Google Throws a YUGE Wrench Into the AI Boom… Memory Stocks Drop as Demand Math Changes

Shares across the memory chip stock sector fell Thursday after Google highlighted research tied to a new algorithm that could significantly reduce the amount of memory required for artificial intelligence workloads.

In Seoul, Samsung and SK Hynix each declined more than 6%. In the U.S., Micron Technology, Western Digital, and SanDisk fell at least 5%, extending losses from the prior session.

The selloff comes after a big run. Memory stocks have been ripping for months as AI demand choked supply, drove prices higher, and boosted margins across the industry. Through Wednesday, shares of SK Hynix and Samsung were up more than 50% this year, while Kioxia Holdings had more than doubled.

Now Google’s research throws a wrench into that story.

The company said its “TurboQuant” algorithm can reduce memory requirements for large language models by as much as sixfold. If scalable, that would directly address one of the key constraints in AI infrastructure: the amount of high-bandwidth memory needed to train and run models.

For investors, the concern is that if AI systems can operate with materially less memory, hyperscalers may not need to procure as much capacity as previously expected. That, in turn, could ease pricing pressure in a segment that has benefited from persistent shortages.

Those concerns are emerging at a time when spending expectations remain elevated. Major cloud providers, led by Amazon.com and Google, are expected to invest roughly $650 billion this year in data center buildouts, much of it tied to Nvidia accelerators and the surrounding memory ecosystem.

Looking further out, the picture gets even murkier.

Some analysts see efficiency gains as a positive. Morgan Stanley’s Shawn Kim notes that lowering memory requirements reduces the cost of running AI models, which can increase usage and improve returns. In that case, demand grows alongside efficiency rather than falling.

That view ties back to Jevons Paradox… the idea that making something more efficient can lead to more of it being used. Analysts at JPMorgan Chase and Citigroup have pointed to the same dynamic.

Even with that in mind, supply hasn’t caught up yet. Executives across the industry continue to point to limited capacity, and Chey Tae-won recently said the memory shortage could last through the end of the decade.

Viewed through that lens, demand in the near term may not change much, especially given how much capital is already being deployed.

At the time of publishing this article, Stocks.News holds positions in Google as mentioned in the article.