TrendForce indicates that AI development is shifting from large model training to Agentic AI applications centered on inference, driving a structural expansion in memory demand. As supply shortfalls are unlikely to be resolved in the short term, prices are being pushed upward. TrendForce has significantly revised upward its global memory market value forecast, raising the projected 2026 value to $889.3 billion and adjusting the 2027 estimate from $842.7 billion to over $1.28 trillion, representing a year-on-year growth of approximately 44%.

Analyzing changes in DRAM demand, inference requests in Agentic AI systems are shifting from single instances to continuous cycles. The capacity requirement for KV Cache expands proportionally with the Context Window. If recomputation is needed, computational costs would grow exponentially. Therefore, efficient KV Cache management has become crucial for enhancing AI inference performance, directly boosting demand for HBM and DRAM. Concurrently, as Agentic AI workloads significantly increase demands on CPU scheduling, data preprocessing, and memory management, the CPU-to-GPU configuration ratio in new-generation AI server platforms is gradually increasing from the previous approximate 1:8 to 1:4 or higher. For instance, the NVIDIA NVL72 rack adopts a 1:2 configuration. The rise in CPU usage will expand the demand for Server DRAM configurations, simultaneously driving up procurement volumes and contract prices. Furthermore, the increased wafer consumption for HBM production is constraining the available capacity for Conventional DRAM. Amid expanding demand, this strengthens suppliers' pricing power in contract negotiations, with the upward price trend estimated to be sustainable through 2027. Consequently, TrendForce has raised its 2026 DRAM market value forecast to $618.7 billion, with a year-on-year growth rate of 303%. The 2027 value is projected to further increase to $903.3 billion, growing 46% year-on-year.

In the NAND Flash sector, the combined capital expenditure of the world's nine major cloud service providers continues to rise, with an estimated growth of 79% in 2026. Capital intensity is also expected to increase to 34%, reflecting a shift in investment strategy from "capacity expansion as needed" towards large-scale deployment of AI infrastructure to secure long-term advantages. The underlying driver is also Agentic AI, including AI agents driving a doubling of enterprise usage, with token consumption for heavy users surging to four times the original level. The increasing complexity of media generation is also substantially raising token consumption. Faced with massive and continuously growing memory demand, the higher-cost HBM is unsuitable for large-scale deployment, while HDDs are constrained by access speeds and power consumption, unable to meet the real-time processing needs of AI data centers. This scenario creates growth space for NAND. Whether it's SCM SSDs, HBF, or SLC/pSLC SSDs, various high-performance SSD solutions are comprehensively penetrating all layers of AI inference, training, and agent workloads, becoming key components with strong growth momentum.

Due to robust demand and tight supply, NAND Flash prices are well-supported. TrendForce has revised its 2026 global NAND Flash market value forecast upward to $270.6 billion, representing a year-on-year surge of 280.7%. The 2027 value is further projected to expand to nearly $379.4 billion, sustaining a year-on-year growth rate of 40.2%.