Storage sell-off is a mistaken panic? Morgan Stanley: Traditional cyclical selling logic doesn't apply, TurboQuant panic is a cognitive bias!

```

U.S. memory chip stocks have recently suffered sharp sell-offs due to concerns over demand triggered by Google's TurboQuant memory optimization technology. However, Morgan Stanley believes the market is mispricing a structural shortage story using the logic of traditional cyclical stocks.

In its latest report, the bank maintained its overweight rating on U.S. memory chip stocks, viewing the recent sell-off as normal profit-taking rather than a signal of a cyclical peak. Memory has evolved from a beneficiary of AI demand to the core bottleneck in AI infrastructure expansion. The tight supply-demand situation is expected to last for several years, and current valuations remain attractive.

Morgan Stanley believes concerns over increased capital expenditure, demand disruption, and Google's memory optimization technology have been overinterpreted by the market.

Regarding Google's TurboQuant memory optimization technology, Morgan Stanley analyst Joseph Moore regards it as an incremental improvement: its purpose is to expand the context window and enhance model capabilities, not to reduce memory demand. It has no substantial impact on the memory market.

The impact of Google's TurboQuant technology is overinterpreted

The report specifically responds to Google's "TurboQuant" memory optimization technology, which triggered the recent decline in memory stocks. Google introduced a data compression algorithm for key-value cache (KV Cache), announcing it will soon launch. The widespread rumor that "Google will reduce memory usage by sixfold" has further pressured memory stocks.

After discussions with industry professionals, Morgan Stanley believes this is an incremental technical evolution with no substantial impact on memory demand. The report explains that KV Cache is usually stored in high-bandwidth memory with fixed capacity; if extra cache offload is needed, it is typically transferred to 18TB LPDDR5 memory in a rack, also with fixed capacity. More efficient KV Cache use may have some effect on third-tier storage (SSD or HDD), but industry feedback generally indicates such improvements are mainly to expand the context window and enhance model capabilities, not to reduce memory costs.

The report also cites Google's own case: Gemini 1.5 Pro tested a 10 million token context window to good effect, but was not released due to high inference costs. As similar optimizations lower costs, they are expected to support more intelligent, higher-compute products, rather than cut storage demand.

AI devours capacity, memory becomes bottleneck for compute expansion

Morgan Stanley's report notes this memory cycle is fundamentally different from historical patterns. The report points out that AI’s consumption of DRAM is so large that supply shortages are now affecting other end markets—both PC and smartphone production are constrained by insufficient memory supply.

For three years, the market assumed there was slack in DRAM supply, but this buffer has disappeared. AI data centers’ demand for high-bandwidth memory (HBM) continues to climb, HBM4’s complexity will further absorb capacity, and once the Rubin Ultra platform launches next year, related memory capacity demand will double. Meanwhile, rack-based low-power DDR5 and enterprise storage demand are also exploding.

Morgan Stanley notes that AI capital expenditure growth exceeds 50%, and as AI’s share of total spending keeps rising, this trend strengthens every year. Greater capital spending may boost supply, but this is not the traditional 3%-5% growth seen by smartphones and servers—its scale is incomparable.

The report cites OpenAI pausing its Sora AI video generation application as proof of insufficient compute supply. Morgan Stanley sees this event as strong evidence for its investment thesis—as token counts grow at double-digit rates weekly, compute supply is severely insufficient and demand far exceeds supply. Video generation is extremely storage- and HDD-intensive; storage shortages may even be one reason Sora paused.

The report concludes: For goods with demand so strong it cannot be fully met, it is difficult to take an overly pessimistic stance at current valuation levels.

Micron target $520, SanDisk target $690

Morgan Stanley maintains its overweight rating on Micron Technology, with a target price of $520, about 25 times through-cycle EPS of $21, offering roughly 36% upside from the current price of $382.09. The bull case target is $700 (28 times through-cycle EPS of $25), the bear case is $240. The bank 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 sets a target price of $690, reflecting 23 times through-cycle EPS of $30, up about 1.8% from the current price of $677.86. Bull case target: $875, bear case: $350. The bank 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 notes both companies’ current annualized free cash flow could reach 15%-25% of their market value. Even accounting for cyclical factors, more than two years of sustained high profitability is enough to support substantive stock price increases.

Risk Disclosure and DisclaimerThe market has risks; investment requires caution. This article does not constitute personal investment advice and does not take into account any individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions are suitable for their particular circumstances. Investments made accordingly are at your own risk. ```