Google unveiled TurboQuant, a new compression method that could reduce the amount of memory required to run large language models (LLMs) by six times, according to its announcement on Tuesday [1]. This breakthrough focuses on reducing the key value cache, which stores past calculations of an AI model, thereby making AI models more efficient—a major goal for leading AI labs [1]. The announcement triggered a sharp sell-off in memory chip stocks, as investors feared that fewer memory chips may be required in the future to support advanced AI models [1].
On Thursday, shares of SK Hynix and Samsung, the world's two largest memory chipmakers, fell 6% and nearly 5% respectively in South Korea. Japanese flash memory company Kioxia dropped nearly 6%, while Sandisk and Micron shares declined in the U.S. on Wednesday and continued to trade lower in U.S. premarket on Thursday [1]. The sell-off was attributed to concerns that Google's research could slow chip demand, especially as memory chips have been critical for training large LLMs from companies like Google, OpenAI, and Anthropic [1].
Matthew Prince, CEO of Cloudflare, likened Google's TurboQuant breakthrough to efficiency gains made by Chinese AI firm DeepSeek last year, which also caused a massive sell-off in tech stocks. Prince noted that there is significant room to optimize AI inference for speed, memory usage, power consumption, and multi-tenant utilization [1].
Despite the immediate market reaction, analysts such as Ray Wang from SemiAnalysis argued that Google's research may not necessarily lead to reduced chip demand. Wang explained that addressing the value cache bottleneck could result in more capable AI hardware and more powerful training models, which would require better hardware and potentially higher memory usage in the future [1]. He emphasized that improving model performance could ultimately drive greater demand for memory chips [1].
The memory chip sector has experienced a blistering rally, supported by significant demand and supply shortages, pushing memory prices to unprecedented levels and boosting profits for Samsung, SK Hynix, and Micron. Samsung shares have risen nearly 200% over the last period, underscoring the sector's strong fundamentals despite the recent sell-off [1].
CONCLUSION
Google's TurboQuant breakthrough has triggered a sharp sell-off in memory chip stocks due to concerns over reduced demand, but analysts suggest that long-term demand for memory chips may remain robust as AI models become more powerful. The immediate market reaction reflects investor uncertainty, yet the sector's strong fundamentals and ongoing supply shortages continue to support memory chip prices and profits.