Amazon’s third quarter answers the first half of the AI-investment question but not the second. AWS revenue grew 39.3% year over year and segment operating income rose 55.4%, evidence that demand and operating leverage are real. Yet Amazon’s trailing-12-month free cash flow, under the company’s non-GAAP definition, swung to a US$42.4 billion outflow as net property-and-equipment purchases climbed 81.8% to US$210.7 billion. The result is not proof that AI infrastructure earns a poor return. It does mean investors now need growth, margins and future cash conversion—not the AWS growth rate alone—to justify the capital intensity.
Three findings
- AWS added US$13.0 billion of quarterly revenue and US$6.3 billion of operating income year over year; the implied incremental operating margin was 48.8%, an AHNIM calculation rather than a company metric.
- Amazon generated US$168.3 billion of trailing operating cash flow, but its US$210.7 billion of net property-and-equipment purchases exceeded that inflow. The company said the investment increase primarily reflected AI spending.
- The financial test moves forward: management projects AWS revenue growth above 40% in 2027 and 2028, while investors still need evidence that installed capacity converts into durable free cash flow after the current buildout.
AWS accelerated and became more profitable at the same time
| Measure | Q3 2025 | Q3 2026 | Change |
|---|---|---|---|
| Amazon net sales | US$180.169bn | US$201.944bn | +12.1% |
| Amazon operating income | US$17.422bn | US$28.076bn | +61.2% |
| Consolidated operating margin | 9.7% | 13.9% | +4.2pp |
| AWS sales | US$33.006bn | US$45.984bn | +39.3% |
| AWS operating income | US$11.434bn | US$17.773bn | +55.4% |
| AWS operating margin | 34.6% | 38.7% | +4.0pp |
The quality of the AWS result is stronger than a headline revenue acceleration alone. Segment operating income grew faster than sales, lifting the AWS operating margin by roughly four percentage points. Subtracting the prior-year amounts from the current quarter gives US$12.978 billion of incremental AWS sales and US$6.339 billion of incremental operating income. Dividing the second figure by the first produces an implied 48.8% incremental operating margin.
That calculation is useful but easy to overextend. It is not a disclosed long-run margin target, and quarterly segment profit can move with capacity utilization, pricing, depreciation, energy, compensation and the mix of higher- and lower-margin workloads. The defensible conclusion is narrower: in this quarter, rapid AWS growth did not require a lower reported segment margin.
Amazon described the US$45.984 billion quarterly AWS figure as an annualized revenue run rate above US$184 billion. That is simply the quarter multiplied by four—US$183.936 billion—not actual trailing-12-month AWS revenue. The four reported quarters sum to approximately US$161.4 billion. We keep those measures separate because an annualized exit rate can overstate the revenue already earned when growth is accelerating.
The cash-flow bridge is the central investment question
| Trailing-12-month measure | September 2025 | September 2026 | Change |
|---|---|---|---|
| Operating cash flow | US$130.7bn | US$168.3bn | +28.8% |
| Net purchases of property and equipment | US$115.903bn | US$210.657bn | +81.8% |
| Free cash flow | US$14.788bn | (US$42.357bn) | (US$57.145bn) swing |
Operating cash flow improved by about US$37.6 billion, but net property-and-equipment purchases increased by US$94.8 billion. The second change overwhelmed the first. The arithmetic reconciles directly: US$168.300 billion minus US$210.657 billion equals a US$42.357 billion free-cash-flow outflow.
This is a better description than saying Amazon “stopped generating cash.” Operations generated substantial cash. Amazon then invested more cash in property and equipment than operations produced. The company said the increase in net purchases primarily reflected artificial-intelligence investment. “Primarily” is important: logistics, fulfillment, transportation, offices and other infrastructure remain inside Amazon’s capital base, so we do not attribute every dollar of investment or every dollar of the free-cash-flow decline to AI.
The free-cash-flow measure also has limits. It is non-GAAP, reflects the timing of cash purchases and incentives, and does not by itself capture every finance-lease commitment or unpaid property-and-equipment obligation. It remains useful because it exposes the current cash burden, but it is not a complete project-return calculation.
US$220 billion of 2026 cash capex is a forecast, not the same denominator
Amazon maintained an expectation of approximately US$220 billion of cash capital expenditures in 2026. That forward-looking full-year measure should not be equated with the US$210.657 billion of trailing-12-month net property-and-equipment purchases. The periods differ, and the latter is net of proceeds and incentives. Treating the two numbers as interchangeable would manufacture precision that the disclosures do not support.
For context, third-quarter property-and-equipment purchases were US$77.259 billion, up from US$35.095 billion a year earlier. For the first nine months, gross purchases were US$175.670 billion versus US$92.297 billion. The cash commitment therefore accelerated visibly before applying any long-run revenue assumption.
The economic payoff can arrive later than the cash outlay. Data centers and power infrastructure are paid for before all customer capacity is occupied; depreciation then spreads part of the accounting expense over future periods. That timing supports neither an automatic bullish nor bearish conclusion. It creates a sequence investors can test: cash deployment first, usable capacity second, customer consumption third, and durable operating cash flow last.
The balance sheet can fund the buildout, but financing still matters
At September 30, Amazon reported US$55.783 billion of cash and cash equivalents and US$58.443 billion of marketable securities. Long-term debt was US$173.125 billion, up from US$65.648 billion at December 31. The nine-month cash-flow statement also shows US$115.1 billion of long-term debt proceeds.
It would be wrong to label that entire debt increase “AI financing.” During the same nine months, Amazon reported US$63.753 billion of cash used for acquisitions and other investments, while borrowings can support general corporate purposes and refinance other obligations. The more defensible reading is that the company has increased both investment intensity and balance-sheet financing at a time when AI infrastructure, acquisitions and other needs compete for capital.
The next balance-sheet test is not whether Amazon can issue debt. It is whether incremental operating cash flow eventually grows faster than the recurring capital required to sustain cloud growth. A capacity cycle funded comfortably today can still destroy value if utilization, pricing or useful life falls short. Conversely, a temporary free-cash-flow deficit can create value if installed capacity earns attractive returns for many years.
Management’s 2027–2028 AWS projection raises the bar
Amazon projects AWS revenue growth above 40% in both 2027 and 2028. This is a management outlook, not a completed result or an audited figure. It should also be separated from the formal fourth-quarter company guidance: Amazon expects Q4 net sales of US$240–247 billion, growth of 12–16%, with foreign exchange reducing growth by about 60 basis points, and operating income of US$28.5–33.5 billion versus US$25.0 billion a year earlier.
The cloud projection makes the capital decision more testable. If AWS sustains growth above 40% while maintaining healthy segment margins, utilization of the new capacity should rise quickly. If growth slows before the buildout matures, the same fixed infrastructure could pressure returns. Neither outcome follows from one quarter.
Company-reported customer activity—including agreements involving Meta, NVIDIA, OpenAI, Salesforce, Oracle, Qualcomm and Verizon—supports the demand narrative, but does not disclose contract revenue, deployment timing or incremental profit by customer. Announced agreements should not be counted as already recognized sales. The same distinction applies to Amazon’s statement that Bedrock spending by Fortune 500 companies rose nearly sevenfold and token volume nearly seventeenfold: those are usage indicators, not a disclosed Bedrock revenue line.
Broker targets fell even as most cited analysts kept positive ratings
| Firm | Reported October 30 action | Rating |
|---|---|---|
| Oppenheimer | US$320 → US$285 | Buy maintained |
| Scotiabank | US$325 → US$300 | Outperform maintained |
| Wells Fargo | US$329 → US$326 | Buy maintained |
| KeyBanc | US$350 → US$345 | Buy maintained |
| Mizuho | US$330 → US$335 | Reported increase |
The dispersion is informative. Oppenheimer’s larger reduction was reported alongside concern about near-term margins and cash flow, while KeyBanc and Wells Fargo made smaller cuts and emphasized longer-term AWS growth. Mizuho moved in the opposite direction. This is evidence of different assumptions—not a substitute for a valuation model.
TipRanks’ October 30 snapshot listed 45 Buy and two Hold ratings and an average target of US$324.23. AHNIM is not initiating an Amazon Bear/Base/Bull price target in this event review. We therefore cannot make a comparable house Base-versus-consensus gap calculation, and we do not convert the broker average into an AHNIM fair value.
Our assessment: stronger operating evidence, a harder cash test
The quarter strengthens the operating case behind Amazon’s AI investment. AWS growth accelerated, segment margin expanded, and the incremental profit conversion was strong. It also makes the valuation case more demanding because capital investment grew much faster than operating cash flow.
We would treat the thesis as improving if three things occur together: AWS remains near management’s growth path, segment margin holds up as more AI workloads scale, and free cash flow recovers after the buildout without relying on a sharp reduction in necessary capacity. We would become more cautious if growth decelerates while net property-and-equipment purchases remain structurally above operating cash flow, or if debt continues to rise without a matching increase in durable earnings power.
This framework also clarifies the relevance to infrastructure suppliers. Amazon’s spending supports demand for servers, networking, power and cooling, but hyperscaler capex is not the same as supplier revenue recognized in the same period. Investors should follow customer advances, delivery schedules, working capital and cash collection in our Vertiv Q3 cash-flow analysis, rather than treat Amazon’s budget as a direct one-for-one sales forecast.
Related coverage
Our TorchNeuron analysis identified paid usage and returns on installed capacity as the business milestones behind easier Trainium adoption. The Q3 result adds evidence on aggregate AWS growth and margin, but still does not isolate Trainium revenue or profit. For a hyperscaler comparison, see Microsoft’s Q1 capex and cash-flow test and Alphabet’s Q3 AI-capex analysis.
Investment information, not personalized advice. Company outlooks, broker views and AHNIM calculations are distinguished above. See our research, AI-use and corrections policy.
Sources and original documents
Amazon — Q3 2026 earnings release, Exhibit 99.1 to Form 8-K, filed October 29, 2026
Amazon — Form 10-Q for the quarter ended September 30, 2026, filed October 29, 2026
Amazon — Form 8-K cover filing, October 29, 2026
The Fly via TipRanks — Scotiabank target and rating action, October 30, 2026; secondary reporting
The Fly via TipRanks — Mizuho target action, October 30, 2026; secondary reporting
Response
[…] results update — October 31, 2026 KST: Amazon Q3: AWS grew 39% as AI capex pushed free cash flow negative. This update adds aggregate AWS growth, segment-margin, capital-spending and cash-conversion […]