Buy these memory stocks to play the red-hot CPU trade: Analyst

Mizuho's Jordan Klein is bullish on DRAM and memory stocks, citing rising AI-driven CPU demand and tightening supply with no meaningful new supply expected until 2H27. He highlights Micron at 3-4x buyside EPS and says SanDisk could still benefit from SSD demand and strong earnings potential, while also noting favorable CPU attach trends for AMD, Intel, and Nvidia's ARM-based server CPUs. The article is supportive for memory and select semiconductor names, but the impact is mainly analyst-driven rather than company-specific.

Analysis

The market is starting to price a second-order memory cycle, not just a CPU cycle: more AI inference and agentic workloads increase the installed base of servers that must be fed with DRAM, NAND, and especially higher-value memory content per box. That matters because memory pricing is the fastest way to convert unit growth into earnings leverage, and the current setup looks like an earnings revision cycle that can last several quarters if supply remains disciplined. The most attractive exposure is not the “story” names, but the lowest multiple names with the cleanest operating leverage to bit-price recovery.

The key mispricing is that investors are still treating this as a GPU-only AI trade, while CPU attach rates rising in agentic workloads pushes a broader capex bill across the stack. That creates a relative-value winner set: memory vendors with constrained capacity and better balance between commodity and specialized product should outperform CPU designers whose upside is already more fully discounted. Meanwhile, the market may be underestimating how much of the incremental demand gets recycled through enterprise refresh cycles rather than hyperscaler-only spending, which extends the duration of the trade beyond a single quarter of AI enthusiasm.

The main risk is timing: if supply relief arrives faster than expected, or if AI capex pauses, the multiple expansion can outrun the fundamental inflection. SNDK is especially vulnerable to crowded ownership and post-earnings disappointment risk, while high-beta chip names tied to the CPU narrative can de-rate if attach-rate assumptions prove too aggressive. The contrarian angle is that the best long may be the least loved, cheapest memory exposure rather than the most obvious AI beneficiary; if memory ASPs are only beginning to inflect, the consensus is still underappreciating how quickly operating leverage can compound over the next 2-4 quarters.

Key Decisions for Investors

• Go long MU on a 3-6 month horizon; best risk/reward in the group because the market is still pricing it like a late-cycle commodity name rather than a leverage-to-pricing reset.
• Use SNDK only tactically: buy on post-earnings weakness or structure as a call spread into earnings to capture upside while limiting downside from crowded positioning.
• Pair trade: long MU / short AMD for 1-3 months to isolate the memory upcycle while reducing exposure to names already benefiting from AI narrative premium.
• Avoid chasing INTC into strength; if agentic CPU attach-rate headlines continue, prefer a small tactical long via calls rather than outright equity given execution and capacity risk.
• If you want broader AI exposure with lower operational risk, favor TSM over ARM/INTC as a relative beneficiary of the CPU-memory buildout, but size smaller because the article implies the bigger alpha is in memory pricing, not compute design.