Amazon shares sink as spending on AI overshadows growth in cloud unit

Amazon reported Q3 revenue of $201.9B, slightly below expectations, and profit of $20.7B, also below expectations and down year over year; AWS sales rose 39.3% to $45.98B, beating targets. Advertising growth of 12% missed analyst expectations, projected holiday-quarter operating income came in below estimates, and free cash flow swung from positive $14.8B to negative $42.4B over the past year. Amazon reiterated expected 2026 capital expenditures of about $220B, and its shares initially fell 5% in after-hours trading.

Analysis

The key issue is not whether AI demand exists, but whether committed cloud demand earns an adequate return on a much larger, front-loaded asset base. Long-dated Anthropic commitments can help underwrite AWS capacity, but they also increase customer concentration and do not prove that realized usage, pricing, and margins will justify the buildout. If hyperscalers add capacity faster than customers monetize AI, depreciation and power costs can pressure returns even while cloud revenue grows quickly.

AMZN has an additional exposure absent from a pure cloud thesis: weaker advertising and the prospect of AI shopping agents bypassing Amazon’s owned discovery surfaces. If commerce shifts away from its app and site, ad inventory and conversion data may lose value; blocking an agent may protect near-term monetization but risks ceding customer access. Holiday execution is the near-term test, while capex productivity and agent-driven commerce are 6–18 month questions.

The contrarian case is that investors may be over-weighting negative FCF, which can reflect investment timing rather than failed demand; contracted compute could eventually support utilization. But revenue growth alone is not evidence of attractive returns. Peers face the same capex-cycle risk, so this is not a clean sector-wide short. The relative concern is AMZN’s combination of heavy investment, weaker non-cloud signals, and a cautious profit outlook.

Key Decisions for Investors

• Avoid chasing the initial after-hours move. For the next 1–3 months, consider a modest AMZN-underweight / MSFT-overweight pair, sized to limit broad hyperscaler exposure; the thesis is relative operating momentum, not that MSFT is insulated from AI capex risk.
• Reassess after holiday-quarter results: monitor operating income versus guidance, advertising growth, AWS growth and profitability, and whether management explains a credible path from capex to utilization and returns. A sustained ad rebound and stronger-than-guided operating profit would weaken the underweight thesis.
• Track Anthropic’s actual cloud usage and the terms and timing of its commitments, rather than treating headline contract value as equivalent to recognized revenue or profit. Any evidence of concentrated customer exposure or delayed utilization would raise the risk of low-return capacity.
• Falsification for the relative trade: AMZN delivers better-than-expected holiday operating performance and stabilizes advertising while AWS growth remains strong; alternatively, MSFT’s cloud momentum or margins deteriorate enough to remove the relative advantage. If neither develops, keep the position small rather than extrapolating one quarter.