Amazon, Alphabet, and Microsoft: Which 2 to Buy and Which 1 to Avoid

The article favors Alphabet and Amazon over Microsoft, citing lower valuations of 17x and 20x earnings, respectively, versus Microsoft's 27x P/E. Alphabet generated $53 billion of Q2 free cash flow despite a $195 billion-$205 billion AI capex plan, while Amazon's trailing-12-month free cash flow fell to negative $7.6 billion amid projected 2026 capex of $220 billion. Microsoft produced nearly $67 billion in fiscal-2026 free cash flow but is viewed as lagging Google Gemini and Anthropic Claude in AI adoption outside its Windows ecosystem, making it a hold rather than a buy.

Analysis

The relevant valuation question is not headline capex, but incremental gross profit per deployed GPU/TPU and the durability of utilization. GOOGL has the cleanest path to monetize AI across search, Cloud and potentially third-party model distribution, so even modest evidence that AI answers do not dilute search monetization could drive multiple expansion. The underappreciated risk is that AI-generated answers raise query costs faster than ad yield; watch quarterly search TAC ex-traffic-acquisition costs, Cloud operating margin, and capex-to-revenue acceleration rather than aggregate free cash flow alone.

AMZN is the higher-beta operational execution trade: AWS demand and advertising can eventually absorb infrastructure depreciation, but negative free cash flow makes the equity vulnerable if retail working-capital normalization or fulfillment investment coincides with slower cloud bookings. Its custom-silicon strategy is strategically negative for NVDA over a 6-18 month horizon only if trainium/inferentia adoption becomes externally visible in AWS workload mix; near-term hyperscaler spending still supports NVDA revenue. Memory tightness creates a more immediate second-order beneficiary in MU, while raising the risk that cloud margin gains lag revenue growth.

MSFT's premium is vulnerable less because of model-quality rankings than because enterprise AI monetization must offset rising depreciation without cannibalizing high-margin Office and Azure economics. A long GOOGL/short MSFT pair is attractive over 1-3 months if upcoming results show Google Cloud margin resilience and sustained search growth, while Copilot paid-seat disclosure remains limited. Contrarian view: the market may be too quick to treat borrowing as balance-sheet stress; these companies retain unusual financing flexibility, so the more likely downside catalyst is a capex-guidance reset or lower utilization, not a liquidity event.

Key Decisions for Investors

• Initiate a 1-3 month long GOOGL / short MSFT dollar-neutral pair; target 10-15% relative return from GOOGL rerating and/or MSFT multiple compression. Exit if MSFT reports material Copilot paid-seat acceleration with stable Azure margins, or if GOOGL search revenue growth decelerates materially while AI-related cost growth persists.
• Maintain AMZN as a smaller, catalyst-dependent long into the next two earnings reports only if AWS growth reaccelerates and consolidated free-cash-flow conversion improves; use a 8-10% downside stop. Risk/reward is attractive only with evidence that AWS operating leverage is offsetting infrastructure depreciation, not merely higher capex.
• Long MU versus NVDA on a 3-6 month relative basis as AI infrastructure bottlenecks shift toward HBM/DRAM availability and pricing. Falsify if hyperscaler memory procurement guidance eases, HBM supply additions compress pricing faster than expected, or NVDA data-center demand materially exceeds consensus despite custom-silicon adoption.
• Set an event alert around GOOGL disclosures on Apple-related AI distribution economics and search monetization. Do not underwrite material earnings upside until revenue-sharing terms, inference costs, and any impact on traffic-acquisition payments are independently disclosed.