Meta states that storage is acting as a brake on AI because its performance increase has been slow compared to AI compute speed tripling every two years. Director of Engineering Sidharth Bajaj and SW Development Engineer Venkatraghavan Srinivasan, writing in Meta’s AI Storage Blueprint at Scale document, say storage bottlenecks are a primary contributor to GPU stalls for AI workloads. Meta’s BLOB storage architecture operates on top of a foundational block layer called Tectonic, which is a regional, multi-tenant storage fabric. To address high-latency metadata fetches that can stall GPUs, Meta engineers rewrote the metadata subsystem and collapsed the metadata spread across different layers into one unified and flat schema backed by ZippyDB, reducing the number of metadata lookups. The API server no longer proxies data to the client; instead, they built a fat client SDK capable of streaming bytes directly from storage servers.