The Daily Brief The race to remember
Dec 3, 2025
The podcast dives into the crucial role of memory chips in AI, breaking down the competition in the DRAM and NAND markets. It discusses the fierce cyclicality of the memory business and the consolidation of players. The spotlight is on the high-bandwidth memory (HBM) sector, where industry giants race for dominance while facing geopolitical risks. Additionally, it explores the affordable housing finance sector, revealing cautious growth forecasts due to environmental factors and market dynamics. The episode wraps up with brief updates on key industry news.
AI Snips
Chapters
Transcript
Episode notes
DRAM, NAND And HBM Roles In AI
- DRAM serves as fast short-term memory while NAND stores data persistently; both scale massively in AI systems.
- AI training uses high-bandwidth memory (HBM) to feed GPUs at terabytes per second and NAND for petabytes of datasets.
History Of Brutal Consolidation
- The memory industry history is littered with exits, like Intel leaving DRAM in 1985 after losing to Japanese firms.
- Over decades, consolidation left Samsung, SK Hynix and Micron controlling over 90% of DRAM by 2012.
HBM Could Break The Commodity Trap
- HBM's 3D complexity and packaging make it higher-margin and harder to commoditise than 2D DRAM.
- SK Hynix led HBM adoption, Samsung is catching up, and Micron emerged as a surprising competitor in HBM supply.