# Google's TurboQuant Algorithm Compresses AI Memory by 6x, Rattles Chip Stocks Worldwide

*2026-03-29 15:02 UTC · Technology · 1 min read*

> Google's TurboQuant compresses AI memory 6x with no accuracy loss, triggering global chip stock sell-offs and Pied Piper comparisons.

Google has unveiled TurboQuant, a compression algorithm that shrinks AI models' working memory by up to six times with zero accuracy loss. The technique compresses key-value caches to just 3 bits per value — down from the standard 16 — and delivers up to 8x performance gains on Nvidia H100 GPUs. The research will be formally presented at ICLR 2026.

The announcement sent shockwaves through the memory chip industry. Micron fell 3%, Western Digital dropped 4.7%, and SanDisk slid 5.7%. In Asia, SK Hynix and Samsung shed roughly 6% and 5% respectively. Wells Fargo analyst Andrew Rocha noted TurboQuant "directly attacks the cost curve for memory in AI systems," raising questions about how much memory capacity the industry actually needs.

The internet quickly drew comparisons to Pied Piper, the fictional compression startup from HBO's Silicon Valley. While TurboQuant remains a research-stage project, its potential to slash infrastructure costs for running large language models has made it one of the most discussed AI breakthroughs of 2026.

## Sources

- [techcrunch.com](https://techcrunch.com/2026/03/25/google-turboquant-ai-memory-compression-silicon-valley-pied-piper/)
- [thenextweb.com](https://thenextweb.com/news/google-turboquant-ai-compression-memory-stocks)
- [tomshardware.com](https://www.tomshardware.com/tech-industry/artificial-intelligence/googles-turboquant-compresses-llm-kv-caches-to-3-bits-with-no-accuracy-loss)

---
TechTools News · https://techtools.cz/tools/news/article/1135-google-s-turboquant-algorithm-compresses-ai-memory-by-6x-rattles-chip-stocks
Česká verze: https://techtools.cz/tools/news/clanek/1182-algoritmus-turboquant-od-googlu-komprimuje-pamet-ai-6x-akcie-cipovych-firem
