The memory mirage

The AI playbook might be getting a rewrite.

On Tuesday, Google introduced TurboQuant, a compression technique the company says can reduce the memory required to run large language models by up to sixfold, significantly improving AI efficiency.

That’s pretty bad news for memory stocks, which had surged on booming AI demand from companies like Google, OpenAI, and Anthropic: Samsung rallied nearly 200% over the past year, while Micron and SK Hynix each climbed more than 300%. But if models require less memory, investors are suddenly worried that demand for high-bandwidth chips could soften.

That has sent these once high-flying stocks tumbling over the last five days: SanDisk dropped 21.73%, Western Digital fell 11.95%, SK Hynix declined 8.80%, and Samsung slid 10.84%.

Micron in focus

Micron, in many ways the poster child for the memory stock boom, has also been making headlines. The company recently posted blowout earnings, with sales surging 196% year over year, adjusted EPS jumping 682%, and margins expanding from 36.8% a year ago to 75%. But the stock still sold off last week, and shares fell another 6.94% this afternoon.

One of the key reasons for the decline is that investors are worried the traditionally cyclical memory market is nearing a peak, prompting profit-taking as pricing power and margins reach elevated levels. Another is the supply-demand imbalance: surging demand has outpaced supply, pushing Micron to ramp capex to $25 billion in FY2026, well above expectations, and raising fears of future oversupply.

A new playbook

From today’s market reaction, it’s clear that investors are feeling jittery about the memory industry. But analysts have a different view—in fact, while Micron has sunk 19.50% over the last week, it saw near-unanimous price target upgrades following its latest earnings report.

Wall Street says this time is different. In addition to a structural shift in demand driven by AI that is more durable, they say the industry is no longer purely cyclical, where supply inevitably floods the market. Cantor Fitzgerald’s CJ Muse went even further, saying undersupply could worsen into 2027, with earnings power continuing to rise.

As for concerns that Google’s breakthrough could reduce memory demand, analysts say making AI cheaper and more efficient could drive broader adoption, ultimately increasing total demand for memory chips. Morgan Stanley analyst Joseph Moore pointed out that Google’s improvements won’t significantly reduce the need for memory chips, since the underlying hardware still requires a fixed amount of memory.

So, while it’s easy to trade on headlines, it might be worth digging deeper—and listening to what the pros are saying.—SY