Mar 26, 2026 - Shares of major memory chip manufacturers declined Thursday after Google announced new algorithms designed to significantly reduce memory requirements for running large language models and vector search systems.
Google revealed techniques including TurboQuant and PolarQuant that lower the amount of memory needed for AI inference and retrieval tasks. The advancements target inefficiencies in key-value caches and model quantization, potentially decreasing computational demands for large-scale AI deployments.
Stocks reacted quickly to the news. Micron Technology fell about 4%, Western Digital dropped around 4.4%, Seagate Technology declined roughly 5.6%, and other storage-related shares saw similar losses in early trading. The moves came amid broader market caution over the prolonged U.S.-Iran conflict and rising oil prices, which have added pressure on technology and transportation sectors.
The announcement follows months of intense demand for high-bandwidth memory and DRAM chips driven by AI data center expansion. Tech companies have projected spending of hundreds of billions of dollars on infrastructure in 2026, with memory identified as a key constraint. Manufacturers have shifted production toward specialized high-margin chips for AI accelerators, contributing to shortages and price increases for other memory types used in consumer devices.
Google, which develops its own Tensor Processing Units for AI workloads, has emphasized efficiency improvements in its Gemini models and research. The new algorithms aim to address bottlenecks in memory usage during inference, where models process user queries and maintain context. Industry observers noted that such optimizations could slow the pace of new hardware purchases by hyperscalers if they allow existing infrastructure to handle greater loads.
The developments occur as the AI sector grapples with supply limitations. Only a few companies produce the most advanced memory chips required for cutting-edge AI systems. Google DeepMind executives have previously described memory supply as a choke point for both research and deployment.
As of Thursday, trading in semiconductor stocks remained volatile. No immediate comments on the impact of Google's techniques came from major memory suppliers. The company has not detailed timelines for integrating the new algorithms into its production systems or cloud services.
