A petabyte (PB) is a unit of digital data equal to 1,000 terabytes (TB), or 10^15 bytes, as defined by the international system of units (SI). Its binary counterpart, the pebibyte (PiB), is 2^50 bytes, about 12.6% more. The petabyte sits below the exabyte (EB) and is the scale at which storage stops being a hardware purchase and becomes an architecture and contract decision.
At a glance
- Decimal: 1 PB = 10^15 bytes = 1,000 TB. Binary counterpart: 1 pebibyte (PiB) = 2^50 bytes.
- The gap between the unit and its binary counterpart is about 12.6%, roughly 126 TB on every petabyte.
- As a rough illustration, large video and imaging archives, multi-year backup retention and analytics data lakes often reach petabyte scale.
- At this scale, egress, retrieval fees and migration time can matter as much as the price per TB stored.
What problem it solves
The petabyte gives a practical unit for data sets that have outgrown single arrays. Planning at this scale shifts the questions: which storage model (block, file or object) suits the data, how it will be protected, how much it costs to read back, and how it would ever be moved. Many organizations reach petabyte scale gradually through retention rather than through a single project, so the unit often appears first in a backup or archive bill.
How it works
Storage models. Petabyte deployments often span multiple systems or nodes. Block, file and object platforms, including SAN-based designs, scale-out file systems and object storage, can all run at this scale; which fits depends on the workload, access pattern and each platform’s limits. Object storage and scale-out file systems are common for large unstructured archives, while SAN-based designs often serve performance-sensitive block workloads. See file systems and object storage for the differences.
Protection overhead. At this scale, mirroring every byte is expensive, so systems often use erasure coding, which spreads data and parity across many drives or sites. The overhead varies with the scheme, and raw-to-usable ratios should be stated in any quote.
Tiering. Large data sets usually mix a small amount of frequently used data with a large amount that is rarely read. Many cloud and on-premises systems move older data to cheaper tiers automatically. Archive tiers can cost far less per TB to store, but often carry retrieval fees, retrieval delays and minimum storage periods.
Data lakes and analytics. A data lake that collects logs, events and files across an organization can grow into petabytes, and analytics engines may scan large portions of it, so query and access charges can matter alongside storage.
Transfer. At a full 1 Gbps, a petabyte takes about 93 days to send; at 10 Gbps, about 9 days. Leaving a cloud at this scale may also incur substantial data egress fees, unless an exit waiver or negotiated terms apply.
See our file and object storage and colocation pages for how buyers compare cloud and self-managed options at this scale.
When it matters for buyers
- Archive and retention decisions. Long retention policies are a common route to petabytes; check what they cost over the full period.
- Cloud versus colocation. At petabyte scale, owning or leasing storage in a colocation facility may compare well with cloud for steady, frequently accessed data; cloud often wins for bursty or rarely accessed data.
- Storage as a service. STaaS offerings can put petabyte-scale capacity on-premises under a consumption contract, typically with committed minimums.
- Exit planning. Migration time and egress should be in the business case before data is committed.
Questions to ask vendors
- What is the price per TB-month for each tier, and what are retrieval, request and early-deletion charges?
- What raw-to-usable ratio does your protection scheme give, and what failures can it tolerate?
- How does capacity grow, and what is the cost and lead time to add the next increment?
- What does it cost, and how long does it take, to export the full data set at the end of the term?
- Are capacities quoted in decimal PB or binary PiB?
How it differs from a pebibyte (PiB)
A pebibyte is 2^50 bytes, about 12.6% more than a decimal petabyte. Most hardware is sold in decimal units, while some software, monitoring tools and cloud bills report binary units, so the same data can be reported as 1 PB in one place and about 0.89 PiB in another. The glossary term Binary Prefixes covers the binary units in more detail.