AIStor: Hyperscale Object Storage

AIStor is a distributed object store engineered specifically for high-performance, exascale AI workloads. This white paper explains the architectural decisions that shape it, beginning with the design philosophy borrowed from hyperscale cloud providers: single-layer, purpose-built object architecture that avoids the compromises introduced by general-purpose storage systems. It covers AIStor's performance characteristics for AI-centric workloads, including data lakehouses, advanced analytics, and training pipelines, its lightweight container-native design, and its deployment flexibility across private cloud, public cloud, Kubernetes distributions, and edge environments. The paper goes beyond feature coverage to explain the architectural reasoning behind each design choice, making it essential reading for storage architects and infrastructure engineers evaluating object storage for exascale AI infrastructure.