Key takeaways
- Memory chip stocks have fallen nearly 10% in recent weeks, with Micron and SanDisk both down more than 10% following Google’s TurboQuant announcement.
- Google’s new TurboQuant technology promises to cut AI memory requirements by six times, worrying investors
- Morgan Stanley describes the recent decline as a constructive correction rather than a fundamental concern
- Memory capacity has emerged as the primary barrier to scaling AI infrastructure, surpassing GPU availability
- Morgan Stanley maintains Overweight recommendations on both Micron and SanDisk with price targets of $520 and $690.
Morgan Stanley continues to support memory semiconductor manufacturers after a significant market downturn that shook investor confidence in late March.
The iShares Semiconductor ETF has seen a roughly 10% decline over the past month. Multiple factors contributed to the contraction, including valuation concerns, questions about the sustainability of demand, and emerging innovations in artificial intelligence.
On March 24, Google introduced a new dubbed compression technology TurboQuant. This innovation is said to reduce memory requirements for running AI models by up to six times. This announcement sparked widespread concern among investors.
Both Micron and Sandisk saw declines of more than 10% in the immediate aftermath of the revelations. Micron finished trading at $357 on March 27, though the stock maintained a 25% gain over the year-to-date period.
Morgan Stanley’s Joseph Moore challenged the negative sentiment in a research letter distributed on March 26.
Moore reiterated the overweight recommendations for both Micron And SanDisk. The company’s price targets remain unchanged at $520 and $690, respectively.
According to Moore, the sell-off represents a “healthy pricing-in of durability concerns” rather than signaling a fundamental shift in market dynamics. The financial institution asserts that the business strength of memory manufacturers is “more sustainable than the market believes.”
Memory capacity is emerging as a fundamental constraint for AI infrastructure
Over the past two years, Nvidia GPUs have dominated the conversation as the critical element driving AI infrastructure investments. While this is still accurate, Morgan Stanley argues that memory has evolved to become the primary defining element.
“Memory is the bottleneck, and increasingly, the bottleneck for building artificial intelligence,” the research team stated. They note that clients are now making upfront payments for larger commitments, indicating how tight supply is.
According to Moore, the excess capacity of DRAM has been fully absorbed. “Everywhere we look we see signs that it is a real bottleneck,” he noted.
AI’s share of semiconductor spending could reach “more than 50%,” according to the bank’s analysis. It seems unlikely that the increase in supply will match the intensity of demand.
Morgan Stanley’s assessment of the impact of TurboQuant
Morgan Stanley specifically analyzed the Google TurboQuant announcement, arguing that market participants misinterpreted its implications.
The compression technique targets KV Cache exclusively, not overall memory consumption. “They are only talking about KV Cache, not memory in general,” the company explained.
KV caches typically reside in high-bandwidth memory, which represents a specialized and restricted category. Morgan Stanley described TurboQuant as a “natural path to productivity improvement,” not a demand-destroying hack.
The investment bank does not expect gross margins to approach 81% indefinitely. However, it sets minimal justification for margin pressure in the near term.
Additionally, Morgan Stanley highlighted the strong prospects for free cash flow production from memory sector companies. The company decided that “duration is all that matters,” and by that standard, the market signals “all look positive.”
Micron and SanDisk maintained their Overweight ratings as of March 26, 2026.
