3 Stocks to Buy on the AI Infrastructure Sell-Off

The article frames a buying opportunity in AI infrastructure names despite a recent pullback, citing continued hyperscaler spending on AI data centers. Micron’s fiscal Q3 results are highlighted: revenue jumped from $9.3B to $41.5B and gross margin expanded to 84.6% from 37.7%, with HBM supply sold out through 2027–2028 and HBM expected to reach a $100B market next year. AMD is positioned to benefit from inference and agentic AI, with its CPU-focused agentic AI roadmap targeting a $120B addressable market, while Nvidia remains supported by CUDA/software and expanding full-stack AI infrastructure.

Analysis

The market is still pricing AI as a single trade, but the earnings pool is fragmenting. As deployments move from model training toward inference and agentic workloads, the economic winners shift from pure accelerator vendors to memory, system integration, and CPU share-takers; that is why the relative upside now looks better in MU and AMD than in the highest-multiple platform names. NVDA still owns the standard, but its next leg depends more on attach rates in networking/software and system sales than on another step-function in GPU scarcity.

MU is the cleanest expression of structural tightness: when supply is effectively spoken for years ahead, the usual memory-cycle inventory risk gets pushed out and multiple support improves. The main falsifier is not demand rhetoric, but evidence that lead times shorten, customer pre-buys normalize, or peers add capacity faster than expected over the next 2-4 quarters. AMD’s opportunity is slower but real; inference economics and CPU-heavy agentic stacks create a share-shift path that is more damaging to INTC than additive to NVDA.

Contrarian view: consensus may be overestimating how much of the incremental AI profit pool stays with the incumbent GPU leader once inference dominates volume. If that mix shift is right, the trade is not to chase the whole basket, but to own the constraint points and short the lagging legacy CPU franchise. The near-term risk is a capex pause from hyperscalers; the 6-18 month risk is capacity catch-up in memory, which would matter far more for MU than for NVDA or AMD.

Key Decisions for Investors

• Long MU / short INTC for 1-3 months: best risk/reward on a widening AI stack margin split; target 10-15% relative outperformance, stop if memory lead times or pricing crack.
• Long AMD / short INTC for 3-6 months: express the inference and agentic-AI CPU share shift; upside improves if datacenter CPU wins convert into revenue, downside if Intel stabilizes server share.
• Do not chase NVDA here; wait for a further pullback or use a 6-12 month call-spread structure only if hyperscaler capex commentary stays firm. The upside is intact, but the multiple is less asymmetric than MU/AMD.
• Use MU strength into rallies to trim only if HBM supply commentary changes; otherwise treat any 5-8% drawdown as add-on territory while the market remains structurally tight.