Seeing the light: Ewigbyte’s optical archive storage technology and strategy

Fresh startup Ewigbyte’s optical archive storage technology and business model has some similarities with, but differs from, that of Cerabyte and we asked Ewigbyte Co-founder and Head of Operations Dr. Ina von Haeften some questions to tease out and understand the differences.

Blocks & Files: How does your technology differ from that of Cerabyte?

Ina von Haeften: Cerabyte and Ewigbyte address the same long-term archival problem; space, but have made different architectural and physical design choices. At a high level, there are similarities: both use ultrashort-pulse (USP) lasers, spatial light modulation, thin glass substrates, and optically readable patterns. These are shared building blocks in photonic storage research.

The differences begin at the recording philosophy and system boundary. Concerning the recording medium and write process, Cerabyte writes data into a ceramic recording layer deposited on glass; the laser modifies the coating while the glass substrate remains unchanged.

Ewigbyte deliberately avoids coatings and writes directly into uncoated glass using UV-range ultrashort pulses, creating nanoscale engraved structures in the glass itself. The data is therefore physically embedded in the substrate rather than stored in a super thin surface layer. This choice results in inherently immutable data representation designed for very long retention (century- to millennial-scale) and enables simpler end-of-life recycling, since the medium is pure glass rather than a composite material.

Regarding the system architecture and business model, Cerabyte is oriented toward selling media and write/read hardware into customer-operated environments. Ewigbyte does not plan to sell writing machines as products. Instead, we operate the hardware ourselves and provide long-term writing, reading, and storage as a managed service.

On otimization priorities, Cerabyte emphasizes a density roadmap down to a few nm. Our view is that for photonic storage, density is not the primary constraint for cold data–instead it will always negatively impact write/read speed. Because glass requires no power, cooling, or active environment once written, capacity can be scaled horizontally using automated storage facilities rather than by pushing density to extremes.

We therefore optimize for write and read throughput at scale, as ingestion speed and retrieval throughput become limiting factors much earlier than physical density in AI-driven data growth.

Blocks & Files: What is the state of your technology in terms of being demonstratable?

Ina von Haeften: We are planning to demonstrate the first operational rack mid of next year.

Blocks & Files: Ewigbyte started up eight months ago. How does that square with developing a photonics technology?

Ina von Haeften: It is important to distinguish between the age of the company and the maturity of the underlying work.

Before founding Ewigbyte, our CEO, Dr. Steffen Klewitz, spent roughly two and a half years working intensively on glass-based photonic storage, developing the technology from early concepts through to working prototypes, building supplier relationships, and exploring system-level storage implications. Ewigbyte therefore started from a more advanced technical blueprint rather than from scratch.

In terms of demonstrability today, we scope our claims carefully. We can demonstrate the core write and read principle at sample level: deterministic data patterns written directly into glass and read back optically using microscope-based readout. These demonstrations have been shown publicly at events such as Web Summit and the IT Press Tour, and to partners, customers, and investors. The emphasis is on deterministic encoding and independent readability at the physical data layer.

What is not yet demonstrated is an end-to-end, automated storage system with robotic handling and production-scale throughput. That is the focus of the current engineering phase. We have selected a system integrator, finalized the MVP architecture, and ordered the next-level prototype, with a target to demonstrate terabyte-scale writing on glass in the next phase (currently planned for September 2026).

Alongside this, we work with research and engineering partners including the University of Vilnius, Fraunhofer ILT, and the Max Planck Institute, and are supported through the TUM Venture Labs and UnternehmerTUM ecosystem.

In short, the technology is demonstrated at the physical data level, with system-scale automation and integration under active development.

Blocks & Files: Are you putting together existing technologies? What are they, and why shouldn’t others do the same?

Ina von Haeften: At the component level, many elements of the system are established, industrial building blocks: USP laser sources, motion and positioning systems, robotics, industrial vision, and control electronics. This is intentional, as it supports manufacturability and long-term serviceability.

The difficulty lies not in owning these components, but in making the full system work deterministically at high throughput over long time horizons. Two areas are particularly non-trivial.

• Optical write process and optics integration: ultrashort-pulse systems are commonly operated in infrared wavelengths with frequency conversion to green or UV, depending on the material interaction required. Our approach relies on a custom USP optical configuration and a tightly controlled process window for direct glass surface structuring. Engineering this for repeatable, high-throughput, “data-grade” writing introduces constraints around stability, alignment, calibration, and quality control. This is a core area of our IP, and we are in the process of filing multiple patents related to the optics.
• System-level know-how: much of the defensibility of archival infrastructure sits in system integration: verification workflows, error budgets, redundancy strategies, and the reliable robotic handling of fragile media over decades. A large part of this resides in engineering know-how and trade secrets rather than in a single component.

Finally, even the choice of glass is non-trivial. “Glass” encompasses many formulations with different optical and mechanical properties, and qualifying the right substrate for long-term, repeatable writing and reading is part of our R&D roadmap.

So, while others can source similar categories of components, the barrier lies in integration discipline and operational reliability, not in assembling a bill of materials.

Blocks & Files: What is the throughput, and what is the capacity of an Ewigbyte glass tablet?

Ina von Haeften: Media capacity: for the first-generation media, the design target is approximately 10 GB per glass tablet, with data written on both sides. This format prioritizes robustness and deterministic readability over maximum density. For volume density considerations, it is important to note that we are planning to work with 100µm glass sheets.

We are planning to move to larger wafers which then will more than double capacity per single carrier.

Blocks & Files: What are the read and write speeds?

Ina von Haeften: These are best separated into media characteristics, local performance, and system-level throughput.

Write and read speed (local): at the write/read head level, the current design target is around 500 MB/s per head for both writing and reading. These are MVP (Minimum Viable Product) targets rather than demonstrated production figures.

Throughput and scaling: throughput is achieved through parallelization. Each machine is designed to operate with up to eight parallel heads, giving a per-machine aggregate target of ~4 GB/s. System-level throughput then scales by running multiple machines in parallel. Over time, facilities are designed to accommodate many such machines concurrently; our long-term model assumes up to ~100 write/read machines operating in parallel.

End-to-end ingest speed also depends on how data arrives — for example, via physical media shipments or wide-area network transfers. Facility placement and connectivity therefore form part of the overall system design, with initial sites planned near major network hubs to support high-volume ingest alongside physical workflows.

Blocks & Files: You say Ewigbyte isn’t designed as media + library hardware, but surely there is hardware. What is the status of this hardware?

Ina von Haeften: Any physical archival system necessarily includes both hardware and software. Our distinction is not about whether hardware exists, but about how the system is exposed and operated.

Ewigbyte combines a hardware stack — optical write/read units, precision handling for glass media, automated storage — with a software layer that abstracts this infrastructure from the customer. We use a mix of off-the-shelf industrial components and custom subsystems, particularly in optics, calibration, verification, and glass handling.

On top of this, we are building software designed to integrate with existing storage platforms, such as object storage systems. This allows customers to archive cold data to different backends — tape, cloud services, or an immutable copy on glass — without changing application workflows.

Our default model is to operate the infrastructure ourselves as a managed service, including the full lifecycle of the glass media. At the same time, long-term archives depend on trust and retrievability. For this reason, we plan to make professional-grade readout equipment available over time so customers can independently access and verify their data, and the architecture also supports on-premises scenarios where required.

Blocks & Files: How is your facility optimized for cold data storage compared to a partitioned data centre with tape?

Ina von Haeften: Tape libraries deployed in conventional data centres still reflect a data-centre-first architecture. Very long-term cold storage benefits from different assumptions.

• Environmental requirements: magnetic tape relies on controlled and stable environmental conditions to meet long-term reliability expectations, which in practice contributes to planned refresh and migration strategies.
• By contrast, the glass medium itself is inert. Once written, the data does not depend on active climate control to remain readable, and there is no inherent need for scheduled migration driven by medium aging.
• Energy model: at rest, stored data consumes zero energy. Power is required only during write/read operations and for automation when media is moved. This differs fundamentally from tape libraries embedded in always-on data-centre environments.
• Fidelity and access: tape is sequential media and can require conditioning after long idle periods. With glass-based storage, the data representation remains physically unchanged over time and is accessed directly. Time to first byte is dominated by physical handling, while read throughput scales with the number of parallel read heads.
• Redundancy and operations: error-tolerant encoding, combined with multiple physical copies and later geographic separation, provides resilience. Facilities are designed for largely automated, “lights-out” operation with minimal on-site staff.
• Facility placement: because there is no need for continuous cooling or high-density power delivery, facilities do not require premium data-centre real estate, while still being placed near major network hubs where ingest bandwidth matters.

Bootnote

Ewigbyte styles itself “ewigbyte” but, being old-school, we have capitalized the first letter.