jason hardy

Storage must shift from passive system of record to active AI data plane that ingests, cleanses, enriches, chunks, embeds, indexes, and governs data at the moment of creation — enabling agents to operate on accurate, real-time context rather than stale data.

Agentic AI creates a new paradigm where storage must continuously prepare, index, govern, and serve multimodal data at creation time — not just store it — requiring co-designed compute, networking, and GPU acceleration inside the storage layer to feed agents at machine speed.

Agents run long-running sessions (hours to years), consume data at GPU speed not human speed, blend SQL and PDFs, and generate persistent context (KV cache, memory, ontology) that must be governed, secured, and served in real time across hybrid environments.

Agents run long-running sessions (hours to years), blend structured and unstructured data, operate at machine speed far exceeding human interaction rates, and generate persistent context (KV cache, memory, embeddings) that must be stored, governed, and served in real time — fundamentally changing enterprise storage requirements.

Nvidia defines a five-tier memory hierarchy (HBM=G1, system RAM=G2, NVMe=G3, high-perf storage=G4, enterprise storage=G4 extended) and introduces G3.5 as a cost-effective KV cache tier that sits between GPU memory and network storage, reducing recompute and boosting token throughput.

A new storage tier (CMX) sits between GPU memory (G1/G2) and high-performance storage (G4) to offload KV cache, eliminating recompute on prefill/decode stages and scaling token output cost-effectively as AI factories grow.

Nvidia's Vera, Bluefield-4, Spectrum-X, and Rubin/RTX Pro GPUs are co-designed with storage partners (Dell, NetApp, IBM, HPE) to create turnkey AI data platforms where compute, network, and storage are integrated at the hardware and software blueprint level.

Vera's 3x memory bandwidth and best-in-class single-thread performance — originally built to feed GPUs — directly benefit storage workloads; Bluefield-4 STX integrates Vera with storage acceleration engines, enabling the AI data platform at the hardware level.