TLDR
- Google’s TurboQuant algorithm announcement triggered significant declines in Micron and Sandisk shares, with the technology compressing AI memory requirements by a minimum of 6x
- Market concerns centered on potential reduced demand for memory semiconductors, sending both equities down more than 15% from peak levels
- Vijay Rakesh from Mizuho Securities maintains Outperform ratings, viewing the market reaction as excessive
- The analyst contends that efficiency gains like TurboQuant typically expand AI deployment, increasing overall memory requirements through Jevons’ Paradox
- AI server NAND content has experienced 100% growth year-over-year, with spot market prices advancing each quarter
Shares of Micron and Sandisk experienced significant selling pressure this past week following Google’s publication of TurboQuant research, detailing a compression technique that reduces AI model memory requirements by a factor of six or more. The technology simultaneously delivers inference speed improvements reaching up to eight times faster, while maintaining model accuracy levels.
Investors interpreted the development as bearish for semiconductor memory manufacturers. Reduced memory consumption per AI application could translate to weaker chip demand from producers including Micron and Sandisk.
Both companies have declined at least 15% from record highs established in late previous month. Trading on Thursday showed Sandisk down 5.9% at $652, with Micron falling 5.5% to $347.78.
Market sentiment faced additional headwinds from President Trump’s Wednesday evening remarks, which failed to clarify timelines for resolving the Iran situation, creating uncertainty that carried into Thursday trading.
Google researchers initially examined TurboQuant concepts in 2025, with updated findings regarding AI inference capabilities published in recent papers.
Mizuho Analyst Challenges Market Consensus
Vijay Rakesh of Mizuho Securities countered prevailing market sentiment in client communications. He maintained Outperform ratings on both Micron and Sandisk while keeping price targets at $530 and $710 respectively.
Rakesh advised clients to “buy the TurboQuant memory pullback,” characterizing concerns about peak memory demand as “overblown.”
His central thesis emphasizes that AI efficiency advancements have historically expanded rather than contracted spending patterns. This economic principle, recognized as Jevons’ Paradox, describes how improved efficiency or reduced cost of a resource typically drives increased total consumption.
Rakesh cited three historical precedents. Virtualized machine technology generated expectations of reduced server requirements, yet drove increased demand. The DeepSeek introduction in 2025 prompted GPU demand concerns, while AI infrastructure investment continued expanding. The transition from copper to optical networking, delivering 10x bandwidth improvements, was anticipated to lower costs but instead fueled higher AI server capital expenditure.
NAND Market Fundamentals Show Strength
Rakesh observed that NAND memory content in AI servers has doubled during the past twelve months. Spot market pricing has maintained upward momentum quarter after quarter.
He maintained that TurboQuant-type compression would “enable larger large-language models, faster inference and better tokenomics, spurring more spending” throughout the AI infrastructure ecosystem.
Combining price appreciation with robust fundamental demand, Mizuho anticipates Micron and Sandisk could exceed already-elevated earnings projections in coming periods.
Sandisk currently trades near $652, below Mizuho’s $710 valuation target. Micron sits around $347, compared to the firm’s $530 objective.
