3 of the Top 5 Performers in the S&P 500 in the First Half Offer 1 Particular Thing for AI Customers. (Hint: It Isn't the GPU, and Nvidia Isn't Among Them.)

AI winners are broadening beyond GPUs: over the past three years Nvidia led logic-chip strength, while in the first half of the year memory/storage names outperformed, with Western Digital up 45% to over $3B revenue and Sandisk data-center revenue up 640% YoY to $1.4B; Micron data-center revenue rose 650% to $11B. The article cites a shift toward inference/agentic AI, with inference expected to drive ~two-thirds of AI compute and increasing data generation and storage needs. Net outlook is constructive for memory and storage providers, while GPU makers remain important for long-term AI compute demand.

Analysis

The market is starting to price a subtler AI spend cycle: not just more compute, but more retained state. That is structurally better for memory and storage names than for pure accelerator exposure, because inference-heavy workloads generate recurring data movement, checkpointing, and retrieval traffic even when GPU unit growth slows. On that frame, MU and SNDK look like the cleaner beneficiaries; WDC benefits too, but its HDD mix means more upside from capacity absorption than from true AI content expansion.

The next 1-3 month catalyst is earnings guidance, not the article itself. What matters is whether hyperscalers confirm that inference capex is broadening into storage and HBM refreshes, which would lift ASPs and utilization before unit growth is visible in reported revenue. The main risk is that this becomes a narrative trade: if customers are still digesting training buildouts, memory names can re-rate ahead of fundamentals and then stall when capex growth normalizes.

Contrarian view: the consensus may be underestimating how much of AI’s value accrues to the picks-and-shovels around data persistence, but it may also be overpaying for any stock with "AI memory" in the story. The quality spread should widen: MU has the best mix and balance-sheet optionality, SNDK has the most operating beta, and WDC is the most vulnerable if the market is extrapolating one strong quarter into a multi-year step-up. A key falsifier is a flattening of data-center revenue or a commentary shift from capacity expansion to efficiency gains, which would cap the upside in the next 1-2 quarters.

Key Decisions for Investors

• Long MU on 1-3 month pullbacks; best risk/reward if the next earnings cycle confirms HBM/data-center mix improvement. Falsify if data-center revenue growth decelerates materially or memory pricing turns down.
• Long SNDK as a higher-beta NAND play for a 3-6 month window, but only on weakness after sharp moves. This is the purest "AI storage" expression, yet it will be the first to give back gains if the trade becomes crowded.
• Prefer MU over WDC in a relative-value pair over the next 1-3 months. MU should capture more of the AI content uplift per dollar of capex; WDC is more exposed to a demand surprise if data retention growth is less linear than expected.
• Do not chase NVDA here; treat it as a hold unless hyperscaler capex commentary re-accelerates. On a 3-6 month horizon, it is more likely to underperform a basket of memory/storage leaders on multiple expansion than to break the uptrend outright.