On Wednesday, the U.S. stock memory sector displayed a lackluster performance. Micron Technology, for instance, witnessed a decline for five consecutive trading days, while other companies such as SanDisk also saw their stock prices tumble in unison. Market observers attributed this downturn to Google’s unveiling of a novel memory compression algorithm, TurboQuant. This innovative technology tackles the memory consumption challenges inherent in large language models and vector search engines. It enables KV cache compression to a precision of 3 bits, with minimal impact on accuracy. For open-source models, TurboQuant can slash KV cache memory usage by approximately sixfold, accompanied by performance enhancements of up to eight times on NVIDIA H100 accelerators. Moreover, its application potential extends to the realm of vector retrieval, with Google planning to officially showcase it in April. Despite the uncertain prospects for its implementation, the market has already preemptively responded to the shifting landscape of AI memory demand. Analysts noted that if this technology gains widespread adoption, the market will reevaluate the scale of memory capacity necessary for AI, although its promotion and rollout still confront substantial uncertainties.