The AI boom broke the memory market, and the bust could be brutal

The article argues the AI-driven memory crunch has broken the long-standing DRAM/NAND pricing pattern, with prices that “should be falling” instead soaring. It suggests the shortage may persist until 2028 and warns the eventual normalization could be harsh for the supply chain. Overall, the immediate implication is supportive pricing dynamics, but the forward risk is a sharp post-hype downturn.

Analysis

The immediate winner is the small set of memory suppliers with true HBM exposure; in an AI buildout, DRAM stops being a commodity and becomes a bottlenecked input with pricing power. That shifts value capture away from system assemblers toward memory makers and the equipment stack (AMAT, LRCX, ASML), because every incremental wafer start and packaging step monetizes the shortage twice: once in ASPs and again in capex.

The less obvious losers are not just handset and PC OEMs, but also hyperscalers and server ODMs whose AI capex is increasingly memory-intensive. If memory stays tight for another 12-24 months, cloud margins can get squeezed even if GPU demand stays strong, because the bill of materials rises while pricing leverage sits with the supplier base; that’s a second-order headwind the market tends to underwrite only after earnings misses.

Contrarianly, the real danger is that investors extrapolate peak scarcity into a 2028 story and miss the classic semiconductor bust mechanism: once capacity lands, pricing can unwind faster than volume growth. The near-term trend can persist for several quarters, but the reversal trigger will be evidence of supply response—HBM yield improvement, capex upshifts, or inventory days normalizing. If those appear before the market re-rates the cycle, the unwind could be violent.

Key Decisions for Investors

• Long MU on pullbacks; prefer 6-12 month call spreads to express the view that pricing power and HBM mix keep earnings revisions positive over the next 2-3 quarters. Falsify if DRAM ASP commentary or inventory days peak sooner than expected.
• Pair trade: long MU / short DELL into the next two earnings cycles. This isolates memory pricing power versus downstream OEM margin pressure from rising memory BOM costs; exit if Dell demonstrates clean pass-through without share loss.
• Add AMAT/LRCX as a secondary beneficiary basket only if upcoming capex commentary confirms incremental memory investment. Best entry is on any post-earnings weakness, with the thesis invalidated by order cancellations or a capex pause.
• Set an alert on any sign of supply normalization by 2H27: HBM lead times shortening, Samsung/SK Hynix capex acceleration, or DRAM contract prices rolling over. That is the point to reduce longs and consider shorting the memory complex.