A two-day sell-off in memory chip stocks has highlighted divergences within the artificial intelligence investment boom. Analysts suggest a breakthrough technology promoted by Alphabet could potentially curb demand for certain types of memory chips while having minimal impact on others. On Friday, shares of flash memory manufacturers, which provide long-term storage for running AI systems, extended their declines. This group includes companies like Kioxia Holdings, which had seen massive gains over recent months. Meanwhile, shares of leading producers of high-bandwidth memory used in Nvidia's AI accelerators stabilized. Samsung Electronics recouped all its losses, while SK Hynix nearly did the same.
Analysts indicate the market is beginning to recognize that Alphabet's "TurboQuant" technology, which improves AI operational efficiency, poses a greater threat to the former category of manufacturers. Analysts including Tiffany Yeh at Morgan Stanley wrote in a report, "By reducing memory footprint and data transfer, TurboQuant significantly boosts inference efficiency. However, it does not reduce the need for core memory like HBM."
In recent months, investors had flocked to manufacturers of flash memory and storage products, convinced of an explosive demand surge as AI goes mainstream. Since late August, shares of companies like Kioxia had soared over 600%. Their performance had, for a time, outpaced traditional storage industry leaders like Samsung, SK Hynix, and Micron Technology, which were early favorites in the AI trade due to their high-margin HBM chips when market focus was on "training" large language models like ChatGPT.
Now, as investors digest the implications of Alphabet's technological breakthrough, flash memory companies are bearing the brunt of an industry-wide sell-off that began this week. Alphabet stated this algorithm can reduce the memory required to run specific parts of large language models by at least six times, helping to lower the overall cost of AI. Investors worry this could reduce memory demand from large data center operators, such as Meta. This, in turn, might depress prices for components also used in smartphones and consumer electronics.
Analyst Jake Silverman noted in a report, "HBM demand, and the DRAM produced by Micron, may remain unaffected as they are needed to store model weights in GPU memory. In contrast, NAND flash demand could face more profound long-term impacts."
Indeed, after significant rallies in the post-ChatGPT AI boom, technology stocks are under intense scrutiny. Concerns about inflation stemming from conflict in Iran are making investors wary of high valuations, leading to profit-taking based on daily headlines. Ed Gomes, Chief Investment Officer at SGMC Capital Pte, commented, "The development of hardware supporting AI technology delivery and application development is a long-term trend spanning years or even decades, not days or weeks. The sell-off triggered by TurboQuant is 'short-term noise' and instead presents a very good buying opportunity."
Comments