For years, the high-performance computing (HPC) world was stuck in a "traffic jam."
They built the on a radical idea: a parallel, distributed architecture that runs on standard x86 servers. Instead of a dedicated storage appliance, Weka turned every server’s local flash into part of a giant, unified pool. It was a "shared-nothing" architecture that used NVMe-over-Fabrics (NVMe-oF) to connect everything at near-memory speeds.
At the heart of the company’s success is , a distributed, parallel file system purpose-built for flash and cloud environments. Unlike traditional Network Attached Storage (NAS), WEKA uses a proprietary VFS driver to support full POSIX semantics with lockless queues. This architecture allows it to deliver the performance of local NVMe storage while maintaining the scalability of a distributed system. Scalability : Seamlessly scales from terabytes to exabytes .
On one side, you had the "Engines": Graphics Processing Units (GPUs) from companies like NVIDIA. These were becoming impossibly fast, capable of processing massive AI models and simulations in the blink of an eye.