The AI wave is reshaping the global storage market across three dimensions: demand structure, technology roadmap, and supply landscape. North American Agent AI and China’s KV cache offloading constitute two major demand engines, QLC products are accelerating their rise and driving the reconstruction of AI infrastructure around QLC, while NAND manufacturers’ capital expenditure shifting toward DRAM and HBM has led to a continuously widening supply-demand gap, with upward price pressure on eSSD expected to persist through the fourth quarter. Against this backdrop, major NAND manufacturers such as Samsung Electronics and SK Hynix (SKHY) are expected to continue benefiting from AI-driven structural growth in storage demand.
The demand structure has also undergone a qualitative shift: high-capacity, low-cost QLC products are accelerating their rise, standing side by side with the high-performance TLC products that previously dominated the market, marking the expansion of eSSD application scenarios from AI model training and inference to a broader range of AI application layers. Industry insiders said that amid NAND manufacturers’ strengthened capacity and pricing controls, upward price pressure on eSSD is expected to persist through the fourth quarter.
Industry experts pointed out that eSSD demand in the second half of the year will be driven by two main threads — the expansion of Agent AI in North America and the application of KV cache offloading technology in China. In North America, with the accelerating adoption of Agent AI, demand for large-scale real-time data retrieval and caching is rising sharply. TrendForce analysis noted that when Agent AI performs autonomous decision-making and actions, it needs to quickly access and store large amounts of contextual data, which will significantly boost demand for high-performance eSSD. Citi Research also pointed out that continuous learning technology requires AI systems to retain historical data while continuously absorbing new information, thereby generating sustained demand for high-capacity storage.
In China, AI companies represented by DeepSeek are actively promoting KV cache offloading solutions, a technology that migrates lower-priority cache data from high-performance memory to high-speed, high-capacity QLC solid-state drives during inference, in order to reduce dependence on expensive high-performance memory while improving model inference efficiency and compressing overall costs.
The core feature of this round of demand growth is the comprehensive rise of QLC products. Cloud service providers are actively leveraging QLC to balance high capacity and cost efficiency, with application scenarios covering vector database construction — a core component of AI search and recommendation systems. TrendForce believes that this shift may drive the reconstruction of AI infrastructure design around QLC, with the scope of impact extending beyond the procurement adjustments of a few hyperscale cloud providers. Citi Research further noted that QLC-based near-GPU storage solutions will gradually gain market acceptance to support AI systems in achieving more efficient data retention and retrieval near accelerators. In addition, demand for advanced storage solutions such as XL Flash and HBF, as well as storage complementary technologies such as CXL, will also rise simultaneously.
From a longer-term perspective, Citi forecasts that high-density eSSD demand will grow by 53% and 41% year-over-year in 2027 and 2028, respectively, and replacement demand from HDD to SSD will also provide additional support for the adoption of high-density storage. Although consumer-grade NAND demand is under pressure due to weakness in the PC and mobile terminal markets, Citi believes that strong enterprise-grade AI demand is sufficient to offset this drag.
