general
Google TurboQuant dents memory-chip stocks, analysts say buy the dip
Google announced TurboQuant, an algorithm that sharply reduces AI cache memory needs and sparked a global selloff in memory stocks, but analysts say the efficiency may lower costs and ultimately boost demand.
Mar 26th 2026 · Republic of Korea
Insights
- Google said TurboQuant cuts key-value cache memory demand by six times using extreme compression.
- The corresponding TurboQuant research paper was published in April.
- Shares of Samsung, SK Hynix and Chinese firms fell after the blog post, with GigaDevice down 5.89 percent and Montage down 3.53 percent in Shanghai.
- Analysts, including Morgan Stanley research head Shawn Kim, said TurboQuant will raise throughput per chip and lower AI inference costs.
- Analysts warned that efficiency gains could increase total memory demand through the Jevons paradox as services become cheaper to use.
Sources
- Google says new TurboQuant compression can lower AI memory usage without sacrificing quality arstechnica.com
- Google unveils TurboQuant, a lossless AI memory compression algorithm — and yes, the internet is calling it ‘Pied Piper’ techcrunch.com
- Google’s TurboQuant AI advance dents memory-chip stocks, but analysts say ‘buy the dip’ www.scmp.com