Pullback in U.S. memory stocks provides an entry point, while AI data center demand continues to support tight supply-demand conditions

Overview

This Morgan Stanley North America semiconductor report focuses on the risk-reward in U.S. memory stocks after the recent selloff. The report argues that the unusual feature of this memory cycle is that it is primarily driven by data center and AI demand, while traditional end markets such as consumer, PC, and smartphones may send mixed signals that cause the market to misread the cycle turning point. The analysts believe that after the pullback, the investment appeal of memory stocks is rapidly approaching that of compute leaders, although the best risk-reward in coverage remains NVDA.O and AVGO.O.

Core views

The core judgment is that memory shortages are still intensifying, and data center memory prices are expected to rise at least 25% QoQ in 3Q, above Morgan Stanley and third-party expectations. Although this represents a second-derivative slowdown relative to the 70% increase in 1Q and more than 40% in 2Q, the market had already fully understood that such high growth rates could not continue indefinitely. The report emphasizes that long-term agreements, down-speccing, and improved memory efficiency do not necessarily mean the cycle is ending; rather, they suggest that customers are redesigning systems around a multi-year memory shortage. If the earnings upside can persist for several years rather than peaking in a single year, valuation support may be stronger.

Analysis framework

The report uses cycle supply-demand analysis, procurement channel checks, price momentum observation, risk-reward scenario analysis, and a through-cycle valuation framework. It provides bull, base, and bear case scenarios for SNDK.O and MU.O respectively, and forms its judgments based on EPS, valuation multiples, industry supply and demand, AI demand, HBM wafer intensity, eSSD penetration, and regional revenue exposure.

Methodology notes

• Valuation methodsThrough-cycle EPS multiple method

  Estimate target price by multiplying through-cycle earnings by a target multiple

  The SNDK.O base-case target price of $1,750 is derived from 28x through-cycle EPS of $62.50; the MU.O base-case target price of $1,200 is derived from 30x through-cycle earnings of US$40.

• Scenario analysisRisk-reward scenarios

  Bull, base, and bear case target price ranges

  The report lists for SNDK.O a bull case of $2,635, a base case of $1,750, and a bear case of $1,100; for MU.O it lists a bull case of $1,650, a base case of $1,200, and a bear case of $675, to characterize AI demand, the earnings cycle, and the risk of price declines.

• Industry cycleSupply-demand and inventory cycle analysis

  Assess cycle sustainability through memory supply-demand, prices, inventories, and capex

  The report believes that AI demand growth, HBM4 complexity, low-power DDR5, and enterprise storage demand are together absorbing supply; NAND capex remains moderate, and while new DRAM supply will increase, it is unlikely to quickly relieve the AI-driven shortage.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• SanDisk Corporation. (SNDK.O)

  Beneficiary of NAND and data center SSD

  Strengths

  AI demand is driving improvement in the NAND market, eSSD demand is increasing, capex remains moderate, and through-cycle earnings and free cash flow conversion support valuation.

  Weaknesses

  Direct AI exposure is lower than that of some compute semiconductor companies, North America accounts for a high share of regional revenue, and the company is still affected by NAND cycle volatility.

  Comparison

  The report states that SNDK.O's 28x multiple is slightly below MU.O's target through-cycle multiple of 28.5x, with the lower AI exposure partly offset by higher FCF conversion.

  Risks

  NAND industry growth comes in below expectations, higher capex leads to additional supply, failure to expand data center share, and Chinese manufacturers continue to gain share.

• Micron Technology Inc. (MU.O)

  Core beneficiary of the DRAM/HBM and AI memory cycle

  Strengths

  DRAM fundamentals are in an unusually strong zone, data center and AI demand are rising, HBM share and wafer intensity support tight supply-demand conditions, and the cycle may last 2-3 years.

  Weaknesses

  The stock price and positioning may be relatively crowded, earnings are sensitive to price turning points, and valuation could compress quickly if demand is misread as inventory buildup.

  Comparison

  The report believes memory risk-reward is catching up, but the best value in coverage still comes from compute companies such as NVDA.O and AVGO.O.

  Risks

  Weakening end demand combined with high inventories could lead to rapid price cuts, HBM demand may disappoint, and intensifying competition could pressure prices.

• NVIDIA Corp. (NVDA.O)

  Core beneficiary of AI compute and data center demand, and also an important driver of memory demand

  Strengths

  Training and inference growth are driving data center revenue, GPU AI PCs may support the gaming business, and there may be potential for a recovery in China revenue.

  Weaknesses

  The report also notes that NVIDIA may reduce LPDDR5 content per rack and push for memory efficiency optimization, which could marginally suppress some memory pricing.

  Comparison

  Morgan Stanley believes NVDA.O and AVGO.O offer the best risk-reward in coverage, superior to but not excluding opportunities in memory stocks.

  Risks

  AI end-market demand falls short of expectations, customers cut GPU purchases, AMD becomes a viable GPU competitor again, or cloud customers develop competing in-house hardware.

• Broadcom Inc. (AVGO.O)

  Beneficiary of AI compute- and networking-related semiconductors

  Strengths

  Stronger AI revenue, recovery in the core semiconductor business, and realization of VMware synergies could support valuation.

  Weaknesses

  Valuation depends on AI revenue and M&A integration execution, and parts of the business are still affected by the core semiconductor cycle.

  Comparison

  The report says AVGO.O and NVDA.O remain the best sources of value, while memory stocks are catching up quickly thanks to the pullback and resilient fundamentals.

  Risks

  Losing networking share to Nvidia Mellanox, insufficient ASIC competitiveness or customer losses, and execution risk related to the VMware acquisition.

Key data

• Data center memory pricesAt least +25% QoQ in 3QThe report states this increase is above Morgan Stanley's estimate and third-party estimates.
• DRAM prices+70% QoQ in 1Q, more than +40% QoQ in 2QThe report uses this data to explain that the slower 3Q price increase is a known second-derivative slowdown.
• Memory revenue scaleMore than $200bn this quarter, versus $46bn in the same period last yearThe report believes such a high revenue base makes it impossible to linearly extrapolate prior momentum.
• AI spending growthMore than 50%, possibly well above 50%Based on processor company commentary, the report believes AI spending has become the main incremental driver of memory demand.
• SNDK.O target price$1,750The base case uses 28x through-cycle EPS of $62.50; the current price appears in the text as $1,354.82, implying about +29.17%.
• SNDK.O scenario pricesBull case $2,635, base case $1,750, bear case $1,100The bull case uses 31x EPS of $85, and the bear case uses 25x EPS of $44.
• MU.O target price$1,200The base case uses 30x through-cycle earnings of US$40.
• MU.O scenario pricesBull case $1,650, base case $1,200, bear case $675The bull case uses 33x earnings of US$50; the bear case assumes a downcycle begins in early 2027.
• MU.O active institutional holder ratio46.5%An ownership positioning indicator disclosed in the report.

Impact & implications

If the report's view is correct, the pullback in memory stocks should not be seen as a signal that the cycle is over, but rather as a buying window triggered by a second-derivative slowdown in growth, capex concerns, and down-speccing concerns. At the portfolio level, SNDK.O and MU.O benefit from AI-driven demand for DRAM, NAND, HBM, and eSSD, but investors still need to distinguish between a slowdown in short-term price amplitude and the persistence of a long-term shortage; meanwhile, NVDA.O and AVGO.O are still viewed by the report as better compute-related value opportunities.

Risks

• DRAM and NAND prices could turn quickly; if end demand weakens and inventories rise, falling prices would compress earnings and valuation.
• Long-term agreements may limit the extent of short-term price upside, causing high near-term earnings expectations to be missed.
• Customers may reduce memory demand per unit through down-speccing, improved memory efficiency, or system redesign, potentially weakening pricing elasticity at the margin.
• Higher DRAM capex or a recovery in NAND capex could increase future supply and alter the duration of the cycle.
• If HBM demand slows or competition intensifies, Micron's pricing and margins could come under pressure.
• Continued share gains by Chinese manufacturers could weaken the competitive position of companies such as SanDisk.
• Positioning in memory stocks is crowded, and drawdowns may recur during this unusual cycle.

What to watch

• Whether 3Q data center memory prices reach or exceed a 25% QoQ increase.
• Whether the ceiling on pricing for newly signed long-term agreements is above 2Q levels, and whether such agreements continue to reinforce cycle persistence.
• Whether cloud customers continue to pay a premium versus 2Q prices for 6-week expedited delivery.
• The extent to which HBM4, Rubin Ultra, and low-power DDR5 demand absorb wafer capacity.
• Whether NAND capex remains moderate or shifts toward an investment cycle large enough to expand supply.
• MU.O's HBM share, gross margin, and AI execution progress.
• SNDK.O's share gains in data center SSD and advanced storage technologies such as HBF.
• Whether weak signals from the consumer, PC, and smartphone markets are merely noise or are beginning to transmit into data center demand.