Agent-Era NAND Storage

Agent-Era NAND Storage is the source’s claim that NAND becomes more important when AI agents run long, recoverable workflows. In 存储三巨头破万亿市值，存储超级周期何时能见顶？｜ S10E13, the guest says agent reasoning chains need intermediate results saved to disk, turning NAND from a final-output store into part of the inference process.

This is different from Persistent Agent Memory in product experience. Persistent agent memory is about what the agent remembers for the user or organization; Agent-Era NAND Storage is about the physical infrastructure that preserves intermediate state, checkpoints, retrieval material, and recoverable work during long agent execution.

Key Claims

• Longer agent workflows increase the need to save and recover intermediate state.
• NAND demand can rise even when HBM remains the low-latency layer for KV cache and active computation.
• The role of NAND shifts from passive storage toward a performance-sensitive component in AI inference systems.
• High Bandwidth Flash is one possible extension of this NAND role, but the source says it still cannot fully replace HBM.

Connections

• AI Data Center Memory Hierarchy - where NAND sits below DRAM and HBM.
• High Bandwidth Flash and SanDisk - high-bandwidth NAND-derived route.
• AI Inference Cost Structure, Agentic Workflow, Context Engineering, and Persistent Agent Memory - agent-workload context.
• Memory Chip Shortage and AI Hardware Supply Chain Pressure - market pressure created by storage demand.