🤖 AI Summary
Phison CEO Pua Khein-Seng warned that an industry-wide NAND flash shortage beginning around 2026 could persist for the next ten years — a far longer disruption than most analysts’ one-year squeeze forecasts. He attributes this to underinvestment in flash after price collapses in prior cycles, a 2019–2020 pullback in capex and a 2023 pivot by big players (Micron, SK Hynix) into higher-margin HBM. Market signals already reflect tightness: TrendForce and vendors such as SanDisk, Micron and Western Digital have announced price freezes or hikes as demand surges from AI data centers.
Technically, Pua argues the root is shifting AI workload economics: training drove a GPU/HBM arms race, but matured models and the push to monetize via inference are creating massive nearline/warm storage demand for model versions, logs and user data — workloads dominated by NAND SSDs rather than HBM. He predicts SSD-to-HDD ratios in data centers rising from ~20% today toward 80–100%, with SSD prices potentially matching HDDs in 5–8 years; HDD lead-times and HDD capacity scaling (HAMR) are also constraints. The implication: sustained high NAND pricing, prolonged capacity expansion cycles, and major cloud capex reallocation into storage infrastructure — a potential “memory supercycle” that materially affects supply chains and AI deployment economics.
Loading comments...
login to comment
loading comments...
no comments yet