The cloud business is growing faster than its cash returns
Amazon's third-quarter results contain two figures that appear contradictory until their time periods and definitions are separated. AWS sales rose 39% to $46 billion, the cloud division's fastest growth in nearly five years. But group free cash flow for the trailing twelve months was negative $42.4 billion, compared with positive $14.8 billion a year earlier. Customer demand is accelerating while construction and equipment purchases are absorbing more cash than operations generate.
The outflow is not the loss reported for a single quarter. Amazon generated $168.3 billion in trailing operating cash flow, up 29%, while net purchases of property and equipment reached approximately $210.7 billion. Management attributes the increase primarily to artificial intelligence infrastructure. Investors should not mistake strong cloud sales for evidence that the newest computing facilities have already repaid their cost.
Operating margins provide the other half of the picture
Consolidated quarterly revenue rose 12% to $201.9 billion. Operating income increased to $28.1 billion, producing a record 13.9% operating margin. AWS operating income climbed to $17.8 billion from $11.4 billion a year earlier. The company is not trying to invent a profitable business model: its established cloud operation is already highly profitable. The uncertainty concerns the return on the next wave of servers, data centres, networking and electricity infrastructure.
Amazon expects about $220 billion of cash capital expenditure in 2026 and says spending will be meaningfully higher in 2027. Those are management plans rather than realised future results. Capacity that comes online quickly and runs near full utilisation can justify high initial spending. Capacity delayed by power connections or built ahead of customer demand creates a much longer payback period.
Retail still finances the infrastructure ambition
North American sales rose 6% to $112.9 billion and international sales increased 5% to $43.1 billion. Comparisons were affected by Prime Day moving into the second quarter; management estimated that normalising its timing would add roughly four percentage points to consolidated third-quarter growth. Retail profitability therefore deserves attention alongside AWS. The group's established commerce operations provide cash and scale that a pure cloud challenger may not possess.
Reported net income slipped to $20.7 billion from $21.2 billion, even though operating income rose. A large investment-related gain in the year-earlier quarter distorted the net-profit comparison. Looking only at earnings per share would obscure operating improvement; looking only at AWS growth would obscure cash demands. A useful valuation has to accommodate both.
Guidance is not a signed order book
Amazon forecast fourth-quarter sales of $240 billion to $247 billion and operating income of $28.5 billion to $33.5 billion. The revenue midpoint fell below market expectations reported by Reuters, prompting concern about the holiday period. Management also projected AWS growth exceeding 40% in 2027 and 2028. Those projections remain conditional on customer spending, capacity delivery, pricing and wider economic conditions.
The $184 billion AWS annualised revenue run rate is simply four times one quarter's sales. It is neither next year's reported revenue nor a contractual backlog. Investors should compare the trajectory of operating profit with the capital required to produce each additional dollar of sales, especially as newer assets begin depreciating.
The next test is capital productivity
The most informative measures will be AWS operating margins, actual equipment spending, utilisation of new facilities and free cash flow over successive quarters. A temporary construction surge could eventually yield stronger cash generation, but sustained investment requirements could keep it weak even as revenue rises. Neither outcome is established by one earnings release.
Amazon's 29 October report confirms exceptional cloud demand and a highly profitable existing business. It also confirms that the next phase of growth is expensive. For shareholders, the relevant question is not whether AI workloads are growing, but whether the assets built to serve them can produce adequate returns after their full cost.