2 "Magnificent Seven" Stocks Down Between 10% and 23% to Buy Right Now

Microsoft is seeing record AI-related capex pressure: roughly two-thirds of $37.5B quarterly capex is for short-lived assets, driving free cash flow down to $5.9B and contributing to a YTD share decline of 23.4% (31.7% from peak); valuation sits at P/E 23.3 (forward 22.3) vs a 10-year median of 33.2 and a 1.0% dividend yield. Amazon plans roughly $200B of capex this year to build AI data centers (up from $131.8B in 2025 and $83B in 2024), and its shares are down ~10.3% since early 2026 (18.5% over the past five months); authors view both pullbacks as buying opportunities but flag near-term cash-flow strain and competitive risk (e.g., Anthropic/Claude impacting OpenAI exposure, ~45% of Microsoft’s $625B backlog tied to OpenAI).

Analysis

The rush to provision massive model hosting creates an uneven payoff surface: hardware and interconnect suppliers capture immediate pricing power while cloud operators retain optionality to monetize higher-margin managed AI services later. That favors firms owning the stack where they can push differentiated APIs and capture software-as-a-service uplifts, but it also expands the attack surface for commoditization if base models or efficient inference techniques materially cut per-query compute needs.

Key tail risks are efficiency breakthroughs and regulatory constraints. A breakthrough that reduces GPU-hours per token by a material percentage would collapse the marginal economics of recent data-center capacity builds within quarters, while antitrust or exclusivity scrutiny of cloud–model tie-ups could force open access terms and compress future monetization; both outcomes are binary and would show up first in utilization and spot GPU pricing.

Consensus is pricing a binary “win/lose” for incumbent software bundles; that’s likely overstated. Enterprise inertia, identity/tenant lock-in, and cross-product billing (security, identity, device management) create multi-year moat effects that blunt rapid share displacement — meaning a multi-quarter earnings hit can coexist with durable long-term cash returns if monetization resumes.

Operational signals to watch: spot GPU pricing spreads, second-hand datacenter rack lease availability, cloud gross margins and billed AI hours, and developer billing growth for first-party model APIs. These are higher-fidelity leading indicators than headline capex or punditry and will give us a 3–12 month read on whether capacity is being monetized or becoming stranded.

Key Decisions for Investors

• Long NVDA via Jan-2027 call spread (buy 2027 Jan 450C / sell 2027 Jan 700C) — target asymmetric payoff to capture continued GPU tightness while capping premium; size 1–2% NAV. Thesis: 12–24 month upside 60–150% if AI provisioning remains GPU-intensive; max loss = premium paid. Close/trim if weekly spot GPU rents decline >30% from current levels.
• Long AMZN shares (core 2–3% NAV) with downside protection: buy stock and hedge with a 12-month ~10% OTM put or fund by selling near-term 1–3 month 5% OTM puts to scale in. Horizon 12–24 months; target 30–50% total return from monetization of installed capacity and services. Stop/exit if two consecutive quarters show negative sequential billed-AI-hours growth and customer churn > baseline.
• Relative-value pair: Long AMZN / Short MSFT equal notional (9–12 month target) — allocate 1–2% NAV net. Rationale: capture asymmetry between scale-driven capacity monetization and application-layer disintermediation risk concentrated in incumbent software. Risk control: unwind if spread moves >15% adverse within 3 months or if MSFT releases a demonstrably superior hosted-model monetization product.
• Protective hedge for portfolio software exposure: buy 3-month MSFT 10% OTM puts sized to cover 0.5–1% NAV as cheap tail insurance. Use this if net long AI-exposed positions exceed 5% NAV; cost is insurance against a rapid re-rating driven by model-efficiency shocks or adverse regulatory rulings.