A piece of glass about the size of a drink coaster, two millimetres thick, can now hold 4.84 terabytes of data. That is roughly the contents of a thousand films, or several million pages of text, sitting in something you could drop in a pocket. The number itself is not the interesting part. The interesting part is the shelf life: in accelerated ageing tests, the medium is projected to keep its data for more than ten thousand years.
The work comes out of Microsoft Research together with the Optoelectronics Research Centre at the University of Southampton, and the results were published in Nature in 2026. The medium is borosilicate glass, the same family of glass used for laboratory beakers and ovenware, chosen because it is chemically inert and tolerant of heat. Data is written with an ultrafast laser that fires into the glass and leaves behind tiny permanent nanostructures — voxels — that change how light passes through that point. Roughly 301 layers of these voxels are stacked through the two-millimetre thickness, giving a volumetric density of about 1.59 gigabits per cubic millimetre. Reading is done with polarised light and a machine-learning decoder that interprets the pattern back into bits.
The physical principle is why the lifetime claim is plausible rather than promotional. Optical media such as CDs and DVDs store data as a dye or a phase change in a plastic substrate, both of which decay. Magnetic tape and hard drives store data as magnetisation that slowly relaxes and is vulnerable to magnetic fields. Silica glass stores data as physical shape inside a material that does not react with air or water. The published work reports that one of these platters survived being baked at 290 degrees Celsius with no loss of data, which is well beyond what any conventional storage medium would tolerate.
Two practical numbers keep the technology in perspective. The first is write speed: about 8.2 megabytes per second, which means filling a single platter takes roughly eighteen and a half days of continuous writing. That is not a backup drive. The second is that this is a write-once medium — you cannot rewrite a voxel. Both limits point in the same direction. This is not a replacement for your hard drive; it is a candidate for the layer underneath it, the copy you write once and hope never to touch again.
That framing matters for anyone thinking in decades rather than years. Every serious archive today faces the same structural problem: the data outlives the format. Tapes need a working tape drive. Disks need a spindle. Cloud storage needs a company that still exists, a subscription that is still being paid, and a data centre with power. Each of those is a chain of dependencies measured in years, while the information we claim to be preserving is measured in centuries. Glass removes nearly all the dependencies except the reader. There is no spinning part, no chemistry, no magnet, no electricity required to hold the state — only a light source and a decoder to get it back.
There is an honest caveat in how Microsoft has described the work. The tone reads like a milestone being banked, not a product being promised. Ten-thousand-year projections come from accelerated ageing, which is an extrapolation, not a wait. And a medium that outlives every machine capable of reading it is, functionally, a fossil unless someone keeps the reader alive. Which is the old archive paradox in its purest form: the harder you make the object to destroy, the more the survival of the contents depends on the continuity of an institution.
The name of the game here is not capacity; it is the absence of maintenance. A tape archive in a climate-controlled vault is a commitment that must be renewed every year by someone with a budget. A glass platter in a cave is not. That difference is what makes the technology interesting to anyone thinking on a horizon longer than a career, and it is also why the project's own researchers describe the work as a foundation rather than a product. The remaining problems are the reader, the write speed, and the fact that nobody has yet demonstrated a migration path into a glass archive from whatever came before it. Solve the medium and you have solved perhaps half the problem.
Still, the capability on the table in 2026 is real. Somewhere between four and five terabytes of data, written into something the size of a beer mat, that does not care about humidity, acid, insects, mould, fire below three hundred degrees, or the passage of a hundred centuries. If a civilisation wanted to leave a record that could survive its own collapse and still be legible to whoever came next, this is the closest thing to that idea that has ever been built in a laboratory. The question it raises is no longer whether we can store something for ten thousand years. It is whether we can decide what deserves to be stored.
https://3026alive.com