Server Storage Hits Inflection as AI Shifts to Inference
Source: Counterpointresearch
Enterprise SSDs now represent nearly half of all NAND flash shipments, a shift driven by inference workloads replacing training as the primary computational burden in production AI systems. Inference requires different performance profiles—latency-optimized rather than throughput-maximized—and different geographies: edge and regional data centers rather than hyperscale clusters. This is forcing manufacturers like Samsung, SK Hynix, and Kioxia to retool production capacity and compete directly with consumer SSD makers on volume.