What Is Amazon S3? Cloud Storage Explained in Plain English

Welcome back to our free AWS learning Series where we are preparing for AWS Machine Learning (ML) Certification exam from scratch and AI. Last time we toured Amazon SageMaker, the machine-learning workshop — today we visit the place where all of its ingredients actually live. Amazon S3 (Simple Storage Service) is AWS's cloud storage: you upload a file, Amazon keeps multiple copies of it in separate data centre buildings, charges you a tiny rent per gigabyte per month, and hands it back whenever you ask, from anywhere on Earth. That's it. That's the product. It launched in 2006 as one of the first AWS services ever, and today it quietly holds hundreds of trillions of files — a good chunk of the photos, backups and app data you touch every day passes through S3 without you knowing. Let's learn about them even if you are a non tech user.

Everything else you'll hear about S3 — buckets, objects, storage classes, Glacier, data lakes — is detail around that one idea: a warehouse you rent shelf by shelf, with staff who never lose things. As usual, we'll walk through it in plain English with Jake and Ethan, because most S3 introductions are written for people who already know three other cloud platforms, and you shouldn't need to be one of them.

The Night Jake's Backup Pendrive Died

Jake called Ethan at 9 pm, which is never a good sign. The shop's ageing desktop — nine years of purchase invoices, supplier contacts, warranty records — had refused to boot. "It's fine," Jake said, in the voice of a man convincing himself. "I have the pendrive."

The pendrive was in the same drawer as the desktop. The drawer was under the water purifier. You can guess the rest of this story. The desktop recovered after a new power supply — Jake got lucky — but the pendrive didn't, and Jake spent the next morning asking Ethan the right question for once: "Where do people keep files so that one bad night can't take them?"

The boring, industrial-strength answer to that question is cloud storage — and the one banks, hospitals, Netflix and your favourite apps all use is Amazon S3. So let's look at what Jake would actually be renting.

A Warehouse You Rent Shelf by Shelf

Ethan's explanation started with what S3 is not: it's not a disk you buy, not a folder that syncs, not a website with smiley previews. It's a storage counter with three operations: give them a file (upload), name the file you want back (download), tell them to bin one (delete). The industry name for this design is object storage — worth knowing only because job listings and cloud computing courses use it constantly. Everything else is built on those three verbs.

The vocabulary is smaller than it sounds:

• Object — a file, plus its name and a bit of information about it. A photo, a PDF invoice, a 4 GB CCTV recording, a database backup. Anything up to 5 TB per file.
• Bucket — the container you create to hold objects. Your room in the warehouse, with your name on the door. Bucket names are unique across the entire world — if someone anywhere owns the name jakes-shop-backups, nobody else can take it.
• Region — the physical place your bucket lives. You choose it when you create the bucket: Mumbai, Virginia, Frankfurt, Sydney. Your data physically sits in data centres in that region and doesn't leave it unless you move it. For Jake, "my invoices stay in India" is one dropdown choice.
• Key — S3's word for a file's full name, like invoices/2026/august/inv-4471.pdf. Here's the mildly mind-bending bit: there are no real folders in S3. That whole path is just the file's name; the slashes are decoration. The console draws folders on screen because humans like folders, but underneath it's one flat shelf of names. You'll never notice day to day — it only matters when someone asks why you can't "create an empty folder" properly.

That's genuinely the whole mental model. Jake makes a bucket in the Mumbai region, drags nine years of invoices into it, and they're now sitting in — this is the part that matters — at least three separate data centre buildings at once.

"Isn't That Just Google Drive?"

Jake's immediate objection, and a fair one. Ethan's answer: same shelf, different customer.

Google Drive is a finished product for humans. It gives you a synced folder, previews, sharing buttons, 15 GB free, and a monthly plan after that. It's excellent at what it does, and if all you want is your documents on two devices, close this tab and use it with a clear conscience.

S3 is a raw material for builders. No sync client out of the box, no pretty previews — but it scales to petabytes without blinking, costs less per gigabyte, plugs into every other AWS service, and is controllable by software down to the finest detail. When a phone app saves your profile photo, when a company keeps seven years of records for the tax office, when a website serves you an image — some bucket somewhere is doing the work. Drive is a shop; S3 is the wholesale market the shops buy from.

The same logic separates S3 from the hard drive in Jake's desktop. A drive is fast and local, but it's one physical object in one building with one power supply — Jake has met this failure mode personally. You also can't install Windows on S3 or run programs from it; it stores files, full stop. Operating systems live on drives. Files you'd mind losing belong in something like S3.

Eleven Nines: The Most Boring Superpower in Computing

S3 is designed for 99.999999999% durability — eleven nines. Numbers with that many digits stop meaning anything, so here it is in shop terms: store ten million files, and the design expectation is that you lose one of them every ten thousand years. Jake's entire invoice archive is about twelve thousand files. He can do that maths with the calculator on his own counter.

The trick isn't magic disks — it's copies. Every object is stored redundantly across at least three physically separate buildings (AWS calls them Availability Zones) within your region, checked continuously for corruption in the background, and silently repaired from the healthy copies when a disk somewhere inevitably dies. Disks fail constantly at Amazon's scale; your files just never notice, because the water purifier would have to leak on three buildings in different parts of the city at the same time.

One honest asterisk, and Ethan made Jake repeat it back: durability protects you from Amazon losing your file. It does not protect you from you deleting it. Delete is delete. The fix is a checkbox called versioning: turn it on and S3 quietly keeps every previous version of a changed or deleted file, so "I overwrote the good copy" and "I deleted the wrong thing" both become five-minute recoveries instead of tragedies. For anything you'd cry about, versioning on. It's the S3 equivalent of the backup advice from every other post on this blog — and yes, you were going to skip it, and no, don't.

Hot Shelves and Cold Vaults: Storage Classes

Jake's twelve thousand invoices split naturally into two piles: this year's (the accountant wants them monthly) and everything older (touched once during an audit, if ever). S3 prices exactly along that line, with storage classes — same warehouse, different shelves:

• Standard — the front shelf. Instant access anytime, roughly 2 US cents per GB per month. This year's invoices, live app files, anything used often.
• Standard-IA (Infrequent Access) — the back room. Same instant access, about half the storage price, but a small fee per retrieval. For files you keep "just in case" and open a few times a year.
• Glacier classes — the cold vaults, for archives. The deepest one, Glacier Deep Archive, costs about a tenth of a cent per GB per month — Jake's entire decade of old invoices for less than one samosa — but getting files back takes hours and costs a retrieval fee. Perfect for "the tax office might ask in 2031"; terrible for "the accountant needs it by lunch."
• Intelligent-Tiering — the honest default when you don't want to think. S3 watches how often each file is actually touched and moves it between warm and cold tiers automatically, for a tiny monitoring fee and no retrieval charges. Choosing this is not laziness; it's usually the right answer.

Better still, you can set lifecycle rules — standing instructions like "anything in invoices/ older than 12 months moves to Deep Archive automatically." Jake sets that once and his filing discipline problem is solved for life, which, Ethan noted, is more than any drawer ever did for him.

What It Costs, With the Traps Marked Honestly

S3 pricing has three ingredients: rent, retrievals, and the exit door. The rent itself, you now know: about 2 cents/GB/month on the front shelf, a tenth of a cent in the deep vault, and new AWS accounts get free allowances to experiment with before any of it bites. Jake's whole archive — call it 60 GB — is about the price of a cup of chai per month on Standard, and nearly nothing archived. So where do the S3 horror-bill stories come from? Three places, all avoidable:

• The checkout fee. Uploading into S3 is free; storing is cheap; downloading large amounts back out to the internet is where AWS charges real money — think roughly 9 cents per GB after a monthly free allowance. Backing up 60 GB costs pennies; pulling all 60 GB back down every week as part of some routine would quietly cost more than the storage. S3 rewards "upload and leave it," which is what backups are anyway.
• The frozen-food mistake. Archive classes charge for retrieval and expect files to stay put (early-deletion fees apply for the first months). Putting frequently-used files in Glacier to save money is like storing your daily groceries in a bank locker — the visits cost more than the rent saved. Cold storage is for genuinely cold files.
• The forgotten bucket. Pennies per month, times years, times buckets nobody remembers creating. Not ruinous, just untidy — a once-a-year look at your own account is plenty.

Requests (each upload/download operation) also cost fractions of a cent — at Jake's scale this rounds to nothing, and you only think about it when software makes millions of requests.

"But I've Read About Leaky Buckets in the News"

So has Ethan — it's a whole genre. Here's the part the headlines skip: S3 buckets are private by default, and have been aggressively so for years. A new bucket additionally gets Block Public Access switched on — a master safety catch that stops the contents being opened to the internet even by mistake. Every leaky-bucket story you've read was someone deliberately flipping a bucket public (usually to make a website work in a hurry) and forgetting what else they'd put in it. The warehouse doors lock themselves; the news stories are about people propping them open with a chair.

Day to day, the sharing tool worth knowing is the presigned URL: a link to one specific file that works for a time you choose — an hour, a day — and then dies. Jake can send his accountant one invoice PDF without his bucket, his other files, or even that file's future being exposed. Files are also encrypted at rest by default these days, which means even the disks themselves are useless to anyone who somehow walked out of a data centre with one.

What People Actually Build on It

Once you see S3 as "the place software keeps files," you start seeing it everywhere:

• Cloud backup and archives — Jake's case: the 3-2-1 backup rule's "one copy off-site," solved for pocket change. Most commercial online backup products are, underneath, exactly this.
• App storage — every profile photo, chat attachment and uploaded document in most apps you use lands in a bucket (or one of its imitators — S3's design became such a standard that competitors advertise being "S3-compatible").
• Static website hosting — a bucket can serve HTML, CSS and images directly, making it one of the cheapest ways on the internet to host a simple site. No PHP, no database — pure files — but for portfolios and documentation, a bucket behind a domain name is the whole architecture.
• Data lakes for machine learning — remember the spreadsheet Jake fed into SageMaker Canvas last time? In real setups, that data lives in S3, SageMaker reads it straight from the bucket, and the trained model is saved back into one. If machine learning is the kitchen, S3 is the pantry — which is exactly why this post exists in the series.

What S3 Will Not Do for You

Fair warnings, in the spirit of this series:

• It won't replace your computer's drive. You can't install programs on it or boot from it. (AWS sells a different service for that — EBS — which is a story for another post.)
• It won't sync your desktop by itself. Out of the box there's no magic folder like Drive or OneDrive; syncing needs a small tool or script. For "upload my backup weekly," that's a one-line command — but it's a line, not a checkbox.
• It's not a database. It stores whole files; it doesn't search inside them or update single rows. Databases are a different aisle of AWS entirely.
• It won't organise you. Eleven nines of durability applied to junk gives you immaculately preserved junk. Lifecycle rules help; naming things sensibly is still your job.

Try It Yourself in Fifteen Minutes

Nothing in this post needs to stay theoretical. With an AWS account: open the console, search S3, click Create bucket, give it a name nobody on Earth has taken and pick your region, then drag any file onto it. That's a working, triple-building-redundant, encrypted file store — from zero, in the time it takes to drink the chai you're now legally required to have with every post in this series. Click the file and download it back to prove it's real. Delete the test file and bucket afterwards if you like, or leave them; a bucket with one photo in it costs effectively nothing.

Jake's version of this story ends properly: a bucket in Mumbai, versioning on, a lifecycle rule shovelling old invoices into Deep Archive, and one scheduled weekly upload from the shop desktop. Total monthly cost: less than the pendrive cost him, once. The drawer under the water purifier now contains only the water purifier's manual, which Ethan considers character growth.

Questions People Actually Ask

What does S3 stand for?

Simple Storage Service — three S's. Launched March 2006, one of the first AWS services ever, and still the one nearly everything else leans on.

How big can files be?

Up to 5 TB per object, and there's no meaningful limit on how many objects a bucket holds or how big it grows. For scale: S3 as a whole stores hundreds of trillions of objects.

Which region should I choose?

The one closest to whoever reads the data most — for most readers here, Mumbai (or Hyderabad). Closer means faster and, where it applies, keeps data in-country. Prices differ slightly between regions; the differences are real but not life-changing at personal scale.

Can I access S3 from Windows like a normal folder?

Not natively, but the free official AWS CLI gives you commands like aws s3 sync for scripted backups, and several third-party tools present buckets as drives. For the "weekly shop backup" use case, one scheduled CLI command is the clean answer.

Is S3 only for companies?

No — pay-per-use means a personal archive of 50 GB costs about a dollar a month, with no minimums. The honest entry requirement is comfort with the AWS console rather than money. If you'd rather have a polished consumer product, Drive and OneDrive remain perfectly good answers; S3 is for when you want the industrial shelf.

What's next in this series?

We keep following the data. It's parked in S3; SageMaker can already cook with it — the natural next stops are the computers themselves (EC2) and the different storage AWS attaches to them (EBS), so the drive-versus-warehouse line from this post gets its own full story.

Published August 2026. Part of the free AWS learning series — previous stop: What Is Amazon SageMaker? A Plain-English Introduction. Hope it is helpful! See you in next post.