2 Semiconductor Stocks Peter Lynch Would Love

The article highlights two AI-linked semiconductor names with low PEG ratios: Marvell at 0.6x and Micron at 0.2x, versus premium valuations of Intel at 3.2x, Applied Materials at 2.5x, and Arm at 2.2x. Marvell’s data center revenue now exceeds 70% of sales and is growing over 50% year over year, while Micron reported fiscal Q2 2026 revenue of $23.86 billion, up 196% year over year, with DRAM revenue up 207%. The piece argues both stocks may still be undervalued relative to their expected earnings growth, despite cyclical and customer concentration risks.

Analysis

The market is likely mispricing where the value accrues in the AI stack: not in the most obvious compute names, but in the “boring” infrastructure layers that get pulled through every incremental server deployment. Marvell and Micron benefit from a second-order dynamic that is easy to miss: as hyperscalers diversify away from single-vendor dependence, custom silicon, networking, and memory content per rack become structurally higher, which can support earnings even if AI capex growth decelerates from peak rates. That makes these names less dependent on model-training cycles than the market assumes.

The key risk is that the current setup is still a capex super-cycle, not a fully self-funding demand regime. If cloud budgets tighten, the first-order hit would show up in custom silicon design wins and memory pricing within 1-2 quarters, and both names would re-rate faster than their cash flow would suggest. Micron’s apparent cheapness is especially cyclical: a low PEG can persist until the memory downcycle arrives, at which point earnings elasticity cuts both ways. Marvell is higher quality but more exposed to customer concentration and timing slippage on design ramps.

Contrarianly, the consensus seems to be treating AI infra as a one-way trade where every supplier benefits equally. That is too simplistic: the real winners are suppliers with pricing power, long-duration content gains, and visibility into 2026-2027 demand, while the losers are those selling commoditized exposure or dependent on one-off order bursts. On that basis, the article’s core insight is not just that MRVL and MU are cheap, but that their valuation discount reflects backward-looking memory skepticism and underappreciation of how AI shifts the industry toward tighter supply, longer contracts, and less volatile utilization.

The setup looks strongest over a 6-12 month horizon, because the market will likely need another quarter or two of sustained margin commentary before it stops anchoring to the old cyclical playbook. If AI capex remains elevated into the next earnings season, these names can outperform on both estimate revisions and multiple expansion; if guidance cracks, the downside will be abrupt. The asymmetry is favorable, but only if position sizing reflects the possibility that a single disappointing cloud budget guide can unwind the thesis quickly.

Key Decisions for Investors

• Go long MRVL on 1-2% pullbacks after earnings/guidance windows; target 15-20% upside over 3-6 months if data center growth remains >40% y/y, with a stop on any sign of design-win delays or customer concentration pressure.
• Initiate a long MU position with a 6-12 month horizon, but size modestly and treat it as a cyclical/value hybrid; upside can be 25%+ if HBM contracts stay tight, while the main risk is a sharp reversion in memory pricing.
• Pair trade: long MRVL / short ARM or AMAT over the next 2 quarters to express preference for underappreciated AI plumbing over richer-duration valuation; this works best if AI capex broadens beyond leading-edge foundry/tooling spend.
• Use a bullish options structure on MU, such as a 9-12 month call spread, to capture convexity into the next HBM supply update while limiting drawdown if memory pricing softens.
• Avoid chasing INTC/AMAT/ARM at current multiples; they require sustained revision momentum to justify valuations above 2x PEG, leaving less room for error than MRVL/MU.