Training, inference, RAG, and AI agents all consume data differently. Yet many organizations try to run them on infrastructure optimized for only one of those workloads, leading to performance bottlenecks, operational complexity, and unnecessary data movement. In this session, attendees will learn how data access patterns evolve across the AI lifecycle, from file and object storage for training, to KV cache for inference, to vector search, retrieval, and persistent memory for agents, and how to architect infrastructure that supports them within a unified platform. Drawing on deployments supporting large-scale AI environments, we will present a practical framework for matching storage, metadata, and memory architectures to specific AI workloads.
Speakers: Alon Horev, Co-Founder & CTO, VAST Data
