📌 AI Inference Workloads Reshape Storage: Enterprise SSD Shortage May Exceed Expectations
The pull of AI on storage hardware is spilling over deeply from HBM in the model training phase to the inference application side. The core driver of this shift lies in the proliferation of AI Agents: long-context mechanisms cause KV cache data volumes to balloon dramatically, and the requirements for high-speed read/write operations are forcing data centers to accelerate the abandonment of mechanical hard drives and transition toward enterprise SSDs.
Volume and price data directly corroborate this substitution trend. Research firm TrendForce projects that enterprise SSD demand in Q4 will grow by more than 80% year-over-year, driving NAND contract prices up 15% to 20% quarter-over-quarter. This robust demand is reshaping the revenue structures of major vendors. For example, market expectations suggest quarterly data center revenue for SanDisk (SNDK) could expand to around USD 8 billion by 2028, accounting for nearly half of its total revenue, tracking the explosive trajectory previously exhibited by Micron.
The more critical divergence lies in the duration of the cycle. Previously, sell-side models generally assumed NAND would return to oversupply by the end of next year, but feedback from upstream original manufacturers points in the exact opposite direction. Yangtze Memory Technologies (YMTC) noted that, constrained by extremely severe capacity and capital expenditure limitations, this shortage could persist for another three years.
It should be noted that this rally is not without hidden risks. The sharp rise in storage prices is inflating full-system costs for PC and smartphone makers. Going forward, close attention must be paid to whether downstream customers engage in double ordering and excessive stockpiling, which could pull forward future demand and amplify the risk of a cyclical boom-and-bust reversal.
▌ Sources