On Thursday, memory sector stocks opened broadly lower. SanDisk Corp. and Micron Technology both dropped more than 3%, while Western Digital and Seagate Technology PLC declined over 2%. The movement follows news that Google has released a series of new algorithms designed to significantly reduce the memory footprint required for running large language models (LLMs) and vector search engines. Google's newly introduced core technologies include TurboQuant, quantized Johnson-Lindenstrauss Transform, and PolarQuant. Among them, TurboQuant addresses the memory burden from high-dimensional data processing by optimizing compression efficiency during vector quantization. The Johnson-Lindenstrauss Transform uses mathematical dimensionality reduction to drastically shrink data size while preserving data relationships, achieving efficient representation with "zero additional memory overhead." PolarQuant leverages polar coordinate mapping to project data into a fixed structure, bypassing traditional data normalization steps and further reducing resource consumption. According to Google Research Scientist Amir Zandieh and Vice President Vahab Mirrokni, as AI becomes more widely used in applications such as large language models and semantic search, the importance of underlying vector quantization technology will continue to grow. These algorithms are expected to fundamentally improve the operational efficiency of AI systems and reduce reliance on high-end storage hardware.
Comments