Petabyte

A petabyte is a unit for the amount of data a computer stores. The smallest commonly used unit is the byte: a single letter in a text takes up roughly one byte. A thousand bytes make a kilobyte, a thousand kilobytes make a megabyte, a thousand megabytes make a gigabyte. A phone today typically has 128 or 256 gigabytes of storage. Going two more steps up this ladder, past the terabyte, you arrive at the petabyte. A petabyte is thus one million gigabytes, or around four thousand well-filled phones.

The order of magnitude data centers think in

For individual people, the petabyte plays no role. Nobody fills up a million gigabytes privately. But as soon as it comes to companies, research, or AI, it becomes the standard unit. Anyone talking about a data center's storage needs speaks in petabytes, just as one states distances between cities in kilometers rather than centimeters.

What matters here is what this scale costs. Data isn’t simply there — it sits on hard drives in buildings that consume electricity and need cooling. A petabyte of storage means, depending on the technology, hundreds of hard drives or several server racks. That’s why the figure shows up in business reports: it’s an indicator of how much infrastructure a provider actually owns.

With AI, a second point comes into play. Large language models learn from enormous amounts of text that are first collected and processed from the internet. These raw collections fall in the petabyte range before junk and duplicates are filtered out. The volume of data is thus, alongside computing power, one of the two major cost blocks in training.

How the unit builds up from bytes

The units always increase by a factor of 1000, similar to grams, kilograms, and tons. The sequence is byte, kilobyte, megabyte, gigabyte, terabyte, petabyte. After that come exabyte and zettabyte. Each level is a thousand times larger than the previous one, which is why the numbers grow so quickly.

There are, however, two competing counting methods. Hard drive manufacturers calculate with a clean 1000, while computers internally often use 1024, because they work in binary. For the second variant there are separate names with an inserted “-bi-”, such as pebibyte. The difference adds up: a pebibyte is about 13 percent larger than a petabyte. That’s why a newly purchased hard drive sometimes appears smaller in the operating system than stated on the packaging.

In practice, a petabyte never sits on a single device. The data is spread across many hard drives, usually copied multiple times so that nothing is lost if one drive fails. Special software combines these parts into a single large storage pool. To the user it looks like one drive; physically, it’s an entire room full of technology.

Where the figure shows up in reports

Most commonly, one encounters the petabyte in reports about data centers and cloud providers. When Amazon, Microsoft, or Google announce new sites, capacities are stated in petabytes or exabytes. Streaming services also cite such figures: the entire film catalog of a major provider roughly fits into a few petabytes, while the data traffic delivered per month is a multiple of that.

In research, too, the unit is part of everyday life. The particle accelerator at CERN generates double-digit petabyte amounts of measurement data every year. Weather services and space telescopes operate in similar ranges. These projects need their own data centers just to store their measurements and later be able to analyze them.

A common misconception is to automatically assume that a large amount of data is valuable. A petabyte of unsorted internet text is worth less for AI training than a carefully vetted fraction of it. The quality and rights status of the data matter just as much as the quantity. The number alone therefore says little about how good a model will ultimately turn out to be.