Google's unveiling of TurboQuant, a new technology that raises memory-use efficiency for artificial intelligence models by up to six times, triggers profit-taking by investors worried about slowing chip demand, sending shares of global memory chipmakers plunging in unison.

CNBC reported on March 26 that Google's TurboQuant dramatically compresses the key-value cache, a storage space for past calculations required to run large language models. Google explained that the approach cuts the memory capacity needed to operate AI models to about one-sixth of the previous level, while sharply improving inference speed.

The development is seen as a signal that the AI industry, which has relied on supplies of high-performance memory chips, could shift through software innovation, delivering an immediate shock to the market.

After the news, heavy selling spread across the global semiconductor market. In the United States, shares of SanDisk and Micron fell. In South Korea, SK Hynix and Samsung Electronics fell about 6 percent and nearly 5 percent, respectively. Japan's Kioxia also slipped about 6 percent, as selling spread across the global semiconductor market. Investors cite as a key factor the possibility that demand could decline for high-specification chips such as high-bandwidth memory, which big tech firms including Google and OpenAI have used as essential for AI model training and inference.

Some experts compared the research with China's DeepSeek, which jolted the market with high-efficiency AI models, and said there is still significant room to optimise AI inference speed, power consumption and memory usage.

Industry analysts also counter that innovation will not necessarily lead to an overall drop in chip demand. Ray Wang, an analyst at SemiAnalysis, said, "If bottlenecks are resolved, AI hardware performance will instead be strengthened and more powerful training models will emerge." He said as models become more advanced, demand for high-end hardware to support them will be difficult to avoid.

Ultimately, the recent slide is increasingly seen as driven largely by profit-taking in memory stocks that had surged steeply. Samsung Electronics gained 200 percent over the past year, while Micron and SK Hynix rose more than 300 percent, leading to a view that investors were looking for a rationale to lock in gains.

Ben Barringer of British asset manager Quilter Cheviot said Google's announcement is a gradual development rather than something that undermines long-term demand prospects, and that firm pricing and a supply shortage in the current memory market continue to support the earnings of major companies.

As a result, Google's adoption of TurboQuant is viewed as part of the AI industry's evolution. It is an innovation that improves efficiency in operating AI models, but it is assessed as unlikely to change the long-term structure of the memory market overnight. Experts said the technology could ease demand for high-end chips to some extent, but demand across the broader market remains solid as AI models become more advanced and growth continues, underpinning the profits of major semiconductor companies.