At JPMorgan’s Technology Summit, Micron Technology stated that strong AI-driven demand is expected to keep the supply of HBM, DRAM, and NAND tight well beyond 2026.
Supply expansion faces structural bottlenecks, including diminishing returns on performance improvements, larger die sizes, and constraints in EUV ramp-up. Even with aggressive capacity additions by manufacturers, the supply gap driven by AI demand cannot be quickly closed in the near term.
The AI compute arms race is pushing the global memory chip market into a multi-year supply shortage.
$Micron Technology (MU.US)$Management stated at JPMorgan’s annual Technology Summit that the supply tightness for HBM, DRAM, and NAND memory chips is expected to persist well beyond 2026, primarily driven by robust demand from AI applications for high-performance memory, while supply-side constraints—stemming from technological bottlenecks—limit rapid capacity expansion.
In a subsequent investment report, JPMorgan cited these remarks and indicated that after hearing Micron’s management presentation, the bank has grown more confident in the multi-year bull case for the AI memory market. The supply-demand gap is unlikely to reverse in the near term, sending a clear signal to investors in the memory sector.
Supply expansion faces structural constraints
According to a JPMorgan report,$Micron Technology (MU.US)$the tightness in the memory market stems from structural factors rather than merely cyclical capacity issues.
Specifically, supply growth is constrained by two main factors: first, performance gains from successive generations of memory chips are gradually diminishing, significantly limiting the room to expand effective supply through technological iteration alone.
Second, the die size of next-generation HBM chips continues to increase, reducing the number of chips producible per wafer and further weakening supply elasticity. Additionally, while the adoption of extreme ultraviolet (EUV) lithography helps improve manufacturing precision for advanced-node DRAM, it simultaneously introduces new constraints related to yield ramp speed and cost.
Given these compounding factors, even aggressive capacity expansions by memory manufacturers will struggle to quickly close the supply gap created by surging AI-driven demand in the short term.
HBM4 mass production accelerates; 1-gamma node becomes the highest-yielding node in history
Regarding product progress,$Micron Technology (MU.US)$several key updates were disclosed.
Micron's management stated that, driven by robust demand from AI applications, the company's 1-gamma process node is expected to become its highest-yielding DRAM node in terms of die output per wafer. HBM memory chips, which are fabricated by vertically stacking multiple layers of DRAM modules into a single package, are widely used in AI GPUs. Micron is continuously integrating EUV lithography into the high-volume manufacturing process for its 1-gamma node.
Regarding the HBM product iteration cadence, Micron disclosed that the production ramp-up speed for HBM4 is twice that of HBM3. Volume production ramp-up for the next-generation HBM4E is expected to commence in 2027, with initial samples utilizing DRAM modules manufactured on the 1-gamma node.
Expanding AI inference demand drives SSD market share gains
In addition to HBM and DRAM, Micron also noted that the evolving nature of AI workloads is creating new growth opportunities for its solid-state drive (SSD) business.
Micron’s management indicated that the continuous expansion of AI context windows and the rapid growth of inference workloads are driving demand for high-capacity, high-performance storage, enabling the company to gain market share in the SSD segment.
Notably, Micron emphasized that its strategy does not involve offering standardized off-the-shelf products; rather, it entails deep collaboration with customers to co-develop storage solutions tailored to specific application scenarios—a model that strengthens customer stickiness and enhances pricing power.
Editor/melody