BofA Names Top Pick on AI Memory Demand

Bank of America reiterated a Buy on Micron Technology (MU) with a $1,550 price target, arguing memory is shifting from cyclical commodity to a structural AI infrastructure spend item. It estimates memory now makes up ~35–40% of cloud AI capex (2–3x historical), while MU trades near ~10x forward P/E versus the expectation of multiple expansion. The firm models $1,040/share for cyclical memory and values AI HBM at 31x CY28E P/E, highlighting risks from deeper-than-expected price declines, China competition, and softer end-demand—even as Micron plans to invest $250B+ in the U.S. through 2035.

Analysis

The key market mechanism is not “memory is hot,” but that AI has turned memory into a capacity bottleneck with quasi-pricing power. That matters because if cloud capex really is 35-40% memory-intensive, MU’s revenue sensitivity shifts from handset/PC end-demand to the much stickier hyperscaler spend cycle, which supports a higher terminal multiple than a normal DRAM name. The second-order winner is the semiconductor tool chain: when customers need more advanced bits per wafer and more HBM qualification capacity, AMAT/LRCX/KLAC can monetize the capex before MU fully re-rates.

Near term, the trade is vulnerable to a narrower read-through than the market may infer. If HBM stays tight but NAND/legacy DRAM soften, investors will increasingly separate MU from more commodity-heavy memory exposure such as SNDK and WDC; that dispersion can persist for 1-3 months as sell-side models catch up. The main falsifier is any sign that AI demand is not broadening enough to offset weakness in smartphones/PCs, or that Chinese entrants force ASP declines faster than expected; that would hit the multiple expansion thesis first, before it shows up in reported EPS.

Contrarian view: the consensus may be underestimating how much of the upside is already in the narrative, not the numbers. MU is likely a better relative-value long than an outright beta long because the stock only needs the market to reclassify it from cyclical to strategic, while the index can still wobble on macro or tech risk-off days. If next earnings confirm long-duration agreements and gross margin floors, the rerating can run for 6-18 months; if not, this becomes a fast-money trade with limited durability.

Key Decisions for Investors

• Long MU / short SNDK for 1-3 months: express the AI-memory mix shift while fading commodity NAND exposure; thesis breaks if SNDK prints stronger pricing or MU guides to weaker HBM allocation.
• Buy 3-6 month MU call spreads into the next earnings/capex update: use defined risk to own the valuation-reset catalyst; cut if contract visibility or gross margin guidance does not improve.
• Overweight AMAT or LRCX versus SOXX over the next 3-6 months: if memory capex remains elevated, toolmakers should outperform the broader semi complex on order intensity and backlog conversion.
• Set a watch item on DRAM/NAND contract pricing and hyperscaler capex commentary: if either rolls over for two consecutive prints, reduce MU exposure before the multiple compresses.