It seems like a bar question: “How many TB does the internet occupy?”. You throw out a number, everyone nods, done. The problem is that the question doesn’t have a single answer — it has at least three, and all differ by orders of magnitude. As developers, this is exactly where it gets interesting: the difference between “data created,” “data stored,” and “indexed web” is exactly the difference we manage daily when designing retention, replication, and storage costs in Azure or anywhere else.
First, the units: from GB to ZB
To speak coherently, we need to agree on the scale. Each step multiplies by 1000 (using the decimal SI convention, the usual one in industry reports):
| Unit | In bytes | Intuitive equivalent |
|---|---|---|
| 1 Gigabyte (GB) | 10&sup9; | a movie, about 250 songs |
| 1 Terabyte (TB) | 10¹2; | a well-equipped laptop SSD |
| 1 Petabyte (PB) | 10¹5; | 1000 TB — what CERN produces in one day |
| 1 Exabyte (EB) | 10¹8; | 1 million TB |
| 1 Zettabyte (ZB) | 10²1; | 1 trillion GB = 1 billion TB |
Remember the key equivalence: 1 ZB = 1 billion TB. When you see “181 zettabytes,” think of 181 billion 1 TB hard drives. At this scale, the question “how many TB?” becomes almost meaningless without specifying what exactly you are counting.
Three questions disguised as one
When someone says “how much does the internet occupy,” they usually mix three completely different things:
- Indexed web — the pages search engines find. It’s the visible part, but surprisingly small.
- Global datasphere — how much data is created, captured, and consumed annually worldwide. It’s the huge figure in news headlines.
- Actual storage — how much data is actually stored on disk somewhere, at some point. It’s much smaller than what is created.
We’ll take them one by one, because each has its own figure and estimation method.
1. Indexed web: smaller than you think
The part of the internet Google can find — the so-called surface web — is estimated by the WorldWideWebSize project at the order of a few billion unique pages. Sounds like a lot, but in bytes it’s negligible compared to the rest: the HTML text of a few billion pages easily fits into a few petabytes.
The reason the indexed web is so “small” is that the weight of the internet is not in text, but in binary content: video, images, streaming, backups, telemetry. One single minute of YouTube 4K weighs as much as tens of thousands of pages of text. And beneath the surface lies the deep web — private databases, app content, secure networks — which is not indexed at all and is orders of magnitude larger than the visible web.
2. Global datasphere: 181 zettabytes and growing
This is the big number you see in reports. IDC (via Statista) tracks what they call the Global DataSphere — the total volume of data created, captured, replicated, and consumed in a year. The trajectory looks like this:
- 2018: ~33 ZB
- 2020: ~59 ZB
- 2024: ~149 ZB
- 2025: ~181 ZB (projection)
- 2028: ~394 ZB (IDC projection)
In other words, the volume of data doubles roughly every four years. A commonly cited estimate says that about 90% of the world’s data was generated in the last two years — not because the past was erased, but because the production rate grows exponentially, so the “last two years” always dominate the total.
Daily perspective: in 2024, about 402 million TB per day were generated — approximately 0.4 ZB. Daily. Of which a good part is stuff no one will ever open a second time.
An increasing part comes from machines, not humans: IoT devices alone are estimated to produce over 70 ZB in 2025. Sensors, cameras, industrial telemetry — data flowing continuously, mostly processed at the edge and discarded immediately.
3. Created ≠ stored ≠ unique
This is the trap most popular articles fall into, and it’s exactly the distinction a developer feels in the bill. The 181 ZB represent data created and consumed, not data stored. The difference is huge for three reasons:
- Transient by definition: a video call, a Netflix stream, an API response — data passes through the network, is displayed, and disappears. It never reaches permanent disk storage.
- Massive replication: the same file exists in dozens of copies — database replicas, CDN caches at the network edge, backups, dev/staging/prod environments. A “1 GB” object can physically occupy 5–6 GB once replicated across regions.
- Redundancy inside storage: any serious system keeps at least three copies. Azure Blob in LRS keeps three synchronous replicas in a single data center; ZRS and GRS multiply even more.
The result: out of the over 180 ZB “of the internet” in a year, what is effectively stored long-term as installed base is a fraction — on the order of tens of zettabytes, not hundreds — and even there, a good part are copies of the same bits.
Where all this data physically resides
IDC divides storage into three layers, a model that maps surprisingly well over how we think about architecture:
- Endpoints — phones, laptops, PCs, IoT devices. Surprisingly much information lives here, but fragmented.
- Edge — institutional servers, telecom towers, CDN POPs, company servers. The layer that brings data closer to the user.
- Core — classic data centers and cloud. This is where the “serious” weight concentrates, in a few hundred hyperscale facilities globally, with the US hosting almost 40% of them.
The transition from endpoint to core is exactly the cloud migration we all experience: about 60% of corporate data already resides in the cloud, not on on-prem servers.
What this means for us as developers
Beyond trivia, these numbers have direct practical implications in how we design systems:
- “Storage is cheap” is a lie at scale. At hobby level, yes. But the factor replication × retention × regions quickly turns gigabytes into real money. Retention policy — what you delete and when — becomes an architectural decision, not an administrative one.
- The cheapest data is the one you don’t write. Verbose logs, telemetry without sampling, columns no one reads — all add up. The “created vs. stored” model reminds us that not everything passing through the system deserves persistence.
- Replication has a cost you must consciously choose. In a multi-region Cosmos DB or a Blob GRS, you pay for each copy. Redundancy is a business choice (RPO/RTO), not a default to blindly check.
In other words, the same forces inflating the global datasphere to hundreds of zettabytes manifest at small scale in any SaaS we build.
How these numbers are actually estimated?
A moment of skepticism is warranted: how does anyone know the world produces “181 zettabytes”? Obviously, no one counts bits one by one. Estimates combine several methods, each with its margin of error:
- From hardware shipments. Installed storage capacity is approximated based on how many HDDs, SSDs, and tapes are sold annually — manufacturers report total shipped capacity. This is the most solid base because it’s physically measured.
- From network telemetry. Traffic (data “in motion”) is estimated from measurements by operators and major traffic exchange points, extrapolated globally. This includes models like the old Cisco VNI.
- From category modeling. Each source’s production is estimated separately — how many hours of video are uploaded, how many IoT sensors are active, how many transactions — and summed with assumptions about bitrate and frequency.
That’s why numbers differ between reports by 10–15%: some count “created and replicated,” others “unique created,” others “installed capacity.” When you see two sources disagree, usually neither is wrong — they measure different things. Just like when two colleagues report “database size” and one includes indexes and replicas, and the other doesn’t.
Bonus: how much does the internet weigh?
Since we’re talking about sizes, the inevitable question: if you could weigh the internet, how much would it be? There are two amusing answers.
As pure information, a famous estimate (popularized by physicist Russell Seitz) says that the electrons encoding, at any moment, the data in motion on the internet would weigh in total about 50 grams — as much as a large strawberry. All human digital knowledge, in one fruit.
As physical infrastructure, another story: cables, servers, data centers, the global network — estimates speak of over 100 million tons. The difference between 50 grams and 100 million tons is exactly the difference between information and the machinery that carries it.
So, how many TB?
The honest answer: it depends on what you count. The indexed web Google sees — a few petabytes of text. Data created and consumed in a year — ~181 zettabytes, that is 181 billion TB, on track for ~394 ZB by 2028. Data actually stored long-term — a fraction of that, full of copies of the same bits.
Next time someone gives you a single number for “how big is the internet,” ask what exactly they are counting. They probably didn’t think about it — but you, now, do.
Sources for numbers: IDC Global DataSphere (via Statista), WorldWideWebSize Project, The Conversation. Numbers for 2025–2028 are projections and vary between sources by ±10–15%.
