Microsoft Project Silica writes 4.84TB into 2mm glass

Microsoft researchers say they have demonstrated a practical method for “writing” data into glass that could remain readable for more than 10,000 years—a claim framed, with a straight face, as preserving humanity’s most important records for whatever comes after us.

According to Euronews, the work—part of Microsoft’s “Project Silica” and presented in Nature—uses a high‑repetition femtosecond laser to encode bits as microscopic deformations (“voxels”) inside a 2‑millimetre‑thick piece of glass. The laser fires at 10 MHz (10 million pulses per second), with each pulse writing one voxel at the system’s maximum repetition rate. By shifting the focal depth, the team writes hundreds of layers through the glass thickness, reaching a reported capacity of 4.84 terabytes per small glass piece—Euronews translates this as roughly two million books.

Reading the medium is performed by sweeping the glass under an automated microscope with a camera, reconstructing the voxel patterns back into data. The pitch is that, unlike hard drives and magnetic tape—whose lifespans are finite and require periodic migration—glass could be a “set and forget” archival substrate.

Technically, this is not magic; it is an engineering trade that prioritizes durability over convenience. The system swaps today’s cheap, fast, frequently replaced storage for a write‑once medium that depends on precise optical writing and machine‑vision decoding. You’re not escaping infrastructure, you’re just choosing a different kind—lasers, microscopes, calibration, error correction, and a supply chain capable of building compatible readers centuries later.

That caveat matters because the “civilisation survival” narrative is a rhetorical solvent. Once you declare a problem existential, the solution tends to become institutional by default: centralized curation, official “most important” datasets, and large custodians who decide what deserves to be etched into glass and what can be left to rot on obsolete formats.

The medium is less interesting than the governance it invites. A glass archive is inherently less permissionless than the internet: it is physically scarce, curated, and—crucially—readable only to whoever controls the specialized tooling and the catalog. The same organizations that struggle to secure ordinary databases will have no shortage of confidence in their ability to curate “humanity’s memory.”

Project Silica may well become useful for cold storage of high‑value records—scientific datasets, legal archives, source code snapshots—especially where longevity beats access speed. But the story to watch is not the voxel density. It’s who gets to decide what goes into the glass, who controls the readers, and how quickly “archiving” becomes yet another justification for building data cathedrals that ordinary people can’t audit, can’t fork, and can’t leave.