Vishwamitra Numbers
Issue #012
The famous 2.5-quintillion-bytes-a-day statistic is roughly 160 times smaller than a 2024-derived daily average, and it was never a peer-reviewed global measurement to begin with.
By Santhosh Vishwamitra
Scroll through any marketing deck, data-analytics blog, or "fun facts" thread on X, you will still find it: "The world creates 2.5 quintillion bytes of data every day." It sounds authoritative. It has a precise unit. It has been shared for years. The problem is that the figure is a historical IBM estimate from the early 2010s, its underlying calculation is not transparently documented in the IBM material that repeats it, and current Global DataSphere forecasts are orders of magnitude larger. A decade-old marketing statistic has outlived its usefulness by a remarkable margin. This article explains what the number actually was, what modern measurement frameworks say instead, and why the gap between the two matters more than either headline figure suggests.
Where the Number Came From
IBM published the 2.5-quintillion-bytes-per-day claim in the early 2010s, alongside the assertion that 90% of the world's data had been created during the previous two years. A 2011 IBM CMO study reproduced the same formulation, connecting the figure to social networks, sales data, market research and other digital sources. The statistic appeared in a corporate and business context, not as a peer-reviewed scientific measurement.
An IBM technical document repeats the figure and cites an older IBM web page, but does not show the underlying calculation. A 2018 Data Science Stack Exchange discussion likewise records that the original source and calculation could not be reconstructed from the material then available. That does not prove the number was fabricated. It does mean the figure should be treated as a historical IBM estimate rather than as a continuously measured global statistic.
The stronger warning applies to the accompanying "90%" claim. A UK government review of commonly quoted data statistics found that the statistic lacked a robust, accessible methodology and was already unverified and outdated. It is therefore not suitable as a current description of the global data landscape.
The safest editorial treatment of the 2.5-exabyte figure is this: IBM popularized an early-2010s estimate of 2.5 exabytes per day. It should not appear in a current-tense sentence as a factual description of what the world produces today.
Why 2.5 quintillion bytes equals 2.5 exabytes, and why that unit matters
In decimal storage units, one quintillion is 1018. One exabyte is 1018 bytes. So 2.5 quintillion bytes equals 2.5 exabytes. One zettabyte (ZB) equals 1,000 exabytes. One zettabyte equals 1021 bytes. The conversion matters because modern institutional forecasts describe the Global DataSphere in hundreds of zettabytes per year. The dramatic-sounding word "quintillion" can obscure how much larger a zettabyte-scale annual total is.
What Modern Measurement Actually Counts
The industry body most often cited for global data-volume estimates is IDC, through its Global DataSphere. IDC's taxonomy defines the framework around four broad activities: creation, capture, replication and consumption. The current taxonomy also contains additional segmentations by device, data type, location, cloud status and other dimensions.
| DataSphere Category |
What It Includes |
Common Misunderstanding |
| Creation |
Newly generated files, messages, photographs, videos, machine outputs and records |
Often assumed to be the only category being counted |
| Capture |
Sensor readings, telemetry, transactions and observations |
Frequently omitted from popular summaries |
| Replication |
Backups, cloud copies, caches, mirrors and other duplication |
Rarely mentioned; can substantially increase gross data volume |
| Consumption |
Streaming, searches, application activity and data movement |
Often confused with permanent storage |
This distinction matters because the DataSphere is a gross activity measure, not a count of unique information. A file can contribute to the measured volume when it is created, copied or consumed without representing a new piece of knowledge each time. IDC's framework should therefore not be treated as a stock of permanently retained information.
What the DataSphere measures is the scale of global digital activity represented by those categories over a year. That is useful, but it requires clear labelling. A derived daily average from an annual DataSphere forecast is not the same thing as a live planetary counter.
The Numbers That Actually Reflect 2024, 2025 and 2026
A 2024 IDC forecast put the Global DataSphere at approximately 149 zettabytes. A June 2026 IDC forecast, as reported by industry sources citing the IDC forecast document, places 2025 at approximately 218.4 zettabytes, 2026 at approximately 274 zettabytes, and 2030 at approximately 718.5 zettabytes. These figures are forecast vintages, not direct measurements of every byte produced in real time.
IDC does not publish an official single daily counter. When an annual figure is divided by 365, the result is a derived daily average. That arithmetic is useful for scale, but it must not be presented as an IDC-measured daily total.
| Figure |
Meaning |
Year / Scope |
Status |
| 2.5 quintillion bytes/day (= 2.5 EB/day) |
IBM historical estimate |
Early 2010s; global |
Historical claim; calculation not transparently documented in the cited IBM material |
| ~149 ZB/year (~408 EB/day derived average) |
Global DataSphere forecast |
2024; global |
IDC forecast reported by secondary institutional sources |
| ~218.4 ZB/year (~598 EB/day derived average) |
Global DataSphere forecast |
2025; global |
June 2026 IDC forecast, reported by secondary industry sources |
| ~274 ZB/year (~751 EB/day derived average) |
Global DataSphere forecast |
2026; global |
June 2026 IDC forecast, reported by secondary industry sources |
| 1.5 ZB/year mobile + 7.3 ZB/year fixed |
End-user internet traffic only |
2025; global |
ITU estimate; distinct from the DataSphere total |
| >220 EB/month mobile traffic |
Mobile network data traffic only |
Q2 2026; global |
Ericsson estimate; a measured subset of global data activity |
| ~718.5 ZB/year |
Projected Global DataSphere |
2030; global |
June 2026 IDC forecast, reported by industry sources |
The comparison is stark. The 2024 Global DataSphere forecast implies a derived daily average of roughly 408 exabytes. The old IBM figure was 2.5 exabytes per day. That is a ratio of approximately 163 to one. The 2025 June 2026 forecast implies roughly 598 exabytes per day, while the 2026 forecast implies roughly 751 exabytes per day. These are arithmetic conversions of annual forecasts, not official daily measurements.
Why ITU internet traffic figures are a separate measurement from the DataSphere
The ITU's Facts and Figures 2025 report estimates global end-user internet traffic at 1.5 zettabytes on mobile broadband and 7.3 zettabytes on fixed broadband, for a combined total of approximately 8.8 zettabytes in 2025. This is directly relevant, but it covers end-user Internet traffic. It does not represent the whole DataSphere, which also includes other forms of creation, capture, replication and consumption. Internet traffic is therefore a measurable subset of the broader digital activity landscape, not a substitute for the DataSphere estimate.
The AI Feedback Loop
For much of the past decade, major drivers of data growth included smartphones, video and connected devices. That picture has become more complicated since the rise of generative AI. AI systems consume data for training and retrieval, while deployed systems also generate logs, checkpoints, model artifacts, evaluation data and other operational records. Synthetic data can add another layer of machine-generated material.
The important point is not that AI has suddenly become the only source of data growth. It has not. IDC-related reporting describes AI as one driver within a broader shift that also includes digital services, richer applications, connected systems and longer retention requirements.
The feedback loop is therefore better described as an infrastructure cycle than as a single causal chain: more AI workloads create more data requirements; some AI workflows also create new data and artifacts; and organizations increasingly have reasons to retain useful historical data for future training, retrieval, compliance or evaluation. The precise share of total 2025 or 2026 DataSphere growth attributable specifically to AI-synthetic data remains uncertain, and this article does not assign a percentage to it.
What Gets Stored, and What Disappears
Not all data represented in the DataSphere is kept. IDC has repeatedly distinguished the annual DataSphere from the much smaller volume of data actually retained in storage. In its 2020 analysis, IDC said less than 2% of that year's newly created or replicated data was saved and retained into 2021; the remainder was largely ephemeral, consumed immediately, cached or overwritten. That historical figure should not be mechanically applied to 2026, but it illustrates the central distinction between data activity and persistent storage.
That distinction matters because the DataSphere is a flow, while persistent storage is closer to a stock. Much of what enters the digital system may disappear quickly. Some data is retained for days or years. Some is copied into multiple systems because it has operational, regulatory, security or future analytical value. The annual DataSphere total should therefore never be treated as the amount of information permanently added to the world's record.
The Infrastructure Consequence
At a derived daily average measured in hundreds of exabytes, the infrastructure implications are not abstract. Data centers require electricity and cooling, while AI workloads add substantial demand for compute and storage. The exact environmental footprint of the global DataSphere cannot be inferred from the DataSphere number alone, because the metric measures data activity rather than energy use. That distinction matters when discussing sustainability.
The tape-storage trend is a useful example of the physical response to rising retention requirements. The LTO Program reported that shipped tape capacity rose 57% year over year in Q1 2026, after 2025 annual shipments fell 9% from the 2024 record. The program linked the market to growing data, cost, energy and cyber-resilience pressures. AI and archive workloads are part of that demand, but the shipment rebound should not be attributed to AI alone.
Tape remains relevant because cold and archival data do not need the same access characteristics as active workloads. Its low operating cost and the fact that a stored cartridge does not require continuous spinning power make it useful for long-term retention. That does not make tape a universal replacement for disk or flash. It makes it one layer in a tiered storage architecture.
This is the infrastructure question the old 2.5-quintillion headline largely hides. Once data volumes are expressed in hundreds of exabytes per day as derived averages, the relevant questions become physical and economic: what must be retained, where should it live, how often must it move, and who pays for the energy and hardware required to keep it available?
Why the Old Number Will Not Go Away
There is a structural reason for the persistence of the IBM figure. A daily number is more emotionally immediate than an annual one. "The world creates 2.5 quintillion bytes every day" is a claim about right now, about this moment, about what is happening as you read this sentence. "The Global DataSphere reached approximately 149 zettabytes in 2024" is a claim about an annual forecast or accounting framework measured by an analyst firm using a taxonomy most readers have never encountered.
The marketing-language version wins on immediacy. The institutional version wins on precision. Most social-media posts and secondary sites are not primarily optimized for methodological nuance. IBM's original phrasing was already optimized for impact when it was first published. It has simply never been replaced by an equally vivid explanation of how modern DataSphere estimates actually work.
That is why the old statistic is worth preserving as history but not as a current measurement. It tells us something about how the early big-data era communicated scale. It does not tell us how much data the world produces today.
Our Approach
We aim to present global events with clarity, accuracy, and balance without taking sides. Every article is based on credible sources and verified information.
We do not follow the "breaking news" race. Instead, we focus on delivering thoughtful analysis and meaningful perspective on important developments.
If you notice any errors, feel free to reach out. We follow a transparent correction and update policy to ensure all changes are clearly reflected for our readers.
Editorial View · Santhosh Vishwamitra
The 2.5-quintillion-bytes-per-day figure should be retired as a current statistic and kept only as a historical marker. Its continued circulation is misleading because it presents an early-2010s estimate as if it were a current planetary measurement. The more honest starting point is this: no continuous global instrument counts every byte the world creates, captures, replicates and consumes each day. What we have instead are annual institutional forecasts, derived daily averages and increasingly detailed measurements of particular subsets such as Internet traffic and mobile network traffic. The useful question is no longer which giant number sounds most impressive. It is what the number actually measures, what it leaves out, and what the resulting infrastructure demands mean.
Stay Connected
Vishwamitra Chronicles Network
Original reporting on science, geopolitics, and history.
Santhosh Vishwamitra
A developer, analyst, and the Founder & Editor of Vishwamitra Chronicles, dedicated to independent, evidence-based research on geopolitics, science, and history. - About the Author
Comments
Post a Comment