Every time you stream a song, back up a phone, or let a camera keep rolling, you add a thin slice to a pile that no one can hold in their head. Researchers call that pile the global datasphere: not a tidy library, but everything created, captured, copied and consumed in a year. The number on that pile is now so large that the unit itself has to change. We have left gigabytes and terabytes behind. The working measure is the zettabyte — a trillion gigabytes.
A decade and a half ago the annual datasphere was still counted in single-digit zettabytes. By 2020 industry trackers put it at about 64 zettabytes. On the latest widely cited IDC path, 2025 sits near 218 zettabytes, with a line toward more than 700 zettabytes by 2030. Older public charts still show 181 zettabytes for 2025 and about 221 for 2026. The vintage of the forecast moves; the shape does not. The curve is steep. Day by day that works out in the region of hundreds of millions of terabytes — a flood of video, logs, sensor ticks, model outputs and copies of copies.
That last point matters more than the headline. The datasphere is not “how much humanity has filed away.” Most of what is counted is used once and overwritten: a live stream, a cache, a replica, a training pass. Only a thin layer is stored for years. Installed storage has grown fast, but it has always lagged creation. A building full of drives is not the same object as the annual flood.
Why the flood? Three engines run at once. First, people: phones, cameras, messages, entertainment. Second, machines: cameras on poles, cars, factories, satellites — devices that write whether anyone is watching. Third, software that writes more software’s fuel. Artificial intelligence does not only read data. It produces traces, synthetic sets, embeddings and logs. Each new model generation asks for more of yesterday’s output. That is a feedback loop, not a one-off spike.
None of this requires a scare story. Power, water and land already constrain the biggest sites. Efficiency still moves: better chips, better cooling, denser packs, less waste heat per useful bit. Those gains are why Earth data centres remain the rational place to put most of the work. The curve being steep is not the same claim as “the archive must leave the planet.” That is a later argument, for a later piece, and it is not what the numbers force today.
What the numbers do force is a change in how an ordinary reader should hear the word “data.” It is no longer a stack of documents. It is a weather system. Some of it is climate — the records we keep on purpose. Most of it is weather — passing through. Confusing the two is how headlines turn a forecast into a prophecy.
There is an older line that still helps the ear: from the dawn of recorded civilisation to the early 2000s, the stored record of our species was measured in a few exabytes. That much is now created in a couple of days. Treat the comparison as a picture, not a laboratory result. Methods have changed. Definitions have changed. The picture is still useful. Almost all of the digital pile is young.
For the person who only wanted to know “how big,” the honest answer is this. We are already in the low hundreds of zettabytes a year of created-and-copied activity. We are pointed at several times that by the end of the decade if growth stays in the low-to-mid twenties percent a year. We store far less than we generate. We analyse far less than we store. The constraint that bites first is not a shortage of ideas. It is watts, cooling, and the discipline to keep only what still has a job.
That is the broader man’s map. The pile is real. The unit is almost comic. The part that lasts is smaller than the part that moves. Understanding starts when those three sentences sit together.
