AbortIncompleteMultipartUpload` with `DaysAfterInitiation: 7` is a two-minute fix that should be on every bucket.
## The free tier depends on when your account was created
Accounts opened before the 2025 free-tier revamp still get the classic 12-month S3 allowance: 5 GB Standard storage, 20,000 GET requests, 2,000 PUT requests and 100 GB/month of internet egress. Accounts created after that land on the credit-based plan, where you get a starting credit balance that expires. Both models exist in the wild right now, so the tutorials you find will contradict each other.
Check yours in the Billing console under Free Tier rather than guessing. The 100 GB/month egress allowance, by the way, applies to every account regardless of plan — that's why setup A in the table above is effectively free.
## Hard-won lesson: 34 million GETs and a missing Cache-Control header
The alert fired at 01:40 UTC on a Tuesday in 2026-02: a client's static site was returning intermittent `503 SlowDown` errors from S3. The real error text in the browser console was `An error occurred (SlowDown) when calling the GetObject operation: Please reduce your request rate.` S3 scales per prefix, so this wasn't a capacity wall — it was a request storm.
Cost Explorer explained it before the logs did:
USE1-Requests-Tier2 34,120,558 Count
USE1-DataTransfer-Out-Bytes 812 GB
Thirty-four million GETs on a marketing site doing maybe 120k page views. The first configuration was the problem: assets had been copied up with `aws s3 cp --recursive` and nothing else, so every object shipped with S3's default headers and zero cacheability. Each page view re-fetched the hero image, the JS bundle and the fonts from us-east-1 to visitors in Singapore and Frankfurt.
Two changes fixed it. First, cache metadata at upload time, split so hashed assets are immutable and `index.html` is not:
```bash
aws s3 cp ./dist/ s3://my-static-bucket/ --recursive
--exclude "index.html" --exclude "*.json"
--cache-control "public, max-age=31536000, immutable"
aws s3 cp ./dist/index.html s3://my-static-bucket/index.html
--cache-control "public, max-age=60"
Second, CloudFront in front of the bucket with the managed `CachingOptimized` policy, so repeat GETs terminate at the edge. Verification was a single command, run three times over an hour:
```bash
curl -sI https://cdn.example.com/assets/app.4f2a1c.js | grep -iE "cache-control|x-cache|age"
# cache-control: public, max-age=31536000, immutable
# x-cache: Hit from cloudfront
# age: 1832
A climbing `age` with `Hit from cloudfront` is the proof that the edge is serving, not S3. Next month, S3 GET volume dropped by roughly 90% and the `SlowDown` errors stopped. The lesson wasn't "add CloudFront" — it was that headers are part of the cost model, and the default upload path leaves them blank.
## When NOT to put this on S3
S3 is the wrong tool in a few situations beginners walk into. Don't mount it as a POSIX filesystem and run SQLite or a live database file on it — file locking and partial writes will corrupt the data. Don't park multi-terabyte compliance archives in S3 Standard; Deep Archive exists at a fraction of the rate. Don't serve a high-traffic public site straight from a bucket with no CDN, as above. And don't use One Zone-IA for anything you can't rebuild from source, because a single AZ failure takes it with it.
If your access pattern is "small team, a few hundred GB, mostly reads", S3 Standard plus a lifecycle rule plus CloudFront covers it and the bill stays boring. If you'd rather not maintain lifecycle rules and anomaly alerts yourself, that's the sort of housekeeping we do at awsgood.com for small ops teams.
## Two questions that come up every week
**Does deleting objects stop the charge immediately?** For Standard, yes — billing is hourly, so the next hour reflects the removal. For IA and Glacier classes, early deletion inside the 30/90/180-day minimum charges you the remainder anyway. Lifecycle transitions are one-way in cost terms; design the rule before you flip it.
**Why is the bucket bigger on the bill than in my listing?** Versions, noncurrent versions, delete markers and incomplete multipart uploads all bill but don't show in `aws s3 ls`. Turn on `ListBucket` metrics or run a version-aware listing with `aws s3api list-object-versions` before you assume the numbers are wrong.
## Sources
- [Amazon S3 pricing](https://aws.amazon.com/s3/pricing/)
- [Understanding and managing Amazon S3 storage classes](https://docs.aws.amazon.com/AmazonS3/latest/userguide/storage-class-intro.html)
- [Managing your storage lifecycle](https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-lifecycle-mgmt.html)
- [Best practices design patterns: optimizing Amazon S3 performance](https://docs.aws.amazon.com/AmazonS3/latest/userguide/optimizing-performance.html)
FAQ
How is AWS S3 storage actually billed?
S3 bills four things: stored
Leave a Reply