SUNNYVALE / NEW YORK - March 28, 2026 - Google’s unveiling of a new data compression method, TurboQuant, has triggered volatility in the semiconductor and data storage markets. The technology, which reduces the memory required to run large language models (LLMs) by a factor of six without any loss of accuracy, was perceived by investors as a threat to hardware demand.
In trading on Friday, shares of Western Digital ($WDC) and Seagate Technology ($STX) showed sharp declines, exacerbated, according to sources, by the covering of short positions opened ahead of the announcement. The drop also impacted Asian memory chip manufacturers, although their product lines are focused on High Bandwidth Memory (HBM), which is used in NVIDIA and AMD accelerators.
"TurboQuant is not about eliminating the need for memory, but about improving efficiency," industry analysts comment. The technology allows models that previously required clusters of multiple GPUs to be run on a single chip, radically reducing latency and freeing up resources.
In the long term, this will lead not to a reduction in hardware purchases, but to an expansion of the market. The freed-up memory capacity will immediately be utilized to increase the context windows of models and improve the quality of generation.
For the enterprise data storage segment (HDDs and SSDs), represented by WDC and Seagate, the reduced cost of processing also carries a positive signal: the growth of RAG (Retrieval-Augmented Generation) applications and the accumulation of specialized datasets will require ever-greater storage capacity, not less.