1 Artificial Intelligence (AI) Stock I'd Never Sell

AMD projects revenue CAGR of 35% over the next three years, with its data-center segment forecast to grow at a 60% CAGR (vs. Nvidia's ~65% in fiscal 2026). The upcoming MI450 AI accelerator and a deal to power Meta's next-gen AI infrastructure are presented as catalysts that could close the gap with Nvidia. Valuation is cited as becoming more attractive—trailing P/E ~77 but on track to a forward P/E of ~31 vs the S&P 500 ~29—supporting a buy/hold thesis from the author.

Analysis

The immediate competitive lever is not just raw silicon performance but the surrounding stack and procurement elasticity. If a second supplier credibly supplies comparable accelerators at scale, hyperscalers and large cloud accounts gain bargaining power that compresses vendor gross margins and forces accelerated price/performance refresh cycles; that in turn reallocates fab demand and HBM/advanced-substrate volumes across suppliers within 6–24 months. Hardware wins will therefore be decided more on software maturity, total cost of ownership (including power and integration), and availability windows than single-benchmarks — expect meaningful customer churn only after 2–4 large-scale deployments validate a new stack.

Second-order supply-chain winners include high-bandwidth memory and advanced packaging/test vendors who see order re-routing, while legacy x86 server incumbents with weak AI roadmaps risk margin degradation as customers re-balance server BOMs. Key risks that would reverse the current momentum are fast follow-on software lock-in by incumbents, execution slippage on node or packaging supply, or a macro capex pull-forward/pushback event; any of these can swing adoption probabilities within a 3–12 month window. Regulatory or multi-year migration costs at hyperscalers also create a non-linear adoption curve — even technically superior products can take 12–36 months to meaningfully affect revenue mixes.

From a portfolio perspective, this environment favors optionality and convex, defined-risk exposure to adoption outcomes rather than concentrated long-duration equity exposure. Hedged pairs and calendar-staggered option structures capture upside from share-shifts while capping downside if incumbents retain software dominance. Monitor three fast indicators as enrollment signals: (1) multi-rack customer procurement wins, (2) public cloud performance-per-dollar disclosures, and (3) sustained order flow to advanced-memory/package suppliers — each typically leads commercial revenue recognition by two quarters.

Key Decisions for Investors

• Pair trade (12–24m): Long AMD equity (size 3–5% portfolio) / Short NVDA equity dollar-neutral. Thesis: capture share-narrowing + valuation re-rating. Target: 30–50% relative outperformance; stop-loss if pair moves 18% against position on a relative basis. Rationale: asymmetric upside if adoption accelerates; protects vs macro sell-offs.
• Defined-risk option (9–15m): Buy a 12-month AMD 15–25% OTM call spread (financed with a smaller OTM call sale). Position cost ~4–7% of notional, max gain ~300% on cost if adoption accelerates. Use as directional, capped-loss exposure to execution beats without long-dated equity downside.
• Event-driven long (6–12m): Accumulate META (3% portfolio) ahead of infrastructure rollouts and potential margin improvement from diversified hardware procurement. Target absolute return 20–35% if cost-per-inference falls 10–20% for their stack; haircut 40% on an ad-revenue shock. Trim on two consecutive quarters of weaker ad CPMs or if procurement disclosures show single-vendor lock-in remains intact.
• Opportunistic short (9–18m): Tactical short INTC via puts or small outright short (1–3% portfolio) if incremental customer procurement shifts and execution misses materialize. Risk/reward: asymmetric — potential 30–50% downside vs limited time-premium if buying puts; stop if company reports clear server AI roadmap wins or Fab utilization guidance improves.