The global memory market is currently undergoing a fundamental transformation that extends far beyond traditional cyclicality. Between 2025 and 2028, the industry is witnessing a "structural crunch" as leading manufacturers aggressively reallocate wafer capacity from commodity DRAM and NAND toward High-Bandwidth Memory (HBM) to satisfy the insatiable appetite of AI infrastructure. This presentation analyzes the critical shift in Capital Expenditure (Capex) and capacity additions, exploring how the transition from consumer-driven demand to AI and general-purpose server dominance has created an unprecedented supply-demand imbalance. We will evaluate the impact of this transition on pricing and availability through 2028. Finally, we look toward the next horizon of production innovation, examining how emerging technologies like Compute Express Link (CXL) and High-Bandwidth Flash (HBF) will redefine the memory hierarchy and offer scalable solutions to the industry’s capacity constraints.

● AI-Generated Data Analysis: A comprehensive forecast of data volumes produced by AI inference services.
● Data Tiering Strategy: Analysis of storage requirements across "Hot," "Warm," and "Cold" data categories.
● Market Share Allocation: Assessing the specific percentage of AI-generated workloads transitioning to NAND-based storage.
● Strategic Supply Outlook: A long-term forecast of NAND supply scenario.

For DRAM suppliers, although HBM profitability may currently trail conventional DRAM due to cyclical margin expansion in commodity products, HBM remains strategically critical because of its unique role in AI systems. Its integration with advanced accelerators, leadership in bandwidth and density, and lack of direct substitutes position it at the core of next-generation compute platforms.
From a market perspective, while conventional DRAM still dominates bit shipments, HBM’s revenue is expanding rapidly alongside AI server growth. Therefore, even if margin leadership rotates in the short term, HBM continues to drive long-term growth, market share differentiation, and structural advantage within the DRAM industry.

Advanced nodes face high technical barriers, with the number of suppliers steadily shrinking. Meanwhile, surging AI demand keeps capacity in chronic shortage, driving price hikes and crowding effects that reinforce a monopolistic landscape dominated by a few players. In contrast, mature nodes are weighed down by massive capacity expansions and weak consumer demand. Geopolitical forces pushing supply-chain localization have further disrupted global capacity distribution, intensifying competition among suppliers. As a result, resources and profits are increasingly concentrated in the advanced segment, while mature nodes struggle to break through in an oversaturated market—highlighting the semiconductor industry’s dual reality of dominance and stagnation. TrendForce projects that the overall foundry market will grow 25% in 2026, with advanced nodes expanding 37%, outpacing the industry average.

Global AI server shipments (including GPU-, FPGA-, and ASIC-based systems) are projected to grow 24% in 2025, with momentum accelerating to around 28% in 2026, driven by continued CapEx expansion from major CSPs building AI infrastructure.
Among AI chip suppliers, NVIDIA’s high-end GPUs remain the primary focus. Blackwell is expected to become the mainstream platform in 2026, followed by the launch of Rubin in 2H26. Meanwhile, CSPs are actively advancing in-house AI ASIC solutions, and the expansion of cloud and edge AI inference applications will further support strong shipment growth in 2026.