The global semiconductor industry is entering a new growth phase powered by artificial intelligence, with both memory and wafer foundry segments heading toward record revenues by 2026. According to TrendForce, limited supply and rapidly rising prices will push the memory market to $551.6 billion, while the global foundry sector is expected to reach $218.7 billion. The forecast highlights a widening revenue gap as the memory industry scales to more than double the size of wafer foundries.
A new AI-driven memory supercycle emerges
TrendForce notes that the current cycle differs significantly from the 2017 to 2019 memory supercycle, which was largely fueled by cloud data center expansion. The present surge is driven by AI workloads and is far more supply-constrained, creating stronger demand resilience and pricing power.
As the AI industry transitions from model training to large-scale inference, real-time responsiveness and faster data access are becoming critical. This shift is accelerating demand for high-capacity and high-bandwidth DRAM in servers. Per-server memory configurations are also rising, further boosting demand.
The growing adoption of NVIDIA’s Vera Rubin platform is intensifying the need for high-performance storage. Enterprises are increasingly deploying SSDs to support data-heavy AI workloads. To balance performance and cost, operators are expected to rely heavily on high-capacity QLC SSDs to handle massive data access requirements.
Another major shift in this cycle is the buyer profile. Cloud service providers are now leading procurement rather than end-device manufacturers. Their rapidly expanding purchases and lower sensitivity to price increases have enabled record price hikes that exceed those seen in the previous supercycle.
Foundry growth remains strong but more measured
While wafer foundries are benefiting from the AI chip boom, their revenue growth is expected to be more gradual compared to the memory sector. This difference is largely due to industry structure and pricing models.
Advanced-node processes command premium prices and have supported recent growth. However, high technical barriers and capital intensity have resulted in a highly concentrated supplier landscape. This limits how quickly capacity can expand.
Mature nodes still represent roughly 70 to 80 percent of total foundry capacity, while advanced nodes account for only 20 to 30 percent. As a result, even though advanced processes carry higher pricing, they contribute a smaller portion of overall revenue.
The foundry sector also relies heavily on long-term contracts, which reduces price volatility. This contract-based model prevents the sharp pricing swings commonly seen in the memory market.
Memory capacity expansion proves more flexible
TrendForce attributes the widening revenue gap between memory and foundries to differences in capacity expansion models. Memory production is more standardized, allowing manufacturers to scale output more efficiently.
Memory products typically require fewer mask layers than logic chips. This enables memory companies to convert capital expenditure into effective production faster than pure-play foundries, which must support a wide range of process technologies from 28 nm to beyond 90 nm.
Because supply shortages are expected to persist as AI demand accelerates, memory suppliers continue to hold strong pricing power. With average selling prices rising due to sustained supply-demand gaps, the memory industry is projected to grow faster than the wafer foundry sector through 2026.
BABURAJAN KIZHAKEDATH
