AI Boom Fuels DRAM Shortage and Price Surge
AI Boom Fuels DRAM Shortage and Price Surge
https://spectrum.ieee.org/dram-shortage
Publish Date: 2026-02-10 09:00:02
Source Domain: spectrum.ieee.org
Summary:
The explosion in demand for Artificial Intelligence (AI) has led to a dramatic increase in the need for DRAM memory, especially for GPUs and other AI systems, which has severely impacted memory supply and prices. DRAM prices have surged by 80-90% due to this AI-driven boom, leaving other sectors like PCs, consumer gadgets, and others struggling with supply shortages and high costs. Memory experts and economists point to the cyclical nature of DRAM production and the lack of new manufacturing facility investments following a supply slump brought on by the COVID-19 pandemic. The AI data-center infrastructure build-out and the reliance on high-bandwidth memory (HBM) are primarily at fault. Major players like Micron, Samsung, and SK Hynix have new fabs and facilities planned, but these won’t alleviate supply issues for several years. The future may bring technological advancements to improve HBM capabilities, but the balance between demand and supply will remain tough for the industry.
Key Points:
- Demand Surge: The enormous demand for GPU and accelerator memory driven by AI is causing massive price hikes in DRAM, diverting supply from other sectors.
- Cyclical Industry: The DRAM industry’s historical tendency towards booms and busts, along with large investments required to build new facilities, has contributed to the current supply shortfall.
- HBM Technology: High-bandwidth memory (HBM), crucial for GPUs and AI systems, is becoming increasingly important and expensive due to its complex 3D chip-packaging technology.
- Coming Capacity: New production facilities by top memory makers are underway but aren’t expected to ease supply issues before 2028 at the earliest.
- High Prices Persist: Economic trends suggest that although supply issues will be resolved over time, price reductions may occur slowly due to high demand for compute capacity.