Key Takeaways
- Memory chip makers Micron and Sandisk experienced significant declines following Google’s disclosure of TurboQuant, an algorithm that reduces AI memory requirements by 6x or more
- Market participants worried the technology would diminish memory chip demand, driving both equities down more than 15% from their latest peaks
- Vijay Rakesh from Mizuho Securities maintains the downturn is excessive and reaffirmed Outperform recommendations for both companies
- The analyst contends that efficiency gains like TurboQuant historically drive greater AI adoption, ultimately increasing memory consumption—a phenomenon called Jevons’ Paradox
- NAND capacity in AI servers has grown twofold over the last twelve months, with spot market prices climbing quarterly
Shares of memory manufacturers Micron and Sandisk experienced a sharp retreat last week following Google’s publication of research detailing TurboQuant, a novel compression technology that reduces AI model memory requirements by a factor of six or more. The breakthrough also delivers inference speed enhancements of up to eight times while maintaining model performance.
Investors interpreted the development negatively for semiconductor memory producers. The prospect of reduced memory consumption per AI application suggested potential weakness in chip demand from industry leaders like Micron and Sandisk.
Both companies have shed at least 15% from their record highs established in late March. Trading on Thursday saw Sandisk decline 5.9% to $652, while Micron retreated 5.5% to $347.78.
Market sentiment was further dampened by President Trump’s Wednesday evening remarks, which failed to clarify when the Iran situation might be resolved, creating additional headwinds for equity markets entering Thursday’s trading.
Google researchers initially examined TurboQuant concepts in 2025, with updated findings on AI inference capabilities released in subsequent publications.
Mizuho Analyst Challenges Market’s Negative Response
Vijay Rakesh, who covers semiconductors for Mizuho Securities, countered the bearish market sentiment in his client communication. He maintained Outperform classifications on both Micron and Sandisk while holding firm on price objectives of $530 and $710, respectively.
Rakesh advised clients to “buy the TurboQuant memory pullback,” characterizing concerns about memory demand peaking as “overblown.”
His central thesis revolves around historical patterns showing that technological efficiency breakthroughs in artificial intelligence typically result in expanded investment rather than contraction. This economic principle, recognized in the sector as Jevons’ Paradox, suggests that when a resource becomes more cost-effective or efficient, aggregate consumption rises rather than falls.
The analyst highlighted three historical precedents. Virtual machine technology was anticipated to decrease server requirements but produced the reverse outcome. DeepSeek’s 2025 debut generated concerns about GPU demand deceleration, yet AI infrastructure investment continued its upward trajectory. The transition from copper to optical networking infrastructure, which delivered 10x bandwidth improvements, was projected to reduce expenditures but instead catalyzed increased AI server capital investment.
NAND Market Fundamentals Show Continued Strength
Rakesh emphasized that NAND memory content deployed in AI servers has expanded by 100% during the previous twelve months. Spot market valuations have demonstrated consistent quarterly appreciation.
He maintained that compression technologies like TurboQuant would “enable larger large-language models, faster inference and better tokenomics, spurring more spending” throughout the AI infrastructure ecosystem.
With pricing power intact and fundamental demand remaining robust, Mizuho projects that Micron and Sandisk may surpass current consensus earnings forecasts in upcoming quarters.
Sandisk currently trades near $652, representing potential upside to Mizuho’s $710 valuation. Micron hovers around $347, compared with the firm’s $530 price objective.
