🤖 AI Summary
A significant shift in the memory semiconductor industry has occurred as manufacturers prioritize the production of high-bandwidth memory (HBM) to meet the soaring demands of AI accelerators, leading to a notable shortage of traditional commodity DRAM types like DDR5. This shift has resulted in skyrocketing prices for DRAM, with mainstream contract prices soaring nearly 18-23% in Q4 2025 alone, and spot prices increasing almost threefold year-on-year. Consequently, suppliers are experiencing a dramatic drop in inventories, with stock levels falling from 31 weeks in early 2023 to just about 8 weeks by late 2025, reflecting panic-induced double ordering among device makers and cloud providers.
This dynamic poses both opportunities and challenges for the AI/ML community. While high-bandwidth memory continues to be in high demand and carries greater production complexity, traditional DRAM is seeing unexpectedly high profit margins as suppliers capitalize on the scarcity. Notably, Samsung is benefiting from its emphasis on conventional DRAM amid this shortage, with analysts predicting that within a few quarters, the profitability of legacy DDR5 could surpass that of HBM. However, the evolving landscape also risks straining various sectors of the electronics supply chain, including server manufacturers and PC makers, who are pressured by rising memory costs and potential delays in production cycles, further complicating the trajectory of AI advancements reliant on timely hardware availability.
Loading comments...
login to comment
loading comments...
no comments yet