Google’s TurboQuant Boosts AI Performance, Mizuho Prices Micron at $530, Sandisk at $710
Google’s TurboQuant data-compression algorithms delivered AI inference performance gains through key-value cache compression, suggesting support for larger language models and faster response times. Mizuho Securities kept Outperform ratings on Micron and Sandisk with $530 and $710 price targets, arguing the recent memory sell-off overstates peak-memory concerns.
1. TurboQuant Achieves AI Inference Improvements
Google’s TurboQuant algorithms, first explored in 2025, have delivered AI inference performance gains by applying key-value cache compression to large language models, enabling faster token processing and potential cost savings on compute resources.
2. Mizuho Maintains Outperform on Micron and Sandisk
Analyst Vijay Rakesh at Mizuho Securities maintained Outperform ratings on Micron and Sandisk, assigning price targets of $530 and $710 respectively, and argued that investor fears of a peak memory cycle are overblown given historical patterns of efficiency gains driving further demand.
3. Implications for AI Infrastructure Spending
The observed improvements in compression and inference speed could support larger, more complex models and better token economics, likely increasing AI server capital expenditures; analysts draw parallels to past technology shifts that boosted infrastructure investment rather than curtailed it.