97% of Global Data is Digital—A New Era of Storage

The most cited measurement of this shift comes from Martin Hilbert and Priscila López, whose 2011 paper in Science (vol. 332), "The World's Technological Capacity to Store, Communicate, and Compute Information," calculated that 99% of stored information in 1986 was analog. By 2007 that share had collapsed to 6%, with 94% already digital. The 97% threshold was crossed around 2014, and by 2026 less than 1% of new information is committed to non-digital media.

From Punch Cards to Petabytes

Storage hardware has compressed by roughly nine orders of magnitude in 70 years. Herman Hollerith's punch cards tabulated the 1890 US census on cardboard. IBM shipped the 726 magnetic tape unit in 1952, then the RAMAC 305 in 1956 — a 5 MB hard drive the size of two refrigerators that leased for about $35,000 a year in 1957 dollars. The compact disc arrived in 1982 at 650 MB, Toshiba introduced NAND flash in 1987, and consumer SSDs went mainstream around 2008.

Today a single Seagate Exos Mozaic drive, released in 2024, holds 30 TB in a 3.5-inch enclosure. That is six million times the capacity of RAMAC in a fraction of the volume, at roughly 1/3000th the inflation-adjusted price per gigabyte.

Zettabytes and the Cloud

IDC's Data Age 2025 white paper projected the global datasphere would reach about 175 zettabytes by 2025, up from roughly 2.7 ZB in 2012. Current estimates put the 2026 figure above 200 ZB, where one zettabyte equals a trillion gigabytes.

• Hyperscale cloud: AWS S3, Azure Blob, and Google Cloud Storage each measure individual regions in exabytes, with object counts in the trillions.
• Enterprise on-prem: shrinking share of the total, but still the home of regulated archives in finance and healthcare.
• Edge and device: phones, cars, and sensors generate the majority of new bytes, most of which are discarded within hours.

The Power Bill

Storing this much data is not free. The International Energy Agency estimated in 2025 that data centers consumed about 415 TWh in 2024, around 1.5% of global electricity, with AI training and inference workloads on track to roughly double that figure by 2030. Cooling water, grid capacity, and on-site gas turbines are now line items in hyperscaler earnings calls, not footnotes.

One gram of DNA could, in principle, store about 215 petabytes — the equivalent of roughly 200,000 large hard drives.

What Comes After Flash

A 2019 collaboration between Microsoft Research and the University of Washington wrote and read 1 MB of data using synthesized DNA strands, demonstrating that one gram of DNA could in theory hold about 215 petabytes. The catch is throughput: writing speeds remain in the kilobytes-per-hour range, several million times slower than a modern SSD, and reagent costs are still measured in thousands of dollars per megabyte.

Glass-based archival media from Microsoft's Project Silica, holographic storage, and racetrack memory are all in the lab. None will displace NAND or spinning rust this decade. The practical near-term story is denser HMR and HAMR drives at the cold layer, QLC and PLC flash at the warm layer, and DRAM plus HBM at the hot layer feeding GPUs.

The 97% milestone marks the point where analog became the exception. The next milestone is whether the planet can power what comes after it.