Glossary
Exabyte (EB)
An exabyte (EB) is one million terabytes, or one thousand petabytes, of data. It is the scale of hyperscale clouds and the largest research and media archives. At that size the data can no longer be moved in any practical window, so the storage has to grow, repair and renew itself in place.
How the unit is counted
Storage vendors and drive makers count in powers of ten, so an exabyte is 1018 bytes. Operating systems often count in powers of two and show the exbibyte (EiB), which is about 15 percent larger. The gap between the two conventions grows with each step up the scale, and at this size it amounts to well over a hundred petabytes, enough to make two capacity reports of the same system disagree.
| Unit | Equal to | In terabytes |
|---|---|---|
| Terabyte (TB) | 1,000 gigabytes | 1 |
| Petabyte (PB) | 1,000 terabytes | 1,000 |
| Exabyte (EB) | 1,000 petabytes | 1,000,000 |
| Zettabyte (ZB) | 1,000 exabytes | 1,000,000,000 |
Who stores an exabyte?
Hyperscale cloud providers, national research programs in genomics, physics and astronomy, large media libraries and a few global enterprises. Most organizations reach the figure, if at all, as a total across many systems rather than in one. For nearly everyone else, petabyte storage is the working scale, and it is where the habits described below first become necessary.
An exabyte cannot be migrated
Eight quintillion bits sent over a dedicated 100 Gb/s link, running at full speed with no overhead, take about two and a half years to arrive. No refresh project can wait that long, so the usual approach of buying a new array and copying everything across stops being an option. The design has to absorb change without moving the whole dataset:
- Protection without full copies. Three replicas of an exabyte would need three exabytes of drives, so systems at this size rely on erasure coding for fault tolerance at a fraction of the overhead.
- Repair as a constant. Among tens of thousands of drives, some are failing on any given day, and rebuilding runs in the background without staff involvement.
- Hardware renewed in place. New servers join and old ones drain while the data stays online, the model behind scale-out storage.
- A flat namespace. Billions of items are addressed by key rather than through folder trees, which is why storage at this scale is almost always object storage.
Processing moves to the data
When the dataset cannot travel, analysis jobs run where it sits and only their results move. That pull, described as data gravity, decides where compute is placed long before capacity planning begins.
Frequently asked questions
How many terabytes are in an exabyte?
One million, in the decimal units storage vendors use. In binary units, an exbibyte holds about 1.15 million terabytes.
Is any single storage system an exabyte in size?
A few are, mostly inside hyperscale clouds and very large archives. More often the figure is the combined capacity of many systems run by one organization.