Structural demand expansion driven by Agentic AI is reshaping the global storage market landscape. Market research firm TrendForce has significantly revised its storage industry forecast upward in its latest report. It now projects the global storage market size to surpass $1.28 trillion by 2027, with supply shortages expected to persist in the near term, potentially sustaining upward price momentum through 2027. TrendForce has raised its global storage market forecast for 2026 from a previous estimate of $551.6 billion to $889.3 billion. The forecast for 2027 has been revised from $842.7 billion to over $1.28 trillion, representing an annual growth rate of approximately 44%. The magnitude of these revisions indicates that AI computing infrastructure development is propelling storage demand to an entirely new level. At the segment level, the DRAM market size forecast for 2026 has been raised to $618.7 billion, reflecting a staggering year-on-year increase of 303%. It is projected to expand further to $903.3 billion in 2027, growing 46% year-on-year. The NAND flash forecast for 2026 has been adjusted upward to $270.6 billion, a roughly 281% year-on-year increase, and is expected to approach $379.4 billion in 2027, maintaining an annual growth rate of about 40%. The core rationale for these upward revisions lies in the shift of AI development focus from large-scale model training to Agentic AI applications centered on inference, leading to sustained structural demand expansion. The supply side is unlikely to close the supply gap in the short term, further strengthening suppliers' bargaining power in contract pricing.

Agentic AI Reshapes Storage Demand Structure

TrendForce notes that the shift in AI development trajectory is the fundamental driver of this storage demand expansion. As AI applications transition from one-time large model training to the continuous inference paradigm of Agentic AI, the characteristics of storage demand are undergoing a fundamental change. At the DRAM level, inference requests within Agentic AI systems have evolved from single queries to continuous iterative loops. Concurrently, KV cache capacity grows proportionally with the expansion of the context window—recalculation, if needed, incurs exponentially higher computing costs. Therefore, efficient KV cache management becomes crucial for AI inference performance, directly driving demand for HBM and DRAM. Furthermore, Agentic AI workloads significantly increase demands on CPUs for scheduling, data preprocessing, and memory management. TrendForce points out that in next-generation AI server platforms, the CPU-to-GPU ratio is gradually shifting from the traditional 1:8 towards 1:4 or even higher, with Nvidia's NVL72 rack adopting a 1:2 configuration. The increased deployment ratio of CPUs will proportionally expand server DRAM capacity requirements, supporting both procurement volumes and contract prices.

DRAM: Supply Constraints Coupled with Demand Expansion, Sustaining Volume and Price Growth Through 2027

Wafer capacity consumed by HBM production continues to constrain the available supply for conventional DRAM. Against a backdrop of sustained demand expansion, this supply constraint further strengthens suppliers' pricing power in contract negotiations, with TrendForce expecting upward price momentum to persist through 2027. TrendForce has raised its DRAM market forecast for 2026 to $618.7 billion, corresponding to an annual growth of approximately 303%. Revenue is projected to increase further to $903.3 billion in 2027, a year-on-year growth of about 46%.

NAND Flash: CSP Capital Expenditure Accelerates, Multi-Layered Demand Drives Growth

In the NAND flash segment, the combined capital expenditure of the world's nine largest cloud service providers continues to climb rapidly. TrendForce projects a growth rate of 79% for 2026, with capital intensity also expected to rise to 34%. The report indicates this trend reflects a shift in industry strategy from demand-driven capacity expansion to large-scale AI infrastructure investment aimed at securing long-term competitive advantage. Agentic AI also serves as a foundational driver for NAND demand. AI agents significantly expand enterprise-scale usage, with heavy users' token consumption increasing by up to fourfold compared to previous levels. The growing complexity of AI-generated content media is also accelerating token consumption growth. Faced with massive and continuously rising storage demand, HBM deployment costs remain relatively high, hindering widespread adoption. Traditional HDDs are limited by access speeds and power consumption, struggling to meet real-time AI data center workload demands. TrendForce believes this situation opens significant growth opportunities for NAND solutions. From SCM SSDs and HBF to SLC/pSLC SSDs, various high-performance SSD technologies are rapidly penetrating multiple ecosystem layers, including AI inference, training, and Agentic workloads, becoming new key growth engines. Against this backdrop, TrendForce has raised its global NAND flash market forecast for 2026 to $270.6 billion, a year-on-year increase of approximately 281%. The market size is projected to expand further to nearly $379.4 billion in 2027, continuing to maintain an annual growth rate of around 40%.