Meta’s BLOB-storage architecture evolved to address two primary challenges: maximizing GPU utilization and maximizing research velocity. Over the past several years, model capabilities and training dataset sizes have experienced exponential growth, and the time between new-frontier-model releases has gone down from months to weeks. While AI compute performance has roughly tripled every two years, storage and interconnect performance growth have been more modest, making storage bottlenecks a primary contributor to GPU stalls for AI workloads. Meta operates hundreds of exabyte-scale storage clusters serving products including Facebook, Instagram, Reality Labs, Meta AI, Ads, Data Warehouse, and internal Databases. Its storage service is built on a horizontally scalable foundational block layer called Tectonic, which provides high durability and availability leveraging erasure-coding techniques and supports tiering across media types. The BLOB-storage layers on top of Tectonic expose a global, infinitely scalable storage fabric. Meta’s modern training stack has been migrating onto the BLOB-storage interface, motivated by the need for unified storage access to massive data lakes and high performance.