# Quobyte

> Quobyte is a software company that turns commodity x86 or ARM servers into a unified, high-performance, software-defined storage system spanning file, block, and object interfaces. Founded in 2013 by two ex-Google engineers who previously built the open-source XtreemFS parallel file system, Quobyte applies hyperscaler design principles - resilient software rather than exotic hardware - to enterprise, HPC, and AI storage. Its Data Center File System combines NVMe and HDD tiers in one architecture and is used by organizations including Siemens Healthineers, robotaxi maker Zoox, the UK's STFC/JASMIN supercomputer, and Yahoo Japan.

- **Founded:** 2013
- **Headquarters:** Santa Clara, California, United States
- **Founders:** Björn Kolbeck (Co-Founder & CEO), Felix Hupfeld (Co-Founder & CTO)
- **Team size:** ~22 employees
- **Products:** Quobyte Data Center File System, Unified Hybrid Storage (NVMe + HDD), GPU Converged Storage, Quobyte v4
- **Notable:** Published the fastest unverified MLPerf 3D U-Net storage result: 24 H100 GPUs per client on 4 standard servers, ~33% faster than the next-best published result., Powers a single 90 PB Quobyte cluster with 1.5bn+ files on the UK STFC's JASMIN supercomputer., Runs the storage behind Siemens Healthineers' 60+ FDA-approved AI models across 1bn+ datasets.

## Products & services

- **Quobyte Data Center File System** — A software-defined, distributed storage system that turns commodity x86 or ARM servers into a unified platform offering file, block, and object access with linear scalability, fault tolerance, and self-healing.
- **Unified Hybrid Storage (NVMe + HDD)** — Combines flash and disk in one architecture with policy-based automatic data placement and cloud tiering - 'NVMe Speed, HDD Cost.'
- **GPU Converged Storage** — Runs the storage software directly on GPU nodes, including NVIDIA Grace Blackwell and Hopper systems, shortening the data path to AI training jobs.
- **Quobyte v4** — Major release adding cloud object storage, ARM support, and end-to-end observability, with an AI-training focus.

## Achievements

- Published the fastest unverified MLPerf 3D U-Net storage result: 24 H100 GPUs per client on 4 standard servers, ~33% faster than the next-best published result.
- Powers a single 90 PB Quobyte cluster with 1.5bn+ files on the UK STFC's JASMIN supercomputer.
- Runs the storage behind Siemens Healthineers' 60+ FDA-approved AI models across 1bn+ datasets.
- Enabled Zoox to scale autonomous-vehicle AI training storage from 2 PB to 30 PB.
- Founders are the creators of the open-source XtreemFS parallel file system.
- Backed by Samsung Catalyst Fund, ALSTIN Capital, Target Partners, and High-Tech Gründerfonds.

## Latest updates

- **2025-06** — Published first public MLPerf Storage figures, claiming the fastest and most efficient result on the demanding 3D U-Net AI-training benchmark.
- **2025-04** — Released Quobyte v4, adding cloud object storage, ARM support, and end-to-end observability with an AI focus.
- **2025** — Rolled out GPU-converged storage running directly on NVIDIA Grace Blackwell and Hopper GPU nodes.
- **2025** — Robotaxi developer Zoox adopted Quobyte to store vehicle sensor and simulation data for AI training.

## Links

- Website: https://www.quobyte.com
- LinkedIn: https://www.linkedin.com/company/quobyte
- Twitter/X: https://twitter.com/quobyte
- GitHub: https://github.com/quobyte
- YouTube: https://www.youtube.com/c/Quobyte
- Facebook: https://facebook.com/quobyte

---

Profile page: https://yespress.io/quobyte
Published by YesPress — https://yespress.io
Last updated: 2026-07-10