Built by Codex Sol 5.6AMZN | Earnings Review — Q2 2026
Amazon delivered a real operating acceleration, not just a headline beat. Net sales rose from $167.7B to $200.6B, while AWS revenue rose from $30.9B to $42.2B. AWS growth accelerated for a fifth consecutive quarter to 37%, the fastest rate in 18 quarters. Advertising accelerated to 26%, and both North America and International grew double digits. Q2 consolidated operating income increased 43% to $27.5B.
The quality of the profit result is good, but not as pristine as the headline. Q2 included about $600M of tariff-related refunds and about $600M of favorable energy-derivative remeasurement. Excluding both, operating income would have been roughly $26.3B, still about 37% above last year and a 13.1% margin versus 11.4% reported a year ago. AWS margin expanded about 650bps as reported and about 520bps excluding the derivative benefit. By contrast, the $5.75 GAAP EPS result is not useful for judging the core business because Amazon recorded $53.4B of pre-tax other income, primarily related to its Anthropic investments. Amazon Q2 8-K, p. 1
The new negative is cash conversion. Trailing operating cash flow rose 33% to $161.4B, but trailing net purchases of property and equipment rose roughly 64% to $169.0B. As a result, trailing free cash flow fell from $18.2B to negative $7.6B. Management raised expected 2026 cash CapEx from approximately $200B to approximately $220B because memory costs rose, while still expecting insufficient capacity through 2027. Q2 transcript, p. 5
Guidance is better than the headline growth rate looks, but the sales hurdle is not soft. Q3 sales guidance is $197B to $202B, or 9% to 12% growth. Prime Day moved into Q2 this year from Q3 last year; management says Q3 growth excluding the event in both periods would be nearly 400bps higher. Current FMP consensus of about $202.0B sits at the guide high, while the FactSet consensus reported on the announcement date was $203.9B. Amazon Q2 release, Associated Press
The pre-print AWS hurdle was materially lower than the result: S&P/Visible Alpha expected $40.5B of AWS revenue and a 33.8% segment margin, versus $42.2B and 39.4% reported—or about 37.9% after removing the disclosed derivative benefit. Management's repeated claim that demand was capacity-constrained therefore beat both the growth and normalized-margin skepticism embedded in consensus. S&P/Visible Alpha Q2 preview
The bull case won the demand argument: AWS scarcity, a $496B backlog and reserved 2027 capacity validate that the build is responding to contracted demand. It has not yet won the returns argument. The next rating change should be driven by cash conversion and normalized AWS economics—not by GAAP EPS marks or a single quarter of reported margin.
| Metric | Q3'24 | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 |
|---|---|---|---|---|---|---|---|---|
| Net sales ($B) | 158.9 | 187.8 | 155.7 | 167.7 | 180.2 | 213.4 | 181.5 | 200.6 |
| Reported sales growth | +11.0% | +10.5% | +8.6% | +13.3% | +13.4% | +13.6% | +16.6% | +19.6% |
| AWS revenue ($B) | 27.5 | 28.8 | 29.3 | 30.9 | 33.0 | 35.6 | 37.6 | 42.2 |
| AWS growth | +19.1% | +18.9% | +16.9% | +17.5% | +20.2% | +23.6% | +28.4% | +36.8% |
| Advertising revenue ($B) | 14.3 | 17.3 | 13.9 | 15.7 | 17.7 | 21.3 | 17.2 | 19.8 |
| Advertising growth | +18.8% | +18.0% | +17.7% | +22.9% | +23.5% | +23.3% | +23.9% | +26.2% |
| Operating income ($B) | 17.4 | 21.2 | 18.4 | 19.2 | 17.4 | 25.0 | 23.9 | 27.5 |
| TTM free cash flow ($B) | 47.7 | 38.2 | 25.9 | 18.2 | 14.8 | 11.2 | 1.2 | −7.6 |
Management repeatedly argued that supply, not demand, constrained AWS. The pre-print S&P/Visible Alpha hurdle was 31% growth and a 33.8% margin; Q2 delivered 36.8% and 39.4% reported, or roughly 37.9% ex-derivative. The debate has moved from reacceleration to whether capacity can be installed without eroding returns.
Segment operating quality
AWS consensus test — the result beat both growth and margin skepticism
| Metric | Pre-print S&P/Visible Alpha | Q2 actual | Variance | Normalized read |
|---|---|---|---|---|
| AWS revenue | $40.5B | $42.2B | +$1.7B / +4.3% | Reported growth was 36.8% versus roughly 31% implied by consensus. |
| AWS operating margin | 33.8% | 39.4% | +556bps | About 37.9% after removing the $600M derivative benefit, still about +414bps above consensus. |
This is the cleanest management-contrarianism evidence in the print: capacity-constrained demand converted faster than the Street expected, and efficiency more than absorbed early AI mix pressure even after the accounting benefit is removed.
| Segment / metric | Q2'25 | Q2'26 | YoY change | Interpretation |
|---|---|---|---|---|
| AWS revenue | $30.9B | $42.2B | +36.8% | Demand acceleration is broad across AI and core cloud workloads. |
| AWS operating income | $10.2B | $16.6B | +63.6% | Reported margin 32.9% → 39.4%; about 38.0% excluding the derivative benefit. |
| North America revenue | $100.1B | $116.2B | +16.1% | Prime Day helped timing, but everyday essentials and delivery speed remained strong. |
| North America operating income | $7.5B | $9.1B | +21.4% | Margin expanded about 34bps despite fuel and line-haul pressure. |
| International revenue | $36.8B | $42.2B | +14.8% | Healthy growth, but margin was roughly flat at 4.1%. |
| International operating income | $1.5B | $1.7B | +14.9% | Profit grew with sales; no additional margin inflection. |
Verdict — accelerating where it matters, deteriorating in cash conversion. The AWS YoY growth rate improved by about 840bps in one quarter, advertising by about 230bps, online stores by about 270bps and third-party seller services by about 200bps. Reported consolidated sales growth rose about 300bps to 19.6%. That operating breadth is unusually strong. The offset is seven successive declines in trailing free cash flow across the eight observations shown, driven by an infrastructure build whose financial payback remains ahead rather than in the reported numbers.
| Quarter | Revenue actual | Consensus | Surprise | GAAP EPS actual | Consensus | Surprise / quality |
|---|---|---|---|---|---|---|
| Q3'24 | $158.9B | $157.3B | +1.0% beat | $1.43 | $1.14 | +25.4% beat |
| Q4'24 | $187.8B | $187.3B | +0.2% beat | $1.86 | $1.49 | +24.8% beat |
| Q1'25 | $155.7B | $155.1B | +0.3% beat | $1.59 | $1.37 | +16.1% beat |
| Q2'25 | $167.7B | $161.8B | +3.7% beat | $1.68 | $1.31 | +28.2% beat |
| Q3'25 | $180.2B | $177.9B | +1.3% beat | $1.95 | $1.57 | +24.2% beat |
| Q4'25 | $213.4B | $211.5B | +0.9% beat | $1.95 | $1.97 | −1.0% miss |
| Q1'26 | $181.5B | $177.3B | +2.4% beat | $2.78 | $1.63 | +70.6% beat* |
| Q2'26 | $200.6B | $197.0B | +1.8% beat | $5.75 | $1.82 | +215.9% beat* |
* Q1 and Q2 2026 EPS comparisons are distorted by investment valuation marks; Q2 included $53.4B of pre-tax other income primarily related to Anthropic.
| Window | Revenue beat rate | EPS beat rate | Average revenue surprise | Analytical read |
|---|---|---|---|---|
| Last 12 quarters | 11 / 12, 91.7% | 11 / 12, 91.7% | +1.76% | Consistent beater; revenue surprise magnitude is not widening. |
| Last 4 quarters | 4 / 4, 100% | 3 / 4, 75.0% | +1.60% | Top-line execution remains dependable; EPS quality has worsened. |
FMP's ebitAvg of $15.7B is not a comparable operating-income consensus, so this review does not present the implied 74.6% variance as a clean Street beat. That figure is retained only in the trace workspace as a low-confidence fallback. A proper point-in-time operating-income comparison requires Bloomberg or Visible Alpha.
earnings and analyst-estimates, pulled August 7, 2026; actuals from Daloopa. Segment expectations are shown separately in the AWS consensus test above.| Metric | New guide | Midpoint | External expectation | Midpoint variance | YoY at midpoint |
|---|---|---|---|---|---|
| Net sales | $197B–$202B | $199.5B | $202.0B current FMP; $203.9B FactSet at announcement | −1.2% vs. FMP; −2.2% vs. FactSet | +10.7% reported; ~+14.7% ex-Prime Day |
| Operating income | $22.5B–$26.5B | $24.5B | No high-quality comparable source available | N/M | +40.6% vs. $17.4B |
| GAAP EPS | Not guided | N/M | $1.96 FMP | N/M | N/M |
There was no previous Q3 2026 guide, so this is not a revision table. The range assumes roughly 80bps of FX pressure and no energy-derivative remeasurement benefit. Amazon also gave no formal AWS revenue, segment margin, free-cash-flow or full-year 2027 guidance. Q2 8-K, p. 4
Current consensus is at the high end; the at-announcement hurdle was above it. The calendar bridge explains the growth rate, not the dollars.
Guide execution and forward signals
| Item | Evidence | Investment implication |
|---|---|---|
| Q2 execution versus prior guide | Sales actual $200.6B vs. $194B–$199B; operating income $27.5B vs. $20B–$24B | Actuals cleared both high ends; about $1.2B of expense benefits helped operating income, but do not explain the full clearance. |
| AWS capacity | $496B backlog; 2027 capacity largely reserved; some 2028 capacity reserved | Demand visibility is stronger than the reported quarter, but installation cadence gates revenue. |
| 2026 cash CapEx | Approximately $220B, raised from approximately $200B because of memory costs | Higher demand confidence and higher execution/return risk arrive together. |
| Asset economics | Servers generally break even in under three years; most current AI capacity carries contracts of at least five years; data centers have lives of 30 years or more | Management offers a credible duration framework, but not cohort ROIC or sensitivity to input inflation. |
| Stores calendar | Prime Day shifted from Q3'25 into Q2'26 in most large countries | Do not confuse the Q3 reported growth slowdown with a comparable demand collapse. |
Management's recent numerical guidance has been conservative: across the nine source-checked realized quarters from Q2 2024 through Q2 2026, both sales and operating income finished above the high end in 7 quarters and inside the range in 2. Neither metric fell below the guide low. That makes the Q3 range more useful as a floor-and-execution framework than as a neutral forecast, although an external sales consensus at the high end prevents an easy setup.
| Realized quarter | Prior-quarter sales guide | Sales actual / outcome | Prior-quarter OI guide | OI actual / outcome |
|---|---|---|---|---|
| Q2'24 | $144.0B–$149.0B | $148.0B / inside | $10.0B–$14.0B | $14.7B / above |
| Q3'24 | $154.0B–$158.5B | $158.9B / above | $11.5B–$15.0B | $17.4B / above |
| Q4'24 | $181.5B–$188.5B | $187.8B / inside | $16.0B–$20.0B | $21.2B / above |
| Q1'25 | $151.0B–$155.5B | $155.7B / above | $14.0B–$18.0B | $18.4B / above |
| Q2'25 | $159.0B–$164.0B | $167.7B / above | $13.0B–$17.5B | $19.2B / above |
| Q3'25 | $174.0B–$179.5B | $180.2B / above | $15.5B–$20.5B | $17.4B / inside |
| Q4'25 | $206.0B–$213.0B | $213.4B / above | $21.0B–$26.0B | $25.0B / inside |
| Q1'26 | $173.5B–$178.5B | $181.5B / above | $16.5B–$21.5B | $23.9B / above |
| Q2'26 | $194.0B–$199.0B | $200.6B / above | $20.0B–$24.0B | $27.5B / above |
The nine-quarter window is the longest adjacent period independently re-pulled and source-linked for both revenue and operating-income guidance in this review. It differs from the 12-quarter consensus-surprise window because the data questions are different: FMP provides a 12-quarter reported-versus-consensus series, while the management-range audit requires a linked prior-quarter guide for both metrics.
| Fiscal year | 2021 | 2022 | 2023 | 2024 | 2025 |
|---|---|---|---|---|---|
| Net sales ($B) | 469.8 | 514.0 | 574.8 | 638.0 | 716.9 |
| Sales growth, ex-FX | 21% | 13% | 12% | 11% | 12% |
| AWS revenue ($B) | 62.2 | 80.1 | 90.8 | 107.6 | 128.7 |
| AWS revenue growth | 37.1% | 28.8% | 13.3% | 18.5% | 19.7% |
| Operating income ($B) | 24.9 | 12.2 | 36.9 | 68.6 | 80.0 |
| Operating margin (calculated) | 5.3% | 2.4% | 6.4% | 10.8% | 11.2% |
| Free cash flow ($B) | −9.1 | −11.6 | 36.8 | 38.2 | 11.2 |
Annual GAAP EPS is deliberately omitted: the 2022 split and volatile non-operating investment marks make a five-year per-share series less decision-useful than operating income and cash conversion.
The trajectory has three distinct regimes. First, 2021–2022 combined rapid sales and AWS growth with retail overcapacity and negative free cash flow. Second, the 2023–2024 regionalization and cost-reset period produced a large operating-income and free-cash-flow recovery. Third, 2025–2026 reaccelerated AWS and operating income while reopening a larger infrastructure cycle; unlike the prior recovery, cash spend is now rising faster than operating cash flow.
Largest current inflection: AWS growth moved 17% → 20% → 24% → 28% → 37% over five quarters. This is not a base-effect-only bounce: management cites capacity coming online, enterprise migrations, AI workloads pulling CPU/storage/database consumption, and both AI and chips businesses exceeding $25B annual run rates with triple-digit growth. The contrary evidence is economic rather than demand-related: trailing free cash flow moved $18.2B → $14.8B → $11.2B → $1.2B → negative $7.6B over the same five reported quarters.
| Priority | Catalyst / risk | Timing | What to monitor |
|---|---|---|---|
| 1 | AWS capacity converts backlog into revenue | Q3 2026 through 2027 | Growth acceleration, power additions, backlog conversion and whether AI continues pulling core cloud usage. Positive if supply comes online without margin dilution. |
| 2 | Memory inflation versus AI returns | Q3 and 2026 build | Cash CapEx cadence, normalized AWS margin, contract repricing and storage availability. The $20B CapEx increase raises the hurdle. |
| 3 | Prime Day comparison normalizes | Q3 2026 | Ex-event sales rate, units, active customers and cost-to-serve. Current sales consensus leaves little room below the guide high. |
| 4 | Same-day perishables changes shopping frequency | Q3 onward | Active perishable customers grew more than 50% since year-start; qualifying baskets contain about three times more units across 2,300 U.S. cities. Q2 transcript, p. 15 |
| 5 | Amazon Leo initial service | Prior Q3 target; current language says 2026 | First paid service, coverage and terminal availability. The less precise wording is a schedule watch, not yet a missed commitment. |
| 6 | Trainium and application-layer adoption | Q3 and re:Invent, Nov. 30–Dec. 4 | External Trainium economics, commitments, Bedrock consumption, Kiro/Q production use and new services. AWS re:Invent |
| 7 | EU cloud DMA designation | Timing uncertain | Interoperability, switching and contract remedies after the Commission's preliminary gatekeeper view for AWS and Azure. European Commission |
The six analyst exchanges contained 11 separately scored questions: 3 well answered and 8 deflected or incomplete. Management was strongest on strategy and adoption evidence, and least forthcoming on the timing and economics needed to underwrite the capital cycle.
| Topic | Verdict | What management provided | What remains missing |
|---|---|---|---|
| AWS margin sustainability | Deflected | 650bps reported and 520bps ex-derivative expansion; efficiency and capacity optimization drivers | Normalized range, AI/core mix sensitivity and forward cost cadence |
| Need for an Amazon frontier model | Answered | Bedrock can win on model choice; owning a model improves control of cost, priorities and speed | No capability or launch timing, but the strategic question was answered |
| H2 capacity cadence | Deflected | Reiterated target to double power capacity by end-2027 | H2 versus H1 additions and the portion of Q2 acceleration caused by new supply |
| 2027 long-lived data-center spend | Deflected | Most 2027 capacity and some 2028 capacity are reserved | 2027 cash CapEx or a deceleration framework |
| External Trainium timing and ROIC | Deflected | Active discussions; a real chance of future external sales | Timing, allocation tradeoff and comparative return |
| Application layer | Answered | Direct strategic rationale and examples across Q, Transform, Connect and Continuum | Product economics were not requested or supplied |
| Funding the build | Deflected | Debt was issued and Amazon has multiple options | Funding mix, leverage guardrails and duration |
| Existing-contract inflation protection | Deflected | Signed contracts retain pricing; new contracts reflect current costs | Repricing lag, contract-duration mix and margin sensitivity |
| RPO implications for 2028 capacity | Deflected | Backlog is incorporated in the capital plan and more deals are expected | No update to the power-capacity target or post-2027 build framework |
| Grocery adoption | Answered | Scale, city coverage, active-customer growth and units per basket | International cohort comparison was omitted |
| Grocery results by country / geography | Deflected | U.S. adoption, Whole Foods geographies and 2,300 U.S. cities | No country comparison or international cohort data |
Most decision-relevant omissions: 2027 CapEx, H2 capacity cadence, inflation exposure on contracted workloads, and external Trainium returns. Those are exactly the inputs required to move from “demand is strong” to “the capital build creates attractive incremental returns.”
Q2 2025: management described Nova as Amazon's own frontier model.
Q2 2026: management said Amazon is pursuing its own frontier model and expects one of the leading models to be Amazon's in the future.
The frontier threshold can move, so this is not proven false under every interpretation. Amazon nevertheless did not bridge the terminology or identify a model-generation change. Do not give Amazon credit for proprietary frontier-model parity today; underwrite Bedrock's model choice and infrastructure position instead.
| Tension reviewed | Why it is not yet a contradiction | Watch item |
|---|---|---|
| Cash CapEx raised to $220B after Q1 said the plan was largely unchanged | It is an explicit forecast revision as costs changed | Whether Q1's known memory inflation should have been incorporated earlier; risk of another revision |
| Leo shifted from “Q3” to “this year” | Q3 remains inside the current wording | Failure to start Q3 service would turn softer language into a missed commitment |
| AWS margin included a derivative gain | Management explicitly reconciled 650bps reported to 520bps underlying | Lack of a forward normalized range is a disclosure gap, not a contradiction |
| Trainium sales outside AWS moved from possibility to active exploration | The statements describe progression | Timing and relative ROIC remain unanswered |
Credibility verdict: Amazon's numerical sales and operating-income ranges have been consistently conservative, while longer-duration capital and product-timing disclosure is less precise. Trust the near-term range mechanics; require more proof on multi-year cash returns and milestones.
| Signal | Read-through |
|---|---|
| AI infrastructure scarcity | Positive for power, cooling, networking, GPU/custom silicon and storage demand; negative for near-term hyperscaler free cash flow and customers waiting for capacity. The reservation evidence is stronger than generic AI-TAM claims. |
| Memory, HDD and SSD inflation | Positive demand/pricing signal for component categories; margin risk for fixed-price cloud contracts until utilization and repricing catch up. No specific memory supplier was named. |
| Consumer resilience is essentials-led | Grocery, convenience and low-price engagement are strong; this is not evidence that discretionary demand is uniformly strong. |
| Fuel and line-haul inflation | Negative for retail/parcel fulfillment margins, partly offset by an FBA surcharge and Amazon's scale. |
| Tariffs | Most suppliers are importers of record; Amazon largely absorbed direct increases where it bore them. The $600M refund is non-recurring and should not be annualized. |
| Enterprise cloud | Transformation plans and AI pull-through imply durable cloud budgets, but the call does not prove broader enterprise adoption will match frontier-lab demand. |
Named ecosystem implications
| Parties | Directional read-through |
|---|---|
| Anthropic and OpenAI | Multi-year, multi-gigawatt Trainium commitments validate custom silicon at frontier-lab scale; also raise concentration and execution dependence. |
| Nvidia | Amazon expects Nvidia demand on AWS to persist, positive for the ecosystem; Trainium scale and possible external sales create a longer-term share tension. |
| Uber, Pinterest and other named Trainium adopters | Evidence that custom-silicon demand extends beyond two frontier labs, though management supplied no consumption or contract values. |
| Salesforce/Slack, Atlassian/Jira, Microsoft Teams, ServiceNow, Adobe, Moody's and Snowflake | Amazon Q can drive usage into partner systems but also seeks to become the orchestration layer above their interfaces. |
| P&G, 3M, Lands' End and American Eagle | Positive validation for Amazon Supply Chain Services; a potential share risk to incumbent third-party logistics networks. |
| NBA, WNBA, NFL and NASCAR | Sold-out inventory across major sports supports Amazon Ads and rights monetization; rights-cost returns remain undisclosed. |
| Microsoft Azure | The EU preliminary gatekeeper view validates scale but increases regulatory risk for both AWS and Azure. |
Upgrade evidence: AWS sustains at least high-20s growth as capacity comes online; normalized AWS margins remain structurally above the prior-year level; trailing free cash flow inflects upward despite the $220B build; management provides a 2027 CapEx framework and clearer contract-level return protection.
Downgrade evidence: AWS growth slows mainly because of execution rather than demand; memory/storage inflation compresses normalized AWS margins; the capital plan rises again without a corresponding backlog or return bridge; signed workload pricing produces a longer-than-expected margin lag; Leo or other disclosed milestones slip without explanation.
The next-quarter scoreboard: Q3 sales near or above the $202B guide high, Prime Day-normalized Stores growth, AWS capacity additions and growth, AWS margin excluding accounting gains, cash CapEx, and a more complete 2027 capital framework.
Data sourced from Daloopa (company_id 15), Amazon's Q2 2026 earnings materials and transcript, and FMP consensus snapshots pulled August 7, 2026. Where higher-quality Bloomberg or Visible Alpha consensus was unavailable, the limitation is explicit and no estimate was invented.