Key Takeaways
- Memory chip stocks experienced approximately 10% declines last month, with Micron (MU) and Sandisk dropping sharply following Google’s TurboQuant announcement
- Google’s new technology purports to cut AI memory requirements by up to 6x, triggering investor concerns
- Morgan Stanley analysts view the recent downturn as a normal correction rather than a fundamental shift
- Memory capacity has emerged as the primary constraint in AI infrastructure expansion, surpassing GPU limitations
- Morgan Stanley maintains Overweight recommendations on both Micron and Sandisk with targets of $520 and $690 respectively
Morgan Stanley remains committed to its positive outlook on memory semiconductor manufacturers despite significant recent volatility that shook investor confidence in late March.
The iShares Semiconductor ETF experienced approximately 10% depreciation throughout the previous month. Multiple factors contributed to the decline, including valuation anxieties, demand uncertainties, and emerging artificial intelligence technologies.
Google introduced TurboQuant on March 24, a novel compression technique that allegedly decreases memory requirements for AI model operation by as much as six-fold. The announcement triggered widespread investor concern.
Both Micron and Sandisk experienced losses exceeding 10% in the immediate aftermath of the disclosure. Micron’s shares settled at $357 on March 27, though the stock maintained a 25% gain year-to-date.
Joseph Moore, an analyst at Morgan Stanley, countered the market pessimism in a research report distributed on March 26.
Moore confirmed Overweight ratings for both Micron and Sandisk, maintaining price targets at $520 and $690 respectively.
According to his assessment, the selloff represents “a healthy pricing in of durability concerns” instead of a fundamental transformation in demand dynamics. The investment bank contends that memory company resilience is “more durable than the market thinks.”
Memory Capacity Emerges as Primary AI Infrastructure Constraint
Throughout the previous two years, Nvidia’s graphics processing units dominated discussions surrounding AI infrastructure investment. While GPU importance persists, Morgan Stanley identifies memory as the emerging limitation.
“Memory is a bottleneck, increasingly the bottleneck, to AI builds,” the research team stated. They highlighted that clients are now securing large-volume agreements through advance payments, demonstrating how constrained supply has become.
According to Moore, DRAM availability has been completely absorbed. “Everywhere we look we see indications that it is a true bottleneck,” he noted.
Artificial intelligence could represent “well north of 50%” of total semiconductor expenditure, the bank projected. Production increases are unlikely to satisfy demand at such elevated levels.
Morgan Stanley’s Analysis of TurboQuant Impact
Morgan Stanley specifically examined Google’s TurboQuant technology, asserting that market participants misinterpreted its significance.
The compression approach exclusively targets KV Cache memory rather than comprehensive memory utilization. “They are just talking about KV Cache memory, not memory overall,” the analysts clarified.
KV Cache occupies high-bandwidth memory, a specialized and constrained category. Morgan Stanley characterized TurboQuant as “normal course productivity improvement,” rather than a demand-threatening breakthrough.
The investment bank acknowledges that gross margins approaching 81% are unlikely to persist indefinitely. However, analysts identify minimal catalysts for near-term margin compression.
Morgan Stanley also emphasized substantial free cash flow generation prospects for memory manufacturers. The firm determined that “duration is all that matters,” and by that standard, signals “all appear positive.”
As of March 26, 2026, Micron and Sandisk retained their Overweight ratings from Morgan Stanley.
