At the Supermicro Open Storage Summit 2025, theCUBE’s Rob Strechay hosts an in-depth discussion with Vince Chen of Supermicro, Praveen Midha of SanDisk, Kevin Kang of AMD and Balaji Venkateshwaran of DDN on how to architect storage for the era of agentic AI. The conversation examines the unique storage demands of AI agents that can plan, reason and act autonomously, highlighting why low-latency, high-bandwidth infrastructure is essential for real-time inference, retrieval-augmented generation and multi-agent orchestration. Kang details AMD’s full-stack approach spanning EPYC CPUs, Instinct GPUs, SmartNICs and ROCm software to accelerate data movement from storage to GPU memory while optimizing for bandwidth and IO. Midha breaks down how KV cache management and flash-based data lakes can slash time-to-first-token latency and improve efficiency, while new SSD swim lanes address both performance-intensive and capacity-focused workloads. Venkateshwaran shares how DDN’s Infinia data intelligence platform handles multi-protocol, multi-tenant environments at scale, delivering sub-millisecond object storage latency and tangible 22x RAG pipeline speedups. Chen outlines Supermicro’s role in delivering pre-validated, scalable AI data center architectures with partners, designed for fast deployment, seamless scaling and high energy efficiency. From minimizing data movement and power draw to ensuring scalable, future-proof architectures that can adopt next-gen GPUs without forklift upgrades, the panel offers concrete strategies for building storage backbones capable of keeping pace with AI innovation.