3 Stocks to Buy on the AI Infrastructure Sell-Off

The article frames a potential AI-stock buying opportunity as AI infrastructure demand remains strong despite recent pullbacks, citing hyperscalers continuing to plan heavy AI data-center spending. It highlights Micron’s fiscal Q3 results with revenue rising from $9.3B to $41.5B and gross margin expanding to 84.6% from 37.7%, supported by demand for HBM and supply sold out through 2027-2028. It also argues Nvidia remains a core AI infrastructure platform (GPUs, networking, inference via Groq), while AMD is positioned for growth in inference, agentic AI, and CPU demand (citing a $120B addressable market). Overall, the outlook is constructive but explicitly raises the risk that AI spending could eventually slow.

Analysis

The cleaner expression of this setup is not the headline AI incumbents, but the bottlenecks around them. As the spend mix shifts from model training toward inference and agentic workloads, the highest incremental margin accrues to memory and CPU attach rather than raw GPU unit share, which makes MU and AMD more attractive on a risk/reward basis than NVDA at current expectations. NVDA still captures the platform tax, but its multiple is already discounting a long runway of capex intensity; upside is now more dependent on keeping the ecosystem closed than on simply winning every workload.

MU is the strongest structural beneficiary because scarcity plus contractual visibility should keep earnings power elevated even if hyperscaler budgets wobble. The important second-order effect is that HBM capacity is becoming a planning constraint for the whole AI stack, so any delay in MU capacity ramp could ripple into server OEMs, network gear, and even GPU shipment timing. INTC is the obvious loser if AMD keeps taking server CPU share in AI racks, because agentic workloads expand the CPU content per dollar of data-center spend.

Catalyst-wise, the next 1-3 months matter most: hyperscaler capex commentary, HBM pricing, and any evidence that inference deployment is broadening faster than expected. The contrarian risk is that consensus is still too focused on GPU scarcity and underappreciating how much of the margin pool can migrate to memory and system integration. What would falsify the bull case is a guide-down in data-center spend or any sign that HBM lead times and take-or-pay demand are normalizing faster than expected.

Key Decisions for Investors

• Long MU on the pullback for a 3-6 month mean-reversion + earnings-revision trade; best risk/reward in the group because supply scarcity and contractual visibility support estimates. Falsify on HBM lead times shortening or any downward revision to 2025-26 margin guidance.
• Pair trade: long AMD / short INTC over 6-12 months to express the CPU share-shift in AI racks and agentic workloads. Use INTC as the funding leg; thesis breaks if AMD server share stalls or Intel shows credible data-center execution acceleration.
• Reduce chase risk in NVDA; only add on weakness, not strength, because the multiple already prices in prolonged hyperscaler capex. If holding exposure, consider financing via covered calls or a partial hedge versus SOXX/SMH for the next earnings cycle.
• Watch-list, not a trade yet: monitor hyperscaler capex updates and HBM pricing over the next quarter before increasing AI-beta exposure. A capex pause would hit NVDA first on multiple compression and MU second on estimate risk.