Nvidia's Artificial Intelligence (AI) Chips Still Need Memory. Here's Why the Micron Sell-Off Has Gone Too Far.

The article argues the market overreacted to Alphabet’s TurboQuant announcement, which briefly triggered a sell-off in memory stocks by suggesting AI workloads would need less memory. The author contends TurboQuant mainly reduces working-memory footprint during inference and does not shrink models or remove the need for high-bandwidth memory (HBM). Nvidia’s GPU architectures (Blackwell and Vera Rubin) were built expecting larger HBM stacks, and Micron’s years of specialized engineering make it a hard-to-replace supplier. Market impact: the TurboQuant narrative likely caused headline-driven, premature selling in Micron and broader memory names, but underlying demand dynamics and integration complexity should continue to support HBM suppliers and Nvidia’s data-center ecosystem.