Amazon EBS Pricing (2026): gp3 vs gp2 vs io2, and How to Cut It

Amazon EBS pricing is simple to read and easy to overpay, for two reasons that sit in the rate card itself. First, EBS bills provisioned capacity, not used capacity, and bills it whether or not the volume is attached to anything, so a 500 GB volume that is 30 percent full and was detached from an instance terminated in March costs exactly what it did in production. The second is that gp2, the default volume type for a decade, still costs $0.10 per gigabyte-month while gp3 costs $0.08 and performs better at every size, and a large share of estates have never migrated. Between orphans and gp2, most AWS bills carry an EBS line 30 to 50 percent larger than it needs to be.

This guide gives every EBS volume type's rate in US East as of October 2026, explains how IOPS and throughput are charged, compares gp3, gp2, and io2 with worked examples, covers snapshot pricing and the archive tier, and lists the five changes that cut the bill. Rates are from the official EBS pricing page; for the file-storage alternative, see our comparison of EFS and EBS.

The EBS Price List

US East (N. Virginia), October 2026.

Snapshots: $0.05 per GB-month in the standard tier, billed on the blocks changed since the previous snapshot rather than on the full volume, so a chain of daily snapshots of a lightly changing volume costs a fraction of the volume size; $0.0125 per GB-month in the archive tier, which stores a full snapshot for a minimum of 90 days, with a $0.03 per GB restore charge and up to 72 hours to restore. Fast snapshot restore, which removes the first-read latency after a restore, bills per volume per Availability Zone-hour. Data transfer for EBS within an Availability Zone is free.

gp3 vs gp2: Why Migration Is Free Money

gp2 was designed when performance had to be bought by size: 3 IOPS per gigabyte, so a 100 GB volume got 300 IOPS and burst to 3,000 from a credit bucket, and a volume that needed 10,000 IOPS had to be provisioned at 3.3 TB whether or not the data filled it. gp3 decoupled them. Every gp3 volume gets 3,000 IOPS and 125 MB/s regardless of size, and more can be provisioned independently for $0.005 per IOPS-month and $0.04 per MB/s-month. The capacity rate is 20 percent lower.

The migration is an in-place volume modification: no downtime, no data copy, no instance change. There is no workload for which gp2 is the better choice, and an estate that has not migrated is paying 20 percent for nothing on every volume. Volumes that need performance bought through size on gp2 save far more.

gp3 vs io2: When the Expensive Volume Earns Its Price

io2 costs 56 percent more per gigabyte than gp3 and charges $0.065 per provisioned IOPS-month against gp3's $0.005, thirteen times as much. What it buys is a ceiling of 256,000 IOPS and 4,000 MB/s against gp3's 16,000 and 1,000, consistent sub-millisecond latency, and 99.999 percent durability against 99.8 to 99.9 percent. A database that needs 50,000 IOPS has no choice; a database that needs 8,000 and is on io2 because the template said so is paying about $700 a month for 1 TB that gp3 would serve for about $105. The decision is a measured one: look at the volume's actual IOPS and latency over a month before paying io2 rates for them.

Where EBS Bills Grow

• Unattached volumes. Terminated instances leave their non-root volumes behind unless delete-on-termination is set, and they bill at full rate. Estates routinely carry hundreds.
• Over-provisioned capacity. EBS bills the size you provisioned, and volumes are provisioned with headroom that is never used. A fleet of 1 TB volumes averaging 35 percent full is paying for 650 GB of air each.
• gp2 by default. Templates and AMIs written before December 2020 still create gp2, and nobody changes a working template.
• io1 and io2 by template. Database modules that provision io2 for every environment, including staging and test, pay provisioned IOPS around the clock for databases that do nothing at night.
• Snapshots without expiry. Backup scripts that take daily snapshots and never delete them, snapshots of volumes that were deleted years ago, and AMIs for retired builds, all at $0.05 per GB-month.

Each of these is invisible on the invoice, which shows EBS as one line, and visible only in an inventory of every volume and snapshot with its size, its attachment state, its age, and its owner, the pattern our guide to where cloud waste hides describes for every service.

A Worked Example

A company runs 2,400 EBS volumes totaling 480 TB, 70 percent on gp2, plus 90 TB of snapshots with no lifecycle policy and 40 io2 volumes for databases. The EBS line is about $56,000 a month: gp2 $33,600, gp3 $11,500, io2 capacity and IOPS $6,400, snapshots $4,500.

Migrating gp2 to gp3 in place saves $6,700 a month. An inventory finds 310 unattached volumes totaling 41 TB, mostly gp2 from terminated instances, worth $3,900 a month once deleted, and 55 TB of snapshots whose source volumes no longer exist, $2,750. Right-sizing the 200 largest volumes, averaging 30 percent utilization, from 1 TB to 500 GB saves $8,000. Of the 40 io2 volumes, 28 are in non-production and average 900 IOPS; moving them to gp3 saves about $4,200. Moving the remaining snapshots older than 90 days to the archive tier saves $1,100. The line falls from about $56,000 to about $29,000, 48 percent, with no change to any application, and the single largest item was a migration that took no downtime.

How to Cut EBS Costs

1. Migrate gp2 to gp3

Modify every gp2 volume in place. Twenty percent on capacity, better performance, no downtime. Update the templates and AMIs so new volumes are gp3.

2. Delete unattached volumes and set delete-on-termination

List volumes with no attachment across every account and region, snapshot the ones anyone might need, delete the rest, and set delete-on-termination on non-root volumes that should not outlive their instance.

3. Expire snapshots and archive the long-lived

Data Lifecycle Manager policies that keep a defined number of snapshots and expire the rest; delete snapshots and AMIs for resources that no longer exist; move compliance snapshots to the archive tier at a quarter of the price.

4. Right-size provisioned capacity

EBS bills provisioned size, so a volume should be sized to what it holds plus reasonable headroom. Volumes can be grown in place at any time, which removes the reason to over-provision.

5. Buy IOPS only where measured demand needs them

gp3's included 3,000 IOPS serve most workloads; provision more on gp3 at $0.005 per IOPS where CloudWatch shows sustained demand above it; reserve io2 for workloads whose measured IOPS, latency, or durability requirements exceed what gp3 can provision, and move non-production io2 volumes to gp3.

6. Match the type to the access pattern

st1 at $0.045 for sequential, throughput-heavy data and sc1 at $0.015 for cold sequential data cost a fraction of SSD. For shared file access across instances, EFS prices per GB used rather than provisioned, which our EFS versus EBS comparison works through.

How Opslyft Helps

Every item above depends on seeing each volume and snapshot with its type, size, attachment state, age, and owner, across every account, which the EBS line on the invoice does not show and the console shows one region at a time. Opslyft's inventory lists them with a cost and an owner on each, flags the unattached, the orphaned, and the io2 in non-production, and routes the cleanup to the team responsible; the broader pattern is in our guide to where cloud waste hides.

The Bottom Line

Amazon EBS pricing is a per-gigabyte rate on what you provisioned, charged whether it is attached, full, or on the right volume type, which is why the EBS line is so often 30 to 50 percent larger than the storage actually in use. Migrate gp2 to gp3, delete what is unattached, expire and archive snapshots, right-size capacity, buy IOPS on evidence, and match the volume type to the workload. None of it touches a running application, and the first step alone returns 20 percent of the gp2 estate. For how the block, file, and object options compare, see our EFS versus EBS guide and our S3 pricing breakdown; for the IOPS behind every number here, our guide to IOPS.