Particle.news
Download on the App Store

Google’s TurboQuant Shakes Memory Stocks as Wall Street Counters Fears

Analysts say the compression targets a narrow cache function, not total memory needs.

Overview

  • Google unveiled TurboQuant this week, a method that compresses the key‑value cache used during AI inference by up to six times, which is the short‑term working memory that stores recent tokens to speed model responses.
  • Shares in memory makers sank over the past five sessions, with Micron down about 20% and SanDisk off roughly 11% as investors reassessed AI memory demand.
  • SanDisk disclosed a $1 billion purchase of 139 million Nanya shares for a 3.9% stake, alongside a supply deal to receive the Taiwanese firm’s DRAM products.
  • Morgan Stanley kept overweight ratings on Micron and SanDisk and said TurboQuant reduces KV‑cache use rather than overall memory, adding that memory remains a bottleneck for AI builds.
  • Micron reported record fiscal second‑quarter revenue of about $23 billion and very strong margins, yet its stock fell as worries over long‑term demand, heavy future capex, and broader macro risks weighed on sentiment.