The pullback in U.S. memory stocks creates a more attractive entry point

Overview

This is a Morgan Stanley industry research report on North American semiconductors, especially U.S. memory stocks. The core view is that memory stocks were recently sold off due to market concerns about slowing second-derivative growth, higher capital spending, and spec-downs in AI racks, but in the authors’ view these concerns were already foreseeable a month ago and do not change the main thesis of tight data center memory supply-demand conditions. The report believes AI buildout and agentic CPU buildout are making memory an increasingly important bottleneck, and demand for DRAM, NAND, HBM, and enterprise SSDs still has substantial room for upward revision.

Core views

The core views of the report include: first, memory shortages are still worsening, with procurement checks among data center customers showing no easing and 3Q data center memory prices at least 25% above 2Q; second, traditional cycle signals may fail this time because demand is driven mainly by AI and data centers rather than 3-5% growth in PCs, smartphones, and traditional servers; third, long-term agreements and customer engineering optimization may reduce the amplitude of price peaks but provide stronger cycle durability; fourth, SanDisk benefits from improving NAND and eSSD demand, while Micron benefits from DRAM, HBM, and AI execution, and both offer better entry value after the pullback; fifth, the report still views NVDA and AVGO as the best risk-reward compute stocks under coverage, but memory is catching up quickly.

Analysis framework

The report uses a combination of industry supply-demand checks, price trends, a risk-reward framework, through-cycle earnings multiples, and scenario valuation. The authors connect data center procurement feedback, DRAM and NAND price changes, AI capital spending, HBM wafer intensity, long-term agreement structures, and customer spec-down behavior with company earnings forecasts, and provide bull, base, and bear cases for SanDisk and Micron respectively.

Methodology notes

• Valuation methodsThrough-cycle earnings multiple

  Estimate target price by applying a target P/E multiple to mid-cycle earnings

  SanDisk’s base case uses 28x through-cycle EPS of $62.50 to derive a $1,750 target price; Micron’s base case uses 30x through-cycle earnings of US$40.00 to derive a $1,200 target price.

• Scenario analysisRisk-reward scenarios

  Bull, base, and bear target price ranges

  SanDisk’s bull case is $2,635, base case is $1,750, and bear case is $1,100; Micron’s base case is $1,200 and bear case is $675, and the report also discusses upside conditions under stronger AI/HBM demand.

• Industry analysisSupply-demand bottleneck analysis

  AI demand constraints on memory supply

  The report believes AI spending is growing by more than 50%, HBM4 complexity is absorbing capacity, Rubin Ultra-related content per system will double next year, and LPDDR5 and enterprise storage demand remain strong, making memory one of the key bottlenecks in AI buildout.

Asset mapping & comparison

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

• SanDisk Corporation. (SNDK.O)

  Core beneficiary, rated Overweight with a target price of $1,750.00

  Strengths

  NAND and eSSD demand are improving rapidly, cloud demand is accelerating, incremental capacity investment remains relatively moderate, and FCF conversion has historically been strong.

  Weaknesses

  Direct AI exposure is lower than Micron’s, long-term agreements may cap short-term price peaks, and the share price has already been affected by crowded positioning.

  Comparison

  The report says SanDisk is valued at 28x through-cycle EPS, slightly below Micron’s target through-cycle multiple of 28.5x.

  Risks

  NAND industry growth falls short of expectations, industry capex rebounds, data center market share fails to improve, and China continues gaining share.

• Micron Technology Inc. (MU.O)

  Core beneficiary, rated Overweight with a target price of $1,200.00

  Strengths

  DRAM fundamentals are in a zone of unprecedented strength, HBM and AI execution capabilities are underestimated, and supply-demand may remain tight over the next 2-3 years.

  Weaknesses

  If customer inventory accumulation is mistaken for real demand, a downcycle could begin in early 2027; price declines could be very rapid.

  Comparison

  Compared with SanDisk, Micron has more direct AI and HBM exposure, and its base-case valuation uses 30x through-cycle earnings of US$40.00.

  Risks

  End demand weakness combined with high inventories causes rapid price declines, HBM demand weakens, and increased competition pressures pricing.

• NVIDIA Corp. (NVDA.O)

  The report views it as one of the best risk-reward compute stocks under coverage and also an important driver of memory demand

  Strengths

  Training and inference are driving data center revenue growth, GPU AI PCs may support gaming, and China revenue could recover.

  Weaknesses

  Market share and gross margin are already high, leaving limited short-term leverage for multiple expansion.

  Comparison

  The report believes NVDA/AVGO still offer the best risk-reward, but memory stocks are catching up quickly after the pullback.

  Risks

  AI end-market demand disappoints, customers cut GPU purchases, AMD re-emerges as a credible GPU competitor, and cloud customers’ in-house hardware competition intensifies.

• Broadcom Inc. (AVGO.O)

  The report views it as one of the best risk-reward compute stocks under coverage

  Strengths

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

  Weaknesses

  Needs to execute on VMware integration, and networking share faces competition from Nvidia Mellanox.

  Comparison

  AVGO and NVDA are viewed as superior to memory stocks on risk-reward, but memory stock valuations have become more attractive after the pullback.

  Risks

  Insufficient ASIC chip competitiveness, customer losses, networking share taken by Nvidia, and VMware integration execution risk.

Key data

• Data center memory prices3Q up at least 25% versus 2QThe report says this increase is above both Morgan Stanley’s estimate and third-party estimates.
• DRAM pricesUp 70% QoQ in 1Q, and more than 40% QoQ in 2QThe report cites SIA data to explain why a slowdown in second-derivative growth is inevitable.
• Memory revenue scaleMore than $200bn this quarter, versus $46bn in the same period last yearThe report uses this data to show that prior price momentum was unlikely to continue linearly.
• SanDisk target price$1,750.00Based on 28x through-cycle EPS of $62.50; current price is $1,354.82.
• Micron target price$1,200.00Based on 30x through-cycle earnings of US$40.00; the rating table shows a current price of $848.95.
• Micron FY Oct 2026e forecastRevenue $128,983mm, Non-GAAP EPS $73.30From the report’s key earnings input table.
• SanDisk FY Jun 2027e forecastRevenue $48,826mm, Non-GAAP EPS $214.73From the report’s key earnings input table.

Impact & implications

If the report’s judgment is correct, the recent pullback in memory stocks does not indicate the end of the cycle, but rather a buyable fluctuation within ongoing AI-driven supply-demand tightness. Long-term agreements, spec-downs, and engineering optimization may suppress short-term price elasticity, but they also imply customers acknowledge memory shortages over the coming years and are trying to lock in supply, which in turn strengthens the investment case for a more durable cycle. For portfolios, the report repositions memory stocks from crowded-trade pullback names to AI infrastructure bottleneck assets, while still preferring the risk-reward of compute names such as NVDA and AVGO.

Risks

• Memory prices could reverse quickly; if end demand weakens while inventories remain high, prices could fall rapidly.
• Long-term agreements may suppress the upside amplitude of prices, causing short-term earnings to fall short of bullish expectations.
• Customer spec-downs and engineering optimization may reduce memory demand per system.
• Higher DRAM capital spending could eventually bring more supply.
• NAND industry growth may fall short of expectations, prompting industry participants to raise capital spending again to compete for share.
• HBM demand may weaken, and increased competition may pressure Micron-related pricing.
• Memory stock positioning is crowded, and noise from non-data-center markets may continue to trigger periodic pullbacks.

What to watch

• Actual transaction prices for data center DRAM, NAND, and HBM in 3Q and beyond.
• Whether cloud customers continue paying a premium above 2Q expected prices for 6-week expedited delivery.
• Whether the price caps in newly signed long-term agreements are above previously agreed-in-principle levels.
• Changes in capital spending and rack memory configurations at NVIDIA and other AI processor companies.
• The extent to which HBM4, Rubin Ultra, and LPDDR5 rack demand absorb wafer capacity.
• Micron’s HBM share and gross margin performance in CY26.
• SanDisk’s share gains in data center SSDs and high-bandwidth flash-related technologies.