What Is Amazon EBS? Cloud Block Storage in Plain English

Welcome back to our free AWS learning series. Last stop we rented the computer itself — Amazon EC2, cloud computing in plain English — and that post ended with a promise: the drive inside the rented computer deserves a post of its own. Here it is. Amazon EBS (Elastic Block Store) is the hard drive of your cloud computer — you pick a size in gigabytes, pay rent of about 8 cents per gigabyte per month, and from inside the machine it behaves exactly like the C: drive in the PC you're reading this on. Open File Explorer on a Windows instance and it's sitting right there. And now the twist that earns this post its keep: that drive is not inside your computer. It lives on separate storage hardware down the hall, talking to your instance over the network — and that small architectural lie is precisely why your files survive when the computer dies.

Everything else about EBS — volume types, snapshots, the gp2-versus-gp3 business, the bill traps — is furniture arranged around that one idea: the drive and the computer are two different rentals that happen to be plugged into each other. Jake and Ethan will do the explaining, as always, because most EBS guides open with an IOPS formula, and you should get a story about a 64-cent bill instead.

The 64-Cent Charge Jake Wanted to Dispute

Quick recap for anyone joining mid-series: Jake runs a phone repair shop, his booking page now lives on a micro EC2 instance instead of a $1,000 tower next to the water purifier, and Ethan is the friend who translates cloud things into shop things. In August, Jake closed the shop for two weeks — a family wedding, the good kind of closure — and did the responsible thing he learned last post: he stopped the instance before leaving. Meter off. Bill should be zero.

The bill was 64 cents. Jake — a man who once chased a customer two streets over a missing screen protector — was ready to open a formal dispute with Amazon Web Services over an amount smaller than his chai budget. Ethan stopped him with one question: "When you stopped the computer... where do you think your booking page went?"

Silence. Then the slow horror of a man realizing his files were somewhere, and he'd never once asked where. The answer is the entire subject of today's post: the files were on the drive — and the drive never stopped existing, because the drive was never part of the computer in the first place. Those 64 cents were its rent: an 8 GB drive, at 8 cents per gigabyte per month. Cheapest lesson in this whole series.

The Drive That Lives Down the Hall

Inside your own PC, the drive is a physical object screwed into the case — computer dies in a power surge, and your data's fate depends on whether that one component survived. Cloud designers looked at that arrangement and refused to copy it. When you launch an EC2 instance, the "drive" it boots from is an EBS volume: a slice of a dedicated storage system elsewhere in the same building, connected to your instance over a fast internal network, with multiple copies maintained automatically on separate hardware inside that building. Your computer thinks it has a local disk. It's being lied to, politely, and the lie is load-bearing:

• The computer can die; the drive doesn't care. Hardware failure, termination, Jake's two-week stop — the volume sits on its own equipment, unbothered. Attach it to a new instance and everything is exactly where it was. In the physical world this would be sorcery; in AWS it's Tuesday.
• Drives attach and detach like USB, at data-center scale. A volume is its own object in the console with its own ID. Unplug it from one instance, plug it into another — a standard move when a machine misbehaves: detach the patient's drive, attach it to a healthy machine, read the logs. (If your problem today is that you can't get into the machine at all, that's the EC2 connection errors post — this trick is its emergency exit.)
• One computer can wear several. A small root volume for the operating system plus a big data volume for the database is the classic arrangement — the same logic as your C: and D: drives, except each is a separate rental you size and pay for independently.
• But — and this matters — the drive lives in one building. An EBS volume exists in one Availability Zone: copies on separate hardware, yes, all under the same roof. Remember the asymmetry from the S3 post: the warehouse keeps your keepsakes in three separate buildings. EBS is faithful; S3 is durable. Different words, different promises, and the snapshot section below is the bridge between them.

The name, decoded: Elastic Block Store. "Block storage" is just the technical term for how ordinary drives work — a grid of numbered blocks with a filesystem arranged on top — as opposed to S3's object storage, where you hand over whole files like parcels at a counter. And "Elastic," same as EC2's E, because the sizing stretches: 8 GB today, 500 GB when the business grows, changed live with a slider. Mostly. There's a one-way valve hiding in that word, and it's waiting in the cost section.

The Speed Grades (and the 20% Nobody Collects)

Every drive rental asks one more question than size: how fast? EBS answers with a short menu, and one row of it hides genuinely free money.

• gp3 — the default, and the right answer for almost everyone. General Purpose SSD, third generation: about $0.08 per GB per month, and every volume — even a tiny one — starts with 3,000 IOPS and 125 MB/s included. (IOPS = input/output operations per second, the count of tiny read-and-write acts a drive performs; a busy database is thousands of them a second.) Need more speed later? Buy speed separately, without buying size. As of 2026 a single gp3 volume stretches to 64 TiB and 80,000 IOPS — four times the old ceiling — which is truck territory Jake will never visit, but it means the default lane doesn't run out of road.
• gp2 — the older generation, still quietly billing millions of people. Here speed is tied to size: 3 IOPS per GB, so a small drive is a slow drive, and people bought hundreds of unneeded gigabytes just to reach the speed they wanted. gp3 broke that link, made speed a separate dial, and cut the rent by 20%. Here's the part worth saying plainly: AWS will never migrate you automatically. Every gp2 volume created years ago is still gp2, still paying $0.10 against gp3's $0.08 for a drive that's slower at small sizes. The switch is one action on a live volume, no downtime, no data loss — Volumes → Modify volume → change gp2 to gp3. If you take exactly one billable action because of this post, that's the one.
• io2 — the fanatic tier. Provisioned IOPS SSD for serious databases: speed guarantees into the hundreds of thousands of IOPS and a durability promise a hundred times stricter than the general-purpose tiers. If you need io2, you have a job title that already knows you need io2.
• st1 and sc1 — the spinning rust. Actual mechanical hard drives, at mechanical-hard-drive prices, for big slow streams and cold piles. Honest niche: by the time data is this cold, it usually belongs in the warehouse next door instead.

One 2026 quality-of-life note, since this blog was around when the old rule annoyed everyone: volume modifications used to carry a six-hour cooldown — one change, then wait. Since January 2026 you can make up to four modifications to a volume in a rolling day, no waiting period. Resize at breakfast, regret it at lunch, fix it by dinner.

What It Costs, With the Traps Marked Honestly

The headline math is friendly: gp3 at $0.08 per GB per month means a 100 GB drive costs about $8 (roughly ₹700) a month — the same rent as the micro instance it's plugged into, a symmetry Ethan finds funnier than Jake does. New AWS accounts come with free credits that cover this kind of experimenting (the old free-tier hours scheme became a credit system for new sign-ups in 2025 — same note as the last two posts). Now the traps, because marking them is the house specialty:

• You pay for the size you reserved, not the space you filled. This is the one that genuinely surprises people. A 500 GB volume holding 10 GB of files bills as 500 GB, every month, because those blocks are carved out and held for you whether you write to them or not. Your own laptop never charged you by the empty gigabyte; EBS does. The cure is cultural, not technical: start small, because growing is a slider — and shrinking is not.
• Volumes grow; volumes never shrink. The one-way valve inside "Elastic." Going from 100 GB to 200 GB is a live modification plus one command inside the machine to stretch the filesystem. Going from 200 GB back to 100 GB is not supported — the path is: make a new smaller volume, copy everything across, swap. Oversizing "to be safe" therefore has a compounding monthly cost, forever, or an afternoon of migration. Jake's booking page lives happily on 8 GB.
• Stop is not free — the 64-cent lesson. Stopping an instance ends the compute charge; the volume keeps billing, because keeping your data alive while the computer is off is literally its job. Small money at Jake's scale, real money at 500 GB scale. Coming back someday: stop. Not coming back: terminate — after the next trap.
• Terminate has an asymmetry that manufactures orphans. By default, terminating an instance deletes its root volume — but any additional data volume survives, detached, in a state the console mildly calls available. Available means: plugged into nothing, serving no one, billing full price. Data centers are full of these orphan drives, quietly charging rent for years for machines that no longer exist. Monthly ritual, thirty seconds: EC2 console → Volumes → sort by state — anything available should either have a reason or a delete.

Snapshots: Photographing the Drive

Since the pendrive funeral in post one, this series has repeated a single rule: a copy that lives in the same place as the original is not a backup. An EBS volume — multiple copies, one building — is faithful hardware, not a backup strategy. Fire, flood, fat-fingered rm: the volume's copies all agree on the same disaster. EBS's answer is the snapshot, and it's the feature where the drive-down-the-hall design pays off twice.

A snapshot is a photograph of the entire volume at one instant, stored in the warehouse-grade, multi-building storage system next door and restorable into a fresh volume at any time. Three properties make it better than it sounds:

• Snapshots are incremental. The first one copies the whole drive. Every one after stores only the blocks that changed since the last. Snapshot a 100 GB volume nightly and you are not paying for 100 GB thirty times — you pay for one full copy plus each night's changes, at about $0.05 per GB per month for what's actually stored. A nightly schedule on a shop's worth of data costs chai money.
• Deleting old snapshots is safe — even the first one. The counterintuitive follow-up everyone worries about: delete the original full snapshot and the incremental chain doesn't collapse. AWS reshuffles the blocks so every surviving snapshot can still restore completely. Keep a week, delete the rest, lose nothing. (For long-term keepers there's an archive tier at a quarter of the price, with a 90-day minimum stay — the warehouse's basement level.)
• A snapshot is also a moving van. Volumes are locked to one building — but snapshots aren't. Restore one into a different Availability Zone, or copy it to another region entirely, and you've done the thing the volume itself can never do: leave the building. Snapshots are how drives travel, how machines get cloned, and how "let me test this risky upgrade on a copy" becomes a ten-minute exercise instead of a prayer.

Jake's setup after the wedding trip, sketched by Ethan on the back of the disputed 64-cent bill: nightly snapshot of the instance's volume (automated — AWS has a scheduler called Data Lifecycle Manager so no one has to remember), keep seven, and the invoices additionally flow to the S3 warehouse weekly, because customer records deserve the three-building promise. Desk, photograph of the desk, vault. That's the whole professional pattern, and a two-person repair shop just implemented it for under a dollar a month.

One honest aside before the limits section, because sharp readers ask: some instance types also come with an instance store — a drive that really is physically inside the host machine. Blazing fast, included in the price, and it forgets everything the moment the instance stops or dies. It's a whiteboard bolted to the rented apartment's wall: brilliant for scratch work, never for anything you'd miss. If you didn't deliberately choose it, you're on EBS, which is the right default.

What EBS Will Not Do for You

Fair warnings, in the spirit of the series:

• It won't leave the building. A volume lives and dies in one Availability Zone. The instance must be in the same building to attach it, and "let's move the drive to Frankfurt" is a snapshot operation, not a drag-and-drop.
• It won't back itself up. Faithful is not immortal — AWS's own figures put annual volume failure rates in the fraction-of-a-percent range, which is superb hardware and a terrible backup plan for data you love. No snapshot schedule, no sympathy. Two clicks in Data Lifecycle Manager buys the sympathy back.
• It won't be shared storage. One volume, one instance — the D: drive of one particular computer, not a network folder for the whole office. (A costly io2 exception called Multi-Attach exists for specialist clustered databases; if you're wondering whether you need it, you don't.) Many machines reading the same files is the warehouse's job, or a different AWS service wearing a costume.
• It won't shrink, and it won't stop billing out of politeness. The two traps worth repeating in the warnings list, because they're the two that cost real money: size is a one-way slider, and a volume bills until you delete it — stopped instance, terminated instance, forgotten available orphan and all.

Try It Yourself in Ten Minutes

Cheapest hands-on lap in the series so far. In the AWS console: EC2 → Volumes → Create volume — size 1 GiB, type gp3, and pick the same Availability Zone as an existing instance if you have one from last post's exercise. Watch it appear as its own object with its own ID: a drive that belongs to no computer. Attach it to the instance (Actions → Attach volume) and there's your D: drive materializing on a running machine. Then take its picture: Actions → Create snapshot, watch it complete under Snapshots. Then clean the room, in order: delete the volume, and the snapshot too if you don't want to keep the souvenir — at these sizes the whole field trip costs about a cent, but the habit is the curriculum. You have now created, attached, photographed and destroyed a cloud drive, which is four more drive operations than most people perform in a lifetime. Chai is, as ever, legally required.

And if you have any gp2 volumes from an older experiment: Volumes → select → Actions → Modify volume → gp3. Live, no downtime, 20% off forever. That one's not even practice — that's just money.

Questions People Actually Ask

What does EBS stand for?

Elastic Block Store. "Block storage" is the technical name for how ordinary drives work — numbered blocks with a filesystem on top — as opposed to S3's parcel-counter object storage. Launched 2008, two years after the S3-and-EC2 class of 2006 furnished the modern internet.

Is an EBS volume the same as my instance's hard drive?

From inside the machine, yes — it's the C: drive, entirely ordinary. Physically, no — it's networked storage down the hall, kept in multiple copies, merely impersonating a local disk. The impersonation is why your files outlive your computer, so nobody's complaining.

How much does EBS cost?

About $0.08 per GB per month on gp3 — 100 GB ≈ $8 a month. Remember the two clauses in fine print: billed on reserved size, not used space, and billed whether the attached computer is running, stopped, or long gone.

What is the difference between gp2 and gp3?

gp2 ties speed to size (3 IOPS per GB); gp3 gives every volume 3,000 IOPS and 125 MB/s at any size, sells extra speed separately, and costs 20% less. The conversion is live and lossless — and never automatic, which is why so much of the world is still overpaying.

Do I keep paying for EBS when my instance is stopped?

Yes — that's Jake's 64 cents. Stop kills the compute meter; the drive keeps charging rent for staying alive, which is exactly what you asked of it. Terminated instances can also leave orphan data volumes billing in the available state — worth a monthly glance at the Volumes list.

What is an EBS snapshot?

A restorable photograph of the whole drive, kept in multi-building warehouse storage. Incremental after the first — only changed blocks are stored — at about $0.05 per GB per month. Also the only way a volume's contents ever leave their building, making it backup and moving van in one feature.

Can I make an EBS volume bigger without downtime?

Yes — live resize while the machine runs, plus one command inside to stretch the filesystem, and since January 2026, up to four modifications per day with no cooldown. The valve is one-way: grow freely, shrink never. Start small.

EBS or S3 — which one should my files live on?

Desk versus vault. Operating systems, programs and databases need a real block drive: EBS. Anything whose loss would hurt belongs in the three-building warehouse: S3. Real systems use both, with snapshots as the bridge — the answer hasn't changed since the pendrive funeral, only the vocabulary has.

What is next in this series?

The pantry, the kitchen, the shelves — next comes the question every shop eventually faces: who gets keys to what? IAM, the permissions office of AWS, where more real-world accidents are born than in any hardware failure. (Impatient? The S3 Access Denied post is a preview of what happens when permissions go sideways.)

If You're Here for the Certification

The standing reminder, since this series grew out of our AWS certification learning series: storage-selection questions are reliable exam currency. "Which volume type for a high-IOPS database?" (io2), "cheapest way to keep point-in-time recovery?" (incremental snapshots), "what happens to the instance store on stop?" (it forgets), and the evergreen EBS-versus-S3-versus-instance-store triage all show up across the associate-level exams — including the machine learning ones, where "where does the training data live and why" is a favorite costume for this exact material. You can now answer those from a story about a 64-cent bill and a wedding, which sticks better than a comparison table. And if you're planning a booking, check the AWS certification changes post first — 2026 moved several deadlines that decide which exam version you should even sit.

Published August 2026. Part of the free AWS learning series — previous stop: What Is Amazon EC2? Cloud Computing in Plain English. I hope you learnt something new! See you in next post.