Google AI breakthrough impacts Samsung, Micron memory chip stocks

Google's latest findings claim to improve the efficiency of AI models

Google AI breakthrough impacts Samsung, Micron memory chip stocks

Google's latest findings claim to improve the efficiency of AI models, thereby raising concerns for memory stock investors who fear a potential decrease in chip demand.

On Thursday, the shares of the world’s top two memory chip manufacturers, SK Hynix and Samsung, declined by 6% and nearly 5%, respectively, in South Korea.

In Japan, flash memory firm Kioxia also saw a close to 6% drop. Similarly, Sandisk and Micron experienced declines in the US the previous Wednesday, with their premarket shares still down on Thursday.

Alphabet's Google revealed TurboQuant on Tuesday, a new compression technique that it claims reduces the memory needed for large language models by six-fold.

This method emphasises minimising the key value cache, which saves the previous calculations of an AI model to prevent redundant processing.

This approach is meant to boost AI models’ efficiency, a crucial target for leading laboratories.

Investors worry that this could decrease the demand for AI memory chips, pivotal for training massive long-language models from firms like Google, OpenAI, and Anthropic.

Matthew Prince, CEO of Cloudflare, likened the research to "Google's DeepSeek," nodding to the efficiency improvements made by Chinese AI company DeepSeek last year, which led to a major tech stock sell-off.

"There's plenty of opportunity to improve AI inference in terms of speed, memory use, power efficiency, and multi-tenant operations," he commented in a Wednesday post on X.

Nonetheless, Ray Wang, a memory expert at SemiAnalysis, stated that Google's research might not necessarily translate to reduced chip requirements.

He highlighted that the value cache is "a major hurdle to overcome for enhanced model and hardware functionality."

Wang noted the difficulty in reducing memory use as model performance continues to improve.

"By dealing with a bottleneck, you're enhancing AI hardware capabilities, making future training models more robust. This enhanced power demands superior hardware support," Wang explained to CNBC.

Even with the stocks dropping on Thursday, the memory market stays robust due to a mix of high demand and limited supply, driving memory prices to new heights and bolstering profits for Samsung, SK Hynix, and Micron.

Samsung shares have surged nearly 200% over the past year, while Micron and SK Hynix both enjoyed gains exceeding 300%.