Microsoft Project Silica is an archival storage technology that encodes data permanently inside quartz and borosilicate glass using femtosecond laser pulses, achieving a density of 4.84 terabytes per palm-sized glass square equivalent to roughly 2 million books stored in an object you can hold in one hand, with accelerated aging tests confirming data stability beyond 10,000 years.
In February 2026, Microsoft Research published findings in Nature confirming a major leap forward: the technology now works on ordinary borosilicate glass, the same material used in kitchen cookware, removing two of the biggest barriers to scale. The Superman movie has already been preserved in glass. Government archive pilots are underway. And the world’s growing mountain of data, projected to reach 175 zettabytes by 2025, finally has a storage candidate that does not decay on a five-year replacement cycle.
Here is exactly how it works, what it costs to compare against existing options, and when you can actually expect glass storage to reach a data center near you.
What Microsoft Project Silica Is and How It Works
Project Silica is a Microsoft Research initiative that writes data into the molecular structure of glass using laser pulses measured in femtoseconds, which are quadrillionths of a second. The result is a write-once-read-many (WORM) storage medium with no moving parts, no power requirement during storage, and no practical degradation timeline under normal conditions.
The physics here are worth understanding. Under normal conditions, a femtosecond laser beam passes through glass without interacting with it. But when those ultrashort pulses are tightly focused on a single interior point, the intense electric field at that focal zone permanently alters the molecular structure of the glass. That modification is called a voxel, the three-dimensional equivalent of a pixel, and each one is typically less than one millionth of a metre on a side.
Microsoft researchers developed two distinct types of voxels for this system. Birefringent voxels encode data through changes in light polarization, using laser-driven micro-explosions inside the glass to achieve a storage density of 1.59 gigabits per cubic millimeter. Phase voxels take a different approach: they modify the refractive index of the glass at the focal point, requiring only a single laser pulse per voxel rather than multiple pulses. That single-pulse method is significantly faster and less expensive to operate, which is a critical step toward commercial viability.
Reading the data back uses polarization-sensitive microscopy with ordinary light, and machine learning algorithms decode the patterns created as light passes through modified voxels. The February 2026 Nature paper also simplified the reader from a three-camera setup to a single camera, which reduces cost and complexity further. Writing speed for phase voxels currently sits at approximately 65.9 megabits per second, far below what you get from a solid-state drive but appropriate for cold archival storage where write speed is not the primary requirement.
The Storage Numbers: How Much Data Fits in One Piece of Glass
A single glass square measuring 120mm by 120mm by 2mm, roughly the size of a drink coaster, holds 4.84 terabytes of data encoded across 301 layers within that 2mm depth. To put that in practical terms, 4.84 terabytes is approximately 200 4K movies, 2 million books, or roughly 1 billion average-sized photos. The raw theoretical potential scales even higher, with estimates suggesting up to 7 terabytes in a DVD-sized glass platter using further optimizations.
The 2 million books figure comes directly from Microsoft Research and was cited in the February 2026 Nature paper. It is calculated against an average book size of roughly 1.5 megabytes as plain text, so the actual usable book count varies depending on format, metadata, and encoding overhead. The headline number holds because it reflects raw data capacity at 1.59 gigabits per cubic millimeter across hundreds of layers, not just a marketing approximation.
The earlier proof-of-concept used expensive fused silica glass and stored Warner Bros.’ 1978 Superman movie, 75.6 gigabytes of data, in a glass piece just 75mm by 75mm by 2mm. The 2026 breakthrough is a different order of magnitude both in capacity and in material, moving from a laboratory-grade substrate to borosilicate glass widely available through commercial supply chains.
Glass Storage vs. Hard Drives, Tape, and DNA: A Direct Comparison
Every storage technology makes tradeoffs between capacity, lifespan, speed, cost, and access pattern. Glass storage is not designed to replace the SSD in your laptop. It is designed to replace magnetic tape in archival systems, and when measured against that specific job, the comparison looks like this:
| Storage Type | Capacity | Lifespan | Estimated Cost | Random Access Speed |
|---|---|---|---|---|
| Silica Glass (Project Silica) | 4.84 TB per coaster | 10,000+ years | High (R&D phase, no commercial price) | Very slow (write ~66 Mbit/s) |
| Hard Drive (HDD) | 20–30 TB | 3–5 years | $15–$25 per TB | ~160 MB/s |
| Magnetic Tape (LTO-9) | 18–45 TB per cartridge | 30 years | $5–$8 per TB | Sequential only, no random access |
| SSD (NVMe) | 4–8 TB consumer | 5–10 years | $50–$80 per TB | ~7,000 MB/s |
| DNA Storage | 45 zettabytes per gram (theoretical) | Millennia | Extremely high (not commercial) | Extremely slow, lab-only |
The honest limitation of glass storage is write speed. At 65.9 megabits per second, writing 4.84 terabytes to a single glass square would take roughly 16 hours. That is not a problem for archival use cases, where data gets written once and read rarely, but it rules out glass as a primary storage tier for any active workload. Magnetic tape, the current benchmark for cold archival storage, lasts about 30 years. Glass storage at 10,000-plus years is roughly 333 times more durable. That durability gap is the entire commercial argument for the technology.
DNA storage is theoretically denser than glass by several orders of magnitude, with supplier Biomemory citing 45 zettabytes per gram as a theoretical ceiling. But writing and reading DNA data requires wet lab processes that are too slow, too expensive, and too fragile for data center deployment at any scale. Glass storage, by contrast, uses established optical and laser manufacturing processes that industrial suppliers already understand how to produce.
How Long Glass Storage Actually Lasts
The 10,000-year figure for Project Silica durability comes from accelerated aging tests conducted by Microsoft Research, in which glass samples with written voxels were exposed to elevated temperatures, humidity, and environmental stress designed to simulate long-term degradation. The voxels showed no measurable data loss under those conditions, leading to the 10,000-year projection.
For context on what that means practically: a hard drive typically fails within 3 to 5 years without climate-controlled storage. Magnetic tape in ideal archive conditions lasts 30 years before requiring migration to new media. LTO-9, the current state-of-the-art tape standard, is rated at roughly 30 years with proper handling. Optical discs such as M-Disc claim 1,000-year archival lifetimes but hold between 25 and 100 gigabytes per disc. Glass storage exceeds all of these by a factor of at least 10, and does so in a medium that is immune to electromagnetic fields, does not require a climate-controlled environment to the same degree as tape, and does not degrade from oxidation or biological contamination.
The glass itself is also physically durable in ways that magnetic media are not. Microsoft Research confirmed that glass samples were subjected to boiling water, microwave radiation, oven temperatures, flooding, scouring, and demagnetization without data loss. The Superman movie glass from 2019 survived all of those tests intact. That resilience profile is why governments and national archives are paying attention.
What Is Already Stored in Silica Glass
The first real-world demonstration of Project Silica came in 2019, when Microsoft Research partnered with Warner Bros. to store the entire 1978 Superman film on a single glass square measuring 75mm by 75mm by 2mm. The encoded file contained 75.6 gigabytes of movie data plus error redundancy codes. After writing, the glass was subjected to the boiling, microwaving, and flooding tests described above. The movie remained fully readable after each test.
Since then, Microsoft has partnered with the Global Music Vault to explore permanent preservation of music recordings in silica glass, a direct application of the technology for cultural heritage archiving. A student project called Golden Record 2.0 also encoded historical audio and imagery in glass as a demonstration of the archival use case.
Government pilots are in progress. Microsoft has confirmed that specialized applications including national archives and intelligence agency data repositories are in the planning or early deployment phase for 2025 to 2027. These are exactly the use cases where 10,000-year durability and electromagnetic immunity justify a high per-unit cost: irreplaceable historical records, classified long-term data repositories, and scientific datasets that need to survive regime changes, natural disasters, and infrastructure failures.
The Internet Archive, which preserves hundreds of petabytes of web history and digital media, represents the type of institution that glass storage was built for. No partnership between Project Silica and the Internet Archive has been officially announced as of March 2026, but the fit is obvious and the archive community is watching the technology closely.
When Glass Storage Might Actually Go Commercial
Microsoft Research declared the research phase of Project Silica complete following the February 2026 Nature publication. That is meaningful progress. But complete research does not equal commercial product. As of March 2026, Microsoft has not announced a product, a price, a manufacturing partner, or a release window for glass storage in any Azure tier or standalone form.
The honest assessment from analysts who cover storage is that glass storage still needs three to four more developmental stages before it can compete with tape in a production data center. Those stages include scaling up laser writing hardware for industrial throughput, bringing borosilicate glass manufacturing into data-center-grade tolerances, building a full error correction and retrieval stack that works at petabyte scale, and establishing a cost-per-terabyte that undercuts tape’s current pricing of roughly $5 to $8 per terabyte.
The borosilicate breakthrough significantly shortens the timeline. Fused silica, the original substrate, is expensive and produced in limited quantities. Borosilicate glass is produced at industrial scale for consumer products and scientific equipment, which means the supply chain for glass storage media already exists. Laser writing hardware is the remaining bottleneck: femtosecond lasers capable of writing at the required precision are not yet manufactured at the volume or cost point that a mass-market storage product requires.
The realistic window for specialized commercial deployment, think national archives and intelligence agency systems, is 2027 to 2030. For broad Azure integration as a cold storage tier, industry observers are looking at the early 2030s at the earliest, assuming continued investment and no major technical setbacks. That timeline may compress if a major hyperscaler other than Microsoft decides to invest in competing glass storage research, as competitive pressure tends to accelerate hardware development cycles.
Frequently Asked Questions About Glass Data Storage
How much data can Microsoft Project Silica actually store?
Microsoft Project Silica currently stores 4.84 terabytes of data in a palm-sized glass square measuring 120mm by 120mm by 2mm, encoded across 301 layers at a density of 1.59 gigabits per cubic millimeter. That is equivalent to approximately 2 million books, 200 4K movies, or 1 billion average photos in a single piece of glass you can hold in one hand.
How does a femtosecond laser write data into glass?
A femtosecond laser fires pulses lasting quadrillionths of a second, focused to a precise point inside the glass. The intense electric field at that point permanently alters the molecular structure, creating a microscopic change called a voxel. Each voxel, often less than one millionth of a metre across, encodes binary data through changes in the glass’s optical properties. The modification is permanent and cannot be erased.
Is glass storage better than magnetic tape for archives?
For permanent archival use, glass storage outperforms magnetic tape in almost every durability metric. LTO-9 tape is rated for roughly 30 years under proper storage conditions, while Project Silica glass holds data for over 10,000 years based on accelerated aging tests. Glass is also immune to electromagnetic fields, heat, flooding, and demagnetization. Tape wins on cost and write speed for now, but glass is the superior choice for truly permanent records.
When will Microsoft Project Silica be commercially available?
As of March 2026, Microsoft has not announced a commercial product, price, or release date for Project Silica. The research phase was declared complete with the February 2026 Nature paper, but three to four additional development stages are needed before market deployment. Specialized government archive pilots are planned for 2025 to 2027, with broader commercial availability unlikely before the early 2030s.
What real data has already been stored in silica glass?
The most prominent example is the entire 1978 Warner Bros. Superman movie, stored as a 75.6-gigabyte file in a 75mm glass square during a 2019 proof-of-concept demonstration. Microsoft has also partnered with the Global Music Vault to archive music recordings in glass, and a student project called Golden Record 2.0 used silica glass to preserve historical audio and imagery. Government archive pilots are in progress as of early 2026.