Your PC Memory Costs 400% More. Samsung, SK Hynix and Micron Chose AI Instead

This is a classic margin-transfer story disguised as a supply story: the pricing power is migrating upstream to the few memory vendors with true allocation control, while downstream device makers absorb the cost inflation. The second-order effect is that AI capex is becoming a gatekeeper for the whole semiconductor stack; HBM is effectively crowding out conventional DRAM/NAND, which means consumer and automotive OEMs are now competing for residual capacity at worse economics. That should pressure gross margin assumptions for PC, handset, and some auto-electronics suppliers over the next 1-2 quarters even if unit demand does not collapse.

The immediate setup favors the memory oligopoly, but the durability of the move is the real question. Over 6-18 months, new fabs and China capacity additions can flip this from shortage pricing to oversupply faster than many models assume, especially once HBM demand growth slows from "sold out" to merely strong. The clean falsifier is a rollover in contract pricing or evidence that hyperscaler capex is being deferred because memory/system costs are forcing ROI discipline; that would hit the whole AI hardware chain, not just commodity memory.

The contrarian point is that the market may be underestimating how much AI can tolerate higher memory costs before deployment slows. If inference/training economics get stretched, the response is usually architecture substitution, not endless spend acceleration, which would cap the upside in the mem-capex complex. In that scenario, the current winner basket remains right tactically, but the duration is shorter than the market is likely pricing in.