Semiconductor stocks sink as Google’s new AI tech threatens memory demand - Micron drops 4.82%, Sandisk tanks 8.35%

Semiconductor stocks slumped on Thursday following Alphabet Inc.’s Google highlighting research into a novel algorithm designed to optimize storage efficiency for artificial intelligence. This technological advancement from Google could potentially mitigate current supply shortages, leading to a downward adjustment in market pricing.

Although the research was originally published last year, the company drew fresh attention to it on social media platform X this week.

According to Google, its TurboQuant algorithm can diminish the memory capacity required for operating large language models by a factor of six, significantly lowering the financial burden of AI training.

Investors are concerned this could dampen the aggressive demand for memory from hyperscale data centers, eventually impacting the pricing of components also found in mobile devices and general electronics.

Stock Market Reaction

Consequently, shares of chipmakers Micron Technology Inc tumbled 4.82%, Western Digital Corp. lost 4.93% , Sandisk Corp. dropped 8.35%, Nvidia slipped 2.60%, and Advanced Micro Devices slid 6.16%.

Memory manufacturers had previously experienced a massive rally as intense investment in AI hardware caused supply deficits, resulting in a surge in both chip valuations and corporate earnings.

Currently, four major hyperscalers, spearheaded by Amazon.com Inc. and Google, are projected to allocate approximately $650 billion this year toward data center construction, involving massive purchases of Nvidia Corp.’s AI accelerators and specialized memory units.

Stocks of hyperscalers Microsoft fell 1.02%, Amazon declined 0.64%, Meta Platforms plunged 6.80%, and Alphabet was 2.19% lower.

In an analysis, Morgan Stanley’s Shawn Kim suggested that the broader industry impact of Google’s research should be viewed as positive because it addresses a vital technical bottleneck. Specifically, it enhances the efficiency of the key value cache used during the inference phase of running AI models.

“If models can run with materially lower memory requirements without losing performance, the cost of serving each query drops meaningfully, resulting in more profitable AI deployment,” Kim wrote.

While JPMorgan analysts noted that some investors might use the news as an opportunity to secure profits, they maintained that there is no immediate threat to overall memory consumption.

Kim further noted that TurboQuant is advantageous for hyperscalers due to the improved return on investment. Over the long term, this could even benefit chip producers, as “a lower cost per token can also lead to higher product adoption demand”.

Nevertheless, South Korean market leaders Samsung Electronics Co. and SK Hynix Inc. both saw their shares fall by at least 6% in Seoul trading.