U.S. memory chip stocks have recently experienced significant selling pressure due to demand concerns sparked by Google's TurboQuant memory optimization technology. However, Morgan Stanley believes the market is mistakenly pricing these stocks using the logic applied to traditional cyclical stocks, overlooking a story of structural shortage. In a newly released report, the firm maintained its Overweight rating on U.S. memory chip stocks, arguing that the recent market selloff represents normal profit-taking rather than a signal of a cycle peak. Memory has evolved from being a beneficiary of AI demand to becoming a core bottleneck for AI infrastructure expansion. The tight supply-demand dynamic is expected to persist for several years, and current valuations remain attractive. Morgan Stanley contends that market concerns over factors such as rising capital expenditure, demand destruction, and Google's memory optimization technology have been overinterpreted. Regarding Google's TurboQuant technology, Morgan Stanley analyst Joseph Moore views it as an incremental evolution aimed at extending context windows and enhancing model capabilities, not at reducing memory requirements. He believes it has virtually no substantive impact on the memory market. The impact of Google's TurboQuant technology has been excessively interpreted. The report specifically addressed the recent pressure on memory stocks caused by Google's "TurboQuant" technology. Google introduced a data compression algorithm for Key-Value Cache and recently announced its upcoming release. Widespread reports suggesting "Google reduced memory usage by sixfold" led to renewed pressure on memory stocks. After discussions with industry experts, Morgan Stanley concluded this is a gradual technological advancement with minimal substantive impact on memory demand. The report explains that KV Cache is typically stored in High Bandwidth Memory, which has a fixed capacity. If additional cache offloading is needed, it is usually moved to rack-level LPDDR5 memory, which also has a fixed capacity. More efficient KV Cache usage might have some effect on third-tier storage, but industry feedback indicates the improvements are primarily for expanding context windows and boosting model capabilities, not for cutting storage costs. The report also cited Google's own example, noting that Gemini 1.5 Pro successfully tested a 10-million-token context window but was not released due to prohibitive inference costs. As optimization technologies like this reduce costs, they are expected to enable more intelligent, higher-compute products rather than decrease storage demand. AI is consuming capacity, making memory a bottleneck for compute expansion. Morgan Stanley's report indicates this memory cycle is fundamentally different from historical patterns. It points out that AI's consumption of DRAM is now so significant that it is causing supply shortages in other end markets—production of PCs and smartphones is being constrained by insufficient memory supply. The prevailing view over the past three years was that DRAM supply had slack, but this buffer has now vanished. Demand for High Bandwidth Memory from AI data centers continues to climb. The complexity of HBM4 will further absorb production capacity, and memory capacity requirements are set to double with the launch of the Rubin Ultra platform next year. Concurrently, demand for rack-level Low-Power DDR5 and enterprise storage is experiencing explosive growth. Morgan Stanley also noted that AI capital expenditure is growing at over 50%, and as AI's share of overall spending continues to increase, this trend strengthens annually. While higher capital expenditure will bring more supply, this is not the traditional supply expansion driven by 3% to 5% growth in smartphone and server markets; the magnitude is incomparable. The report cited OpenAI's pausing of its AI video generation application Sora as evidence of insufficient compute supply. Morgan Stanley views this event as clear confirmation of its investment thesis: with token counts growing at double-digit rates weekly, compute supply is severely inadequate, and demand far outstrips supply. Video generation is extremely intensive in its demands for memory and HDDs, and memory shortages may also be a factor in Sora's pause. The report states that for a commodity where demand is so robust it cannot be fully met, it is difficult to maintain an overly pessimistic stance at current valuation levels. Price targets: Micron Technology to $520, SanDisk to $690. Morgan Stanley maintained an Overweight rating on Micron Technology with a $520 price target. This corresponds to approximately 25 times a through-cycle EPS of $21, implying roughly 36% upside from the current price of $382.09. The bull case target is $700, and the bear case target is $240. The firm forecasts Micron's FY2026 GAAP revenue at $110.489 billion, non-GAAP EPS at $59.36, and non-GAAP gross margin at 77.8%. For SanDisk, Morgan Stanley set a price target of $690, corresponding to 23 times a through-cycle EPS of $30, implying about 1.8% upside from the current price of $677.86. The bull case target is $875, and the bear case target is $350. The firm forecasts SanDisk's FY2026 GAAP revenue at $15.499 billion, non-GAAP EPS at $41.09, and non-GAAP gross margin at 60.2%. The report noted that both companies' annualized free cash flow at current profit levels could reach 15% to 25% of their respective market capitalizations. Even accounting for cyclicality, a sustained period of high profitability lasting over two years is sufficient to support substantial stock price appreciation.