Do Amazon and Alphabet's Combined $420 Billion in Capital Expenditures Make Sense?

Amazon and Alphabet are projected to spend roughly $220 billion and $195 billion-$205 billion, respectively, on 2026 data-center capacity to address sustained AI and cloud-computing demand. AWS Q2 revenue growth accelerated to 37% year over year, its strongest pace in nearly five years, while Google Cloud revenue rose 82% to $24.8 billion, versus AWS revenue of $42.2 billion. The article argues that demand visibility into 2027-28 supports the investment cycle and views Amazon at 25x and Alphabet at 23x estimated 2027 earnings as reasonable valuations with potential to outperform the broader market.

Analysis

The relevant equity question is not whether cloud demand is strong, but whether incremental capacity earns above the cost of capital after depreciation, power, and networking costs. Alphabet has a potentially cleaner path to capex monetization because custom silicon can both lower its own inference cost and create an external product category; that shifts value away from merchant GPU vendors if workloads increasingly optimize for TPUs. Amazon's larger installed base provides contractual stickiness, but its returns are more exposed to hyperscaler price competition and to retail/logistics cash needs if the investment cycle extends.

Over the next 1-3 months, the market will reward evidence that backlog conversion and cloud operating margins are holding despite accelerated depreciation. For 6-18 months, the key second-order issue is power availability: constrained interconnection and generation capacity can delay revenue recognition while capex and depreciation begin immediately, creating an FCF valley even if demand remains robust. This favors grid equipment and power-exposure names such as VRT and ETN more directly than application-layer AI beneficiaries, while creating downside risk for data-center suppliers if hyperscalers defer builds rather than cancel them.

Consensus likely underweights divergence between AI infrastructure demand and AI infrastructure vendor economics. A sustained migration toward proprietary accelerators is a modest negative for NVDA's long-duration multiple even if total AI capex rises, because the market currently values Nvidia on both volume growth and unusually durable pricing power. The bullish thesis is falsified if cloud growth decelerates while capex guidance remains elevated, cloud operating margins compress materially, or management signals power-related deployment delays; these would imply lower returns on invested capital rather than merely a timing shift.

Key Decisions for Investors

• Favor a 6-12 month long GOOGL / short AMZN pair, sized beta-neutral: Alphabet has greater upside if custom-silicon adoption improves cloud margins and its core cash generation absorbs capex more easily. Reassess if Google Cloud growth decelerates materially for two consecutive quarters or if AWS margin/outlook materially exceeds Google Cloud.
• Maintain NVDA exposure only with a hedge via a modest long AVGO or GOOGL overlay over 6-18 months: proprietary accelerator adoption is the principal risk to Nvidia's terminal multiple, even if aggregate AI spending remains high. The hedge should be removed if merchant-GPU pricing and supply allocation remain tight through the next two hyperscaler capex updates.
• Add VRT or ETN on data-center-power bottleneck weakness rather than chase hyperscaler headlines; the 6-18 month thesis is that electrical infrastructure content rises per deployed compute unit. Risk-manage on evidence of project cancellations rather than mere commissioning delays, which can defer revenue without impairing backlog.
• Before increasing gross exposure to AMZN or GOOGL, verify reported capex, depreciation, cloud backlog, and operating-margin trends in filings and earnings materials. If incremental capex is not accompanied by stable-to-rising cloud margins by the next two reporting periods, treat the investment cycle as an FCF and multiple-compression risk rather than a growth catalyst.