Amazon Earnings Preview: Q3 2026

Consensus estimates project technology infrastructure capex to more than double to $762 billion in calendar 2026 from $362 billion, then rise to $1.3 trillion in 2027. The two-year total would exceed $2 trillion, reflecting continued expansion of investment to support AI.

Analysis

The investable question is who captures returns on the capex, not whether infrastructure spend rises. Near term, constrained power delivery, grid equipment, cooling, networking and memory capacity can give upstream suppliers better pricing and order visibility than the buyers funding the buildout. That advantage is conditional: capacity additions, customer concentration and contract terms determine whether incremental demand becomes margin or merely volume.

Over 1–3 months, treat the forecast as a catalyst for order-book and lead-time revisions, not as evidence of realized AI revenue. A relative-value bias is long diversified electrical-equipment, cooling and network-infrastructure suppliers against a basket of large cloud platforms: the latter face higher depreciation and execution risk if utilization and monetization lag investment. Keep exposure measured; the article provides no company-level backlog, valuation or financing data to select names or size the spread confidently.

The contrarian risk is that aggregate capex forecasts are mistaken for durable end demand. If returns on deployed capacity disappoint, projects can be delayed together, turning today’s bottlenecks into oversupply and leaving infrastructure buyers with underutilized assets. Over 6–18 months, watch utilization, incremental AI revenue relative to investment, project cancellations, supplier lead times and cloud-platform guidance. A deterioration in those measures would reverse the supplier thesis; evidence of monetization keeping pace with investment would weaken the short leg. No standalone directional trade is justified by the forecast alone.

Key Decisions for Investors

• Use the forecast as a watchlist catalyst, not a reason to buy broad AI exposure outright; verify supplier backlog, delivery lead times and pricing before increasing risk.
• Consider a modest, diversified long in electrical-grid, data-center cooling and networking suppliers versus a basket of large cloud platforms. Treat this as a relative-value hypothesis, not a company-specific recommendation; avoid initiating until valuation, positioning and order data are checked.
• Reassess over the next 1–3 months on supplier orders and cloud-platform guidance. Reduce or exit the relative trade if lead times normalize alongside order cancellations, or if cloud monetization and utilization materially improve.
• Track the 6–18 month downside: project deferrals, weakening utilization, or AI revenue failing to support continued investment would indicate capex overshoot and raise risk of supplier oversupply.