Former Intel CEO calls HBM "lousy" at Hot Chips 2026 as High Bandwidth Flash looms
Glittering_Depth_722 · reddit · 2026-09-25
In the Hot Chips 2026 Q&A, Irrational Analysis pressed chipmakers on HBM efficiency: stacking to 20 layers at HBM4 might reach 4TB per square cm, but each layer gets only 20% of a single chip's bandwidth — why go taller instead of faster?
The former Intel CEO's blunt answer: "HBM is lousy." The poster takes this as a signal that High Bandwidth Flash (HBF) is coming, predicting people will soon wonder why they paid so much for something so inefficient.
A rare top-level rebuke of the HBM roadmap with potential implications for AI memory economics.
More from Infra
- $360 for 160GB VRAM: building a local LLM server on unlocked CMP 50HX mining cards — Boricua-vet · 2026-09-25
- MLPerf Training v6.1 adds first LLM post-training benchmark: agentic RL on a 397B model — TheKanter · 2026-09-25
- kvcached brings virtual memory to LLM KV cache, deployed on 10K+ GPUs — techNmak · 2026-09-25
- IEEE plenary talk: micro-optimizations across the full stack, from silicon to models — fooobar · 2026-09-25
- Dev builds local AI GTM workflow, argues the next platform entry point is hardware-bound — dotey · 2026-09-25
- US Faces Memory Chip Conundrum as AI-Critical Prices Skyrocket, WSJ Reports — pstAsiatech · 2026-09-25