研报 | AI带动超级循环，存储器产值攀升至晶圆代工2倍以上

Core Insights - The article highlights that the memory and foundry industries are expected to reach record high revenues in 2026, driven by the AI wave, with memory revenue projected to expand significantly to $55.16 billion, more than double that of foundry revenue at $21.87 billion [2][5]. Memory Industry Insights - The current memory cycle, driven by AI demand, shows a more comprehensive shortage compared to the previous cycle from 2017 to 2019, which was primarily driven by cloud data center needs [5]. - The shift in AI focus from model training to large-scale inference applications emphasizes the need for high-capacity and high-bandwidth DRAM, leading to increased demand for server-side memory [5]. - The introduction of high-capacity QLC SSDs is accelerating to balance performance and cost in response to massive data access needs [6]. - The current demand surge is primarily driven by cloud service providers (CSPs), who are purchasing in exponentially larger quantities and are less sensitive to price changes, resulting in record price increases [6]. Foundry Industry Insights - Despite benefiting from strong orders for AI chips, the growth rate of foundry revenue is more stable compared to memory, due to industry structure and pricing mechanisms [7]. - The foundry market is characterized by high technical barriers and capital expenditures, leading to a highly concentrated supply base that limits capacity expansion [7]. - Advanced processes, while high in price, contribute less to overall revenue compared to the more stable mature process market, which accounts for 70%-80% of foundry capacity [7]. Capacity Expansion Insights - The memory industry's capacity expansion is more flexible compared to the foundry industry, contributing to the widening revenue gap between the two sectors [8]. - Memory manufacturers produce standardized products, allowing for more efficient capital expenditure conversion into actual output compared to the diverse product mix in foundry services [8]. - TrendForce indicates that memory manufacturers hold strong pricing power amid ongoing AI demand and supply shortages, with average selling prices (ASP) expected to continue rising [8].