#0成本拿2股SK海力士 South Korean media "Chosun Ilbo" exclusively reported: SK Hynix is proactively slowing down the expansion of HBM4 high-end AI memory production, suspending part of the HBM3E production line upgrades, and reallocating wafers, equipment, and engineers to produce ordinary DDR5 server DRAM.

The market has identified two layers of fatal signals that directly shatter the "AI perpetual high growth" narrative:

1. The output expectations for NVIDIA's next-generation flagship GPU Rubin, which uses HBM4, have been lowered, and major customers' new product stockpiling is below expectations, cooling the incremental demand for high-end HBM;

2. HBM is no longer the most profitable business: In Q1 2026, the operating profit margin of general-purpose DRAM surpassed HBM by more than 15 percentage points. With a surge in ordinary memory prices, management chooses short-term profit-seeking and abandons aggressive bets on next-generation AI storage. The market interprets this as actively bearish on long-term AI computing demand.