The AI-driven semiconductor rally has been remarkable during the current bull market, which started on October 12, 2022. It received a big boost when ChatGPT was introduced on November 30, 2023. Memory chips have been a major driver of this rally, reflecting tight supply and rising content requirements across AI systems.
As AI workloads scale, performance is increasingly determined not just by compute but by memory. High-bandwidth memory (HBM) and advanced dynamic random access memory (DRAM) both are essential for AI training and for AI inference.
In the past, memory was one of the most cyclical segments of the semiconductor industry, characterized by sharp boom-bust dynamics driven by excessive capacity expansion during booms, thus setting the stage for busts. The key question now is whether this cycle might be less pronounced due to the high demand for memory driven by AI infrastructure spending.
Demand for memory is expected to remain elevated into 2027–28 before normalizing as supply catches up. Hyperscaler orders placed well in advance have effectively pulled forward demand visibility, while long lead times for fabs and equipment limit the speed of supply responses.
This dynamic is already reflected in market leadership, with memory chip stocks among the strongest performers globally as AI-driven demand continues to tighten supply (chart). Sandisk is a striking example, having successfully raised money with an IPO just 13 months ago, the stock has risen 1,836% from its IPO price, including a remarkable 158% ytd.
This strength is also showing up at the macro level. South Korea’s equity market, heavily exposed to memory and semiconductor exports, has rallied sharply in both local currency and US dollar terms, up 96% in the last year (chart). Notably, Samsung Electronics and SK Hynix alone account for roughly 43% of the iShares MSCI South Korea ETF (EWY), underscoring how the AI buildout is now influencing entire equity markets rather than just individual companies.
Micron is increasingly leading the semiconductor rally (chart). Micron’s latest results last week justified this leadership. Despite already elevated expectations going into the print, the company delivered a clear “Nvidia-style” beat, with revenue of $23.9 billion (+196% y/y) and EPS of $12.20, well ahead of the $8.79 consensus estimate. Margins expanded sharply, with gross margins reaching ~75%, up 98% y/y, in line with the surge in memory pricing that we’re seeing.
Management’s guidance remained exceptionally strong. Fiscal Q3 revenue is expected to reach ~$33.5 billion, with margins expanding to 81%, underscoring sustained demand across DRAM and NAND.
Crucially, management emphasized that both markets are now supply-constrained. Cleanroom capacity, long construction timelines, and the growing allocation of wafers to HBM all are limiting industry supply, with much of the near-term capacity effectively sold out. This reinforces pricing power and supports the durability of the current memory cycle.
Constraints are also emerging deeper within the supply chain. Photonics is one of the most underappreciated bottlenecks in the AI stack. As clusters scale, data transfer speeds are becoming a limiting factor, increasing the importance of optical components and materials.
Companies such as AXT, Lumentum, Coherent, and Fabrinet sit at the center of this ecosystem, supplying critical inputs for lasers and optical engines used in high-speed data transmission (chart). These segments are characterized by concentrated supply chains and slow capacity expansion, creating the potential for significant operational leverage as utilization rises. Notably, AXT’s share price is already up 284% ytd.
