Amazon Is Spending $220 Billion on Data Centers. Here's Why That Makes The Stock an Incredible Buy

Amazon is allocating roughly $220 billion to AI data-center capital expenditures, pushing free cash flow lower and requiring use of cash reserves and debt, but AWS demand is supporting the investment case. AWS revenue grew 37% year over year in Q2, accelerating from 28% in Q1, while AWS operating income rose 64%; its backlog reached $496 billion and grew at a triple-digit rate. CEO Andy Jassy said AWS lacks sufficient compute capacity for demand expected through 2026 and likely 2027, positioning the build-out as a long-term AI market-share opportunity.

Analysis

The investable question is not whether AWS demand is strong, but whether incremental AI infrastructure earns above Amazon's cost of capital before the hardware cycle turns. AI capacity has a short economic life relative to the long-lived facilities carrying the investment; utilization, pricing per GPU-hour, power availability and depreciation assumptions will determine whether reported AWS margin expansion converts to durable free cash flow. The cited backlog should be discounted until disclosures clarify its cancellation terms, duration and how much represents contracted capacity versus framework commitments.

Near term (days to 3 months), AMZN can outperform if quarterly AWS growth and operating-income guidance demonstrate that deployed capacity is monetizing faster than depreciation. The less obvious beneficiary is NVDA, but Amazon's scale also increases its incentive to shift workloads toward Trainium/Inferentia and negotiate aggressively with GPU vendors; this is favorable for AMZN gross margins but a medium-term risk to NVDA's cloud-service-provider mix and pricing power. Power-constrained data-center markets also create a bottleneck premium for utilities and generation suppliers rather than for generic cloud infrastructure vendors.

Over 6-18 months, the key downside is synchronized hyperscaler overbuild: AMZN, MSFT, GOOGL, META and ORCL are all chasing similar enterprise workloads, while customers retain multi-cloud leverage. If enterprise AI applications fail to produce measurable ROI, capacity utilization can fall before capex moderates, producing a double hit of lower AWS pricing and elevated depreciation. Consensus appears too willing to extrapolate revenue acceleration while ignoring that Amazon's retail/logistics cash generation is increasingly financing a capital-intensive utility-like business, potentially limiting buybacks and keeping the equity multiple capped.

Key Decisions for Investors

• Maintain a tactical long AMZN only through the next earnings print if AWS revenue growth, segment operating margin and management's 12-month capex outlook all improve; target a 10-15% relative upside versus QQQ over 1-3 months. Exit on decelerating AWS growth or a material capex increase without corresponding operating-income guidance, as that would validate a free-cash-flow compression case.
• Express the competitive-risk view as long AMZN / short ORCL in equal beta-adjusted dollars over 6-12 months. Amazon has a more diversified cash engine and a lower reliance on sustaining AI-infrastructure growth; invalidate if Oracle converts signed cloud demand into accelerating operating cash flow without a comparable rise in net leverage.
• Do not add to NVDA solely on AWS build-out headlines. Set a watch trigger around AWS disclosures of custom-silicon adoption and GPU supply commitments; evidence of faster Trainium/Inferentia penetration would favor reducing NVDA exposure or hedging with 6-12 month NVDA put spreads.
• Monitor power availability, not just chip shipments: a widening gap between commissioned data-center capacity and utility interconnection dates would delay revenue recognition across hyperscalers. If this emerges, prefer regulated power/generation exposure over data-center equipment suppliers, whose orders may be installed ahead of monetizable load.