Google's new technology shocks the market, AI memory demand drops sixfold! SK Hynix and Micron cut prices simultaneously.

Google launches TurboQuant algorithm, compressing large language model memory usage by at least 6 times, while boosting inference speed by up to 8 times without sacrificing model accuracy. The market quickly interprets this as “demand-side disruption,” with a straightforward logic: if AI models during inference require significantly less memory, it suggests that future data center demands for DRAM, HBM, and NAND storage may experience structural downward revisions.

After the announcement, memory and storage-related stocks declined simultaneously, including SanDisk (SNDK) down 3.5%, Micron Technology (MU) down 3.4%, Western Digital (WDC) down 1.63%; in Asia, Samsung Electronics fell 4.71%, SK Hynix dropped 6.23%. Some analysts believe that TurboQuant is more likely to change “resource utilization efficiency” rather than simply weaken demand.

Google’s Latest Algorithm: Memory Usage Reduced by Six Times, Inference Speed Increased by Eight Times

According to Google’s research team, TurboQuant is a quantization algorithm designed for large language models and vector search systems. Its core is to significantly compress the most resource-intensive parts of AI models—the “key-value cache” and high-dimensional vector data structures. Tests show that this technology can reduce memory consumption by at least 6 times, while increasing inference speed by up to 8 times without sacrificing model accuracy.

(What changes will NVIDIA Vera Rubin bring? Analyzing the memory wars: SK Hynix, Samsung, Micron, SanDisk)

This breakthrough directly targets the current bottleneck in AI infrastructure. Generative AI’s expansion at the compute level heavily depends on high-bandwidth memory like HBM to store model weights and large KV caches, preventing memory bottlenecks during inference. However, TurboQuant combines methods like PolarQuant and Quantized Johnson-Lindenstrauss (QJL) to achieve compression with almost “zero additional memory overhead,” enabling the same or even more efficient computations with fewer hardware resources.

Google’s Algorithm Impacts Memory! US and Korean Memory Stocks Drop

The market quickly interprets this technology as “demand-side disruption.” Following the announcement, memory and storage stocks declined together, including SanDisk (SNDK) down 3.5%, Micron Technology (MU) down 3.4%, Western Digital (WDC) down 1.63%; in Asia, Samsung Electronics fell 4.71%, SK Hynix dropped 6.23%.

The underlying logic is straightforward: if AI models during inference require significantly less memory, the growth curve for future data center demands for DRAM, HBM, and NAND storage may experience structural downward revisions. Especially as the AI industry shifts from “training-oriented” to “inference-oriented,” the marginal impact of efficiency optimization technologies will be amplified.

However, some believe that TurboQuant is more likely to change “resource utilization efficiency” rather than simply reduce demand. With costs decreasing and latency improving, AI application scenarios could further expand, driving overall computing power demand to continue growing—creating a structure of “unit demand decreasing, total demand increasing.” Large memory manufacturers have already sold out their capacity this year, prompting the question: how big is the ceiling for AI growth?

This article: Google’s new technology shocks the market—AI memory demand reduced by six times! SK Hynix and Micron simultaneously decline. Originally published on Chain News ABMedia.