I think the market is still in the middle innings, not late innings, but the easy money phase is likely over.

Why HBM can keep running

1. Structural undersupply

Micron expects both DRAM and NAND supply to remain tight beyond 2026, while its HBM capacity is effectively sold out under long-term agreements. 

2. HBM crowds out conventional DRAM

HBM uses far more wafer capacity and advanced packaging. As Samsung, SK hynix and Micron Technology prioritise HBM, standard DRAM/NAND supply tightens, lifting pricing across the stack. This is why even storage names like SanDisk are rerating. 

3. Inference is the second wave

Training drove HBM first. Inference clusters, edge AI, AI PCs and memory-rich architectures could extend demand for years. Micron’s CEO calling AI “early innings” is probably directionally right. 

What could end the supercycle

Capacity ramps meaningfully in 2027 to 2028

Custom silicon becomes more memory-efficient

Valuations outrun earnings reality

My take:

HBM supercycle likely has another 18 to 36 months of strength, but stocks like Micron may see violent pullbacks along the way. Narrative intact, valuation risk rising.

Best positioned: Micron > SK hynix > SanDisk, in my view.