The artificial intelligence supercycle is fundamentally reshaping the semiconductor and electronics industries, placing immense strain on the entire supply chain. While companies developing AI accelerators, like Nvidia, are reaping significant benefits from the surge in demand, it’s memory manufacturers poised to capture the largest share of revenue, according to estimates from TrendForce. This isn’t simply a matter of riding the wave; it’s a consequence of differing business models, expansion strategies, and the inherent dynamics of the commodity memory market.
Demand Outstrips Supply
TrendForce projects a striking disparity in revenue between the foundry and memory sectors. Global foundry revenue is expected to reach $218.7 billion, while revenue from 3D NAND and DRAM is forecast to hit $551.6 billion. This means the memory market is more than twice the size of contract chip production. This dramatic shift is directly attributable to the structural changes brought about by the rapid buildout of AI infrastructure. The demand for specific types of memory, particularly high-bandwidth memory (HBM), is creating shortages across the board, driving up prices and creating a favorable environment for memory manufacturers.
The situation is exacerbated by the fact that even low-capacity commodity memory, which isn’t directly required by the AI industry, is becoming significantly more expensive due to the overall tightening of supply. This creates a perfect storm for memory makers to capitalize on the increased demand and limited availability.
The spot price of a 16 Gb DDR5 chip, tracked by DRAMeXchange, illustrates this price surge. Averaging $38 recently, with daily highs of $53 and lows of $25, this represents a substantial increase from the $4.75 average price just one year ago. Similar price increases have been observed in the 3D NAND market.
Fundamental Differences
Analysts at TrendForce, and others, are characterizing the current situation as a “supercycle,” indicating a widespread and potentially prolonged period of high demand and growth. The memory market has experienced similar surges in the past, notably in 2017-2018 with the expansion of hyperscale data centers, and again in 2020-2021 during the pandemic-driven surge in PC purchases. However, those previous cycles were followed by periods of oversupply and price declines as manufacturers increased capacity to meet demand.
The foundry industry, which is significantly more capital-intensive than the memory sector, has historically been more stable. Building new fabrication facilities (fabs) is a lengthy and expensive process. While foundry revenue experienced a decline in 2023, it has since rebounded, with projected growth of 19% in 2024, followed by 25% growth in both 2025 and 2026.
The current AI boom presents a unique scenario. Leading AI developers require cutting-edge hardware, including high-performance processors and large amounts of HBM3E memory, for training their models. Simultaneously, there’s a growing need for powerful inference systems to deploy those models. This sustained demand for CPUs, AI accelerators, memory, and storage, coupled with the willingness of cloud service providers to absorb price increases, is creating a particularly strong environment for memory vendors.
Foundry vs. Commodity
A key distinction lies in the nature of the memory and foundry markets. 3D NAND and DRAM are largely commodities, meaning their prices are highly sensitive to supply and demand fluctuations. A significant portion of memory is sold on the spot market, allowing prices to adjust rapidly to changing conditions. This contrasts with the foundry industry, where long-term agreements and gradual price adjustments are more common.
TrendForce’s projections reflect this dynamic, forecasting an 80% increase in memory revenue in 2024, 46% growth in 2025, and a substantial 134% surge in 2026. Foundry revenue, while still growing, is expected to increase at a more moderate pace.
memory vendors are projected to generate more than twice the revenue of logic chip producers in 2026, a testament to the unique opportunities presented by the AI supercycle and the inherent advantages of the commodity memory market.
The increasing complexity of leading-edge memory technologies, such as HBM4, which utilizes four times more silicon than traditional DRAM, further underscores the challenges in meeting demand. The core question now is the extent to which current price increases are driven by genuine supply constraints versus typical commodity market behavior – the tendency for buyers to increase purchases when prices are rising in anticipation of further increases.
