MW Micron's stock bounces, as an analyst offers a reality check on the recent panic
By Britney Nguyen and Emily Bary
Investors shouldn't be too worried about the breakthroughs in memory efficiency that Google just touted, according to BofA
Micron and Sandisk shares were charging higher on Friday.
Micron Technology shares were on pace to snap a six-session losing streak Friday, with an analyst likening the recent market freakout over memory stocks to last winter's DeepSeek saga that ultimately proved a blip.
Back then, investors were worried about a Chinese artificial-intelligence startup that said it was able to train a sophisticated AI model with far less compute than what other developers seemed to require. Chip stocks initially tumbled, reflecting concerns that other industry players would follow DeepSeek's lead and have less need for expensive hardware. But defenders argued that cheaper AI development would lead to greater AI adoption, boosting chip demand.
Semiconductor stocks went on to bounce. BofA analyst Vivek Arya noted on Thursday that capital-expenditure growth from Chinese AI companies remained strong last year, "suggesting the rise of efficiency may not always lead to a decline in total demand."
Now, Wall Street is jittery about a recent breakthrough at Google. The company said this week that its TurboQuant compression algorithm allows it to significantly cut back on AI memory usage while also recognizing performance benefits. Some investors view the development as potentially problematic for Micron $(MU)$, which has benefited as AI drives greater demand for memory products.
But BofA's Arya doesn't necessarily think the recent "memory panic" squares with reality. To start, "compression techniques such as TurboQuant are not new," and Google has published similar material over the past year and a half.
That's notable because Alphabet $(GOOGL)$ $(GOOG)$, Google's parent company, recently called for an expected doubling of capital expenditures this year. "AI capex remains the ultimate proof point of AI spend/demand, not efficiency measures," Arya wrote.
If Google is able to drive a sixfold increase in memory efficiency, as its blog post noted, that's more likely to drive a commensurate increase in accuracy or the amount of data that can be processed, rather than a massive decline in overall memory consumption, according to Arya. That's in line with the phenomenon known as the Jevons paradox, which says that efficiency and affordability benefits ultimately net greater uptake of a product.
Micron's stock was is up 3% in midday trading on Friday, while Sandisk's $(SNDK)$ stock was up 4%.
Still, TurboQuant is only focused on key-value cache memory, a technique used for an AI model to "remember" data that it has processed so that it can respond quickly. Allowing for more storage space there addresses a critical need as the industry shifts to AI inference, or the process of running models.
Therefore, Google's new compression algorithm doesn't necessarily impact the overall memory needs of AI models.
The KV cache is often stored in the high-bandwidth memory of an AI chip, and the size of that component cannot be changed, Morgan Stanley analyst Joseph Moore explained in a note on Thursday. Nvidia (NVDA) and other AI-chip makers have driven immense demand for HBM, a type of dynamic random-access memory that has become crucial as AI models get larger and more complex. SK Hynix (KR:000660), Micron and Samsung Electronics (KR:005930) are the top three producers in the world of HBM, which requires stacking multiple DRAM dies. The silicon-intense needs of HBM are partly a reason for DRAM shortages.
Google's compression algorithm serves a different function, as it's instead designed more for extending context windows, Moore said, referring to the amount of data an AI model can process at one time.
"There's just no indication that demand for memory or storage is going down," Moore said, citing conversations with industry insiders.
-Britney Nguyen -Emily Bary
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
March 27, 2026 13:24 ET (17:24 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments