Built by Claude FableAMZN | Earnings Review — 2026 Q2 (Claude)
BUY
NASDAQ: AMZN | Q2 2026 reported July 30, 2026 AMC · analysis date August 3, 2026 · Daloopa company_id 15
Revenue Beat
+1.8%
$200.6B vs $197.0B street; +20% YoY — fastest in 20 quarters
AWS
+36.7%
$42.2B — 5th straight accel; 39.4% margin; $496B RPO backlog
Operating Income
$27.5B
Record 13.7% margin; +$3.5B above guide high end
Capex / FCF
$53.1B
FY26 guide raised to ~$220B; TTM FCF −$7.6B
The AI print of the season: everything accelerated at once, and management re-priced the AWS story. Amazon delivered $200.6B revenue (+20% YoY, its fastest growth in 20 quarters; +1.8% vs $197.0B street), record operating income of $27.5B (+43% YoY, 13.7% margin — clearing the top of its own $20–24B guide by $3.5B), and headline EPS of $5.75 — which is not clean: it includes a $53.4B pre-tax non-operating gain, primarily the Anthropic mark; ex-gain EPS is roughly $1.95, +16% YoY vs the $1.82 street comp. Every major revenue line accelerated: AWS +36.7% to $42.2B (5th straight quarter of acceleration, fastest in 18 quarters, +$4.6B sequential add ~80% larger than any prior increase, 39.4% op margin — +650bps YoY, +520bps ex a one-time energy-derivative gain — and a $496B RPO backlog growing triple digits), advertising broke out of a five-quarter ~22% plateau to +26% ($19.8B) with the first hard agentic-monetization stats (shoppers who click a sponsored prompt convert 48% more often and spend 21% more), online stores +15%, 3P sellers +16%, NA +16%, International +15%. The call stacked up new disclosures: AWS TAM upgraded to "very possibly a trillion-dollar annual revenue business"; the first explicit ROIC framework (servers break even in under 3 years on 5–6-year lives, most AI capacity contracted on 5-plus-year terms, 30-plus-year data centers); chips business now a $25B+ run rate growing triple digits with Graviton5 revenue commitments up ~3x QoQ; Amazon confirmed it is "pursuing our own frontier model"; 2027 capacity is "largely reserved" and "quite a bit" of 2028 is already reserved; and off-cloud Trainium sales moved to "actively having those conversations." The bill: Q2 capex $53.1B (+69%), FY26 cash capex raised to ~$220B on memory-cost inflation, TTM FCF now −$7.6B, $67B of new long-term debt in H1 — and Jassy conceded in Q&A that existing fixed-price contracts do not reprice for memory inflation (only new deals do). The Q3 guide of $197–202B (+9–12%) and OI of $22.5–26.5B screens like deceleration but carries ~400bps of Prime Day timing drag plus ~80bps FX — clean growth is ~14–17%, i.e. no underlying slowdown was guided. Net read: the demand debate is over and the street has moved to underwriting returns and funding; the +15.3% move to $271.58 paid for the AWS acceleration and the $496B backlog, not the EPS optics.
AWS revenue YoY growth — acceleration streak
Fifth consecutive quarter of acceleration. The +$4.6B sequential revenue add was ~80% larger than any prior quarterly increase. AWS run rate $169B annualized.
Key metrics trends
| $M except EPS | Q1'24 | Q2'24 | Q3'24 | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 |
|---|---|---|---|---|---|---|---|---|---|---|
| Total net sales | 143,313 | 147,977 | 158,877 | 187,792 | 155,667 | 167,702 | 180,169 | 213,386 | 181,519 | 200,606 |
| — YoY % | +8.6% | +13.3% | +13.4% | +13.6% | +16.6% | +19.6% | ||||
| AWS rev | 25,037 | 26,281 | 27,452 | 28,786 | 29,267 | 30,873 | 33,006 | 35,579 | 37,587 | 42,232 |
| — YoY % | +16.9% | +17.5% | +20.2% | +23.6% | +28.4% | +36.8% | ||||
| North America rev | 86,341 | 90,033 | 95,537 | 115,586 | 92,887 | 100,068 | 106,267 | 127,083 | 104,143 | 116,177 |
| — YoY % | +7.6% | +11.1% | +11.2% | +9.9% | +12.1% | +16.1% | ||||
| International rev | 31,935 | 31,663 | 35,888 | 43,420 | 33,513 | 36,761 | 40,896 | 50,724 | 39,789 | 42,197 |
| — YoY % | +4.9% | +16.1% | +14.0% | +16.8% | +18.7% | +14.8% | ||||
| Advertising | 11,824 | 12,771 | 14,331 | 17,288 | 13,921 | 15,694 | 17,703 | 21,317 | 17,243 | 19,809 |
| — YoY % | +17.7% | +22.9% | +23.5% | +23.3% | +23.9% | +26.2% | ||||
| Subscriptions | 10,722 | 10,866 | 11,278 | 11,508 | 11,715 | 12,208 | 12,574 | 13,122 | 13,427 | 13,730 |
| — YoY % | +9.3% | +12.4% | +11.5% | +14.0% | +14.6% | +12.5% | ||||
| Operating income | 15,307 | 14,672 | 17,411 | 21,203 | 18,405 | 19,171 | 17,422 | 24,977 | 23,852 | 27,461 |
| — YoY % | +20.2% | +30.7% | +0.1% | +17.8% | +29.6% | +43.2% | ||||
| AWS op income | 9,421 | 9,334 | 10,447 | 10,632 | 11,547 | 10,160 | 11,434 | 12,465 | 14,161 | 16,621 |
| — YoY % | +22.6% | +8.9% | +9.4% | +17.2% | +22.6% | +63.6% | ||||
| Diluted EPS ($) | 0.98 | 1.26 | 1.43 | 1.86 | 1.59 | 1.68 | 1.95 | 1.95 | 2.78 | 5.75* |
| — YoY % | +62% | +33% | +36% | +5% | +75% | +242%* | ||||
| Capex (P&E) | 14,925 | 17,620 | 22,620 | 27,834 | 25,019 | 32,183 | 35,095 | 39,522 | 44,203 | 54,208 |
| — YoY % | +68% | +83% | +55% | +42% | +77% | +68% |
* Q2'26 EPS includes the $53.4B pre-tax non-operating gain (primarily Anthropic); clean ex-gain EPS ~$1.95, +16% YoY. Capex row is gross purchases of P&E per the cash-flow statement; net of proceeds, Q2'26 capex was $53.1B (+69% YoY). Historical figures link to Daloopa source cards (company_id 15); Q2'26 figures are from the SEC 8-K Ex-99.1 press release.
- AWS is the story: +36.7% YoY, the 5th consecutive quarter of acceleration (16.9% to 17.5% to 20.2% to 23.6% to 28.4% to 36.8%) and fastest growth in 18 quarters. The +$4.6B sequential add was ~80% larger than any prior quarterly increase; run rate $169B annualized; RPO backlog $496B (~3.3x AWS TTM revenue), growing triple digits; AI and chips businesses each exceed $25B run rates growing triple digits.
- AWS margin re-expanded to 39.4% — +650bps YoY from the 32.9% Q2'25 trough (+520bps ex the ~$600M energy-derivative gain) — with AWS op income +63.6% on +37% revenue, despite the depreciation load from the capex ramp.
- The acceleration is broad, not just AWS: online stores +9% to +15%, 3P seller services +12% to +16%, NA +12% to +16%, International (ex-FX) +11% to +15%, and advertising broke five quarters at ~19–22% to +26%. Only subscriptions were steady (+12%). Part of the retail step-up is Prime Day shifting into Q2.
- Record consolidated margin: 13.7% (vs 11.4% LY), operating income +43%. AWS is now 21% of revenue but 60.5% of operating income and supplied ~78% of the YoY increase in total op income — mix shift is the margin story.
- Headline EPS $5.75 is not comparable: strip the $53.4B pre-tax (primarily Anthropic) gain and clean EPS is ~$1.95, +16% YoY — below the +43% op-income growth because the tax rate normalized (12.8% to 22.5%) and interest expense more than doubled ($516M to $1,314M) on new debt.
- The cost of the AI build is now visible in cash flow: TTM FCF turned negative (−$7.6B, from +$25.9B five quarters ago) even as TTM operating cash flow grew +33% to $161.4B; long-term debt rose from $65.6B (Dec-25) to $128.9B (Jun-26); FY26 capex was guided up again to ~$220B.
- Watch NA margin into Q3: 7.9% printed but includes ~$600M of one-time tariff refunds (~7.4% ex-refund, vs 7.5% LY) — the retail-margin expansion story is thinner ex-items, and Q3 is NA's seasonally weak quarter (4.5% in Q3'25).
This quarter vs consensus
| Metric | Consensus / Guide | Actual | Variance | Read |
|---|---|---|---|---|
| Total revenue | $197.03B (FMP avg; highest est $198.78B) | $200.61B | +$3.57B / +1.8% | Beat — above the highest estimate; roughly double AMZN's +1.0% median beat |
| Revenue vs own guide | $194.0–199.0B | $200.61B | +$1.6B above high end (+0.8%) | 132% through the range; guide implied +15.7–18.7%, actual +19.6% |
| AWS revenue | ~$40.5B press bar (~+31% YoY) | $42.23B (+36.8%) | ~+4.2% | The beat that moved the stock — ~600bps growth beat; highest-margin mix |
| Operating income | Own guide $20.0–24.0B | $27.46B | +$3.46B above high end | Biggest OI guide-clear on record; ~$1.2B of one-time benefits inside, still +$2.3B above high end ex-items |
| EPS (reported) | $1.82 (FMP avg) | $5.75 | +215.9% — not meaningful | Distorted by the $53.4B pre-tax Anthropic gain (largely non-cash; taxes mostly deferred) |
| EPS (clean, ex-gain) | $1.82 | ~$1.88–1.95 | ~+3% to +7% | Solid mid-single-digit operating beat, driven by the OI upside |
| Advertising | ~22% trend | $19.81B (+26.2%) | ~+400bps accel | Broke the five-quarter plateau; agentic + live-sports driven |
Consensus sourcing: FMP quarterly consensus (31 revenue / 29 EPS analysts, Q2'26 row); AWS and press-bar figures are from pre-print press coverage and labeled approximate. FMP's Q2'26 EBIT/EBITDA consensus rows were internally inconsistent and discarded — operating income is judged against the company's own guide.
Beat / miss history (10 quarters)
| Report date | Quarter | EPS est | EPS act | EPS surprise | Rev est ($B) | Rev act ($B) | Rev surprise |
|---|---|---|---|---|---|---|---|
| 2024-04-30 | Q1'24 | 0.83 | 0.98 | +18.1% | 142.65 | 143.31 | +0.5% |
| 2024-08-01 | Q2'24 | 1.03 | 1.26 | +22.3% | 148.67 | 147.98 | −0.5% |
| 2024-10-31 | Q3'24 | 1.14 | 1.43 | +25.4% | 157.28 | 158.88 | +1.0% |
| 2025-02-06 | Q4'24 | 1.49 | 1.86 | +24.8% | 187.34 | 187.79 | +0.2% |
| 2025-05-01 | Q1'25 | 1.37 | 1.59 | +16.1% | 155.15 | 155.67 | +0.3% |
| 2025-07-31 | Q2'25 | 1.31 | 1.68 | +28.2% | 161.78 | 167.70 | +3.7% |
| 2025-10-30 | Q3'25 | 1.57 | 1.95 | +24.2% | 177.91 | 180.17 | +1.3% |
| 2026-02-05 | Q4'25 | 1.97 | 1.95 | −1.0% | 211.45 | 213.39 | +0.9% |
| 2026-04-29 | Q1'26 | 1.63 | 2.78 | +70.6%* | 177.28 | 181.52 | +2.4% |
| 2026-07-30 | Q2'26 | 1.82 | 5.75 | +215.9%* | 197.03 | 200.61 | +1.8% |
* Anthropic-gain-distorted (H1'26 non-operating income $69.4B, of which $53.4B in Q2; clean Q2'26 surprise ~+3–7%, clean Q1'26 ~+4%). Pattern: revenue 9 beats in 10 (median +1.0%) and EPS 9 beats in 10 (median ~+24.5% on non-distorted quarters). Q2'26 is the first quarter to clear the top of both the revenue and OI guidance ranges simultaneously.
Guidance deep dive
Q3 2026 guide
| Q3 2026 (base: Q3'25 $180.2B / OI $17.4B) | Low | Mid | High |
|---|---|---|---|
| Net sales | $197.0B | $199.5B | $202.0B |
| Optical YoY growth | +9.3% | +10.7% | +12.1% |
| + ~400bps Prime Day timing shift | +13.3% | +14.7% | +16.1% |
| + ~80bps FX headwind ("clean" growth) | +14.1% | +15.5% | +16.9% |
| Operating income | $22.5B | $24.5B | $26.5B |
| OI growth vs $17.4B reported base | +29.1% | +40.6% | +52.1% |
| OI growth vs $21.7B charge-adjusted base* | +3.6% | +12.8% | +22.0% |
| Implied op margin | 11.4% | 12.3% | 13.1% |
* Q3'25's $17.4B included $4.3B of special charges ($2.5B FTC settlement + $1.8B severance); management itself adjusted to $21.7B at the time but compares the new guide against the unadjusted base. The clean Q3 midpoint of ~15.5% underlying growth matches Q2'26 ex-Prime-Day (~+15.3%) almost exactly — no underlying deceleration is being guided; the 9–12% optical range is a calendar artifact. Guide assumes no energy-derivative remeasurements and ~$600M of tariff refunds not repeating.
Guide history
| Qtr | Revenue guide ($B) | Actual (YoY) | vs high end | OI guide ($B) | OI actual | OI vs high end |
|---|---|---|---|---|---|---|
| Q1'25 | 151.0–155.5 | 155.7 (+8.6%) | +$0.2B above | 14.0–18.0 | 18.4 | +$0.4B above |
| Q2'25 | 159.0–164.0 | 167.7 (+13.3%) | +$3.7B above | 13.0–17.5 | 19.2 | +$1.7B above |
| Q3'25 | 174.0–179.5 | 180.2 (+13.4%) | +$0.7B above | 15.5–20.5 | 17.4 / 21.7 adj | +$1.2B above (adj) |
| Q4'25 | 206.0–213.0 | 213.4 (+13.6%) | +$0.4B above | 21.0–26.0 | 25.0 | within range |
| Q1'26 | 173.5–178.5 | 181.5 (+16.6%) | +$3.0B above | 16.5–21.5 | 23.9 | +$2.4B above |
| Q2'26 | 194.0–199.0 | 200.6 (+19.6%) | +$1.6B above | 20.0–24.0 | 27.5 | +$3.5B above |
- Revenue has beaten the HIGH end of guidance six quarters running (avg +$1.6B, range +$0.2B to +$3.7B) — the high end is the de facto base case. OI has cleared the high end in effectively 5 of 6, with the beat magnitude accelerating: +$2.4B (Q1'26) to +$3.5B (Q2'26).
- Handicapped Q3'26 landing: revenue ~$203–204B (+12.5–13% optical, ~+17% clean) and OI ~$28.5–29.5B (~14% margin), EPS ~$2.15–2.26 ex-any-Anthropic-marks. Street already prices the sandbag: FMP Q3 consensus is $203.3B — above the guide's $202B high end — with EPS ~$2.00. Beating guidance alone will not be enough in October; the incremental surprise has to come on OI.
- The real guidance event was capex: FY26 raised to ~$220B from ~$200B (ladder: ~$125B FY25 to ~$200B to ~$220B), attributed entirely to memory-cost inflation — a cost-push raise, not +10% more compute. Jassy: demand exceeds even $220B of supply in 2026 "and I believe this dynamic will also be true in 2027, too."
- FY26 consensus EPS ($9.44) is a broken benchmark: reported H1'26 EPS is already $8.53 (gain-inflated) and reported FY26 tracks ~$13 on Anthropic marks, while clean (ex-gain) FY26 EPS runs ~$8.2–8.5.
Upcoming catalysts
| Horizon | Catalyst | Status | Confirmation signal |
|---|---|---|---|
| Next print (2026-10-29) | Sixth straight quarter of AWS acceleration (bar: 36.7%) | UPDATED | Q3 AWS growth ≥ ~37%; sequential add vs the record +$4.6B; margin ≥ ~36% ex-items |
| $496B RPO conversion and organic bridge — QoQ jump largely the pre-announced $100B+ Anthropic deal folding in | UPDATED | Q3 10-Q RPO figure, recognition timing, ex-Anthropic growth; new mega-deals before 10/29 | |
| Q3 one-timers roll off: ~$1.2B of Q2 benefits (~$600M tariff refunds in NA, ~$600M energy-derivative gain in AWS) | NEW | NA margin ex-refund vs 7.4–7.9%; derivative-reversal risk in AWS | |
| Anthropic mark reprices every round — both directions; stake now large enough to swing the P&L quarterly | NEW | Any Anthropic funding round before 10/29; 10-Q carrying-value disclosure | |
| Amazon Leo initial service — language moved from "in a few months" (Q1) to "this year" (Q2), a soft slip | SLIPPING | Launch announcement + pricing; Q4 cost capitalization start | |
| 6–12 months | Off-cloud Trainium sales — "actively having those conversations" (was "over the next couple of years" in Q1) | ESCALATED | First named third-party chip/rack customer; how it is priced/reported |
| Trainium commitment number — $225B+ (Q1) was NOT updated on the Q2 call | NOT UPDATED | New commitment number on Q3/Q4 call; Trainium3 ramp; Trainium4 timing (~mid-2027) | |
| Advertising holding 26%+ — sponsored prompts (+48% conversion), sold-out NFL/NBA inventory, Ads Agent in 11 new countries | UPDATED | Q3 ad growth; agentic-ad metrics; DSP/off-Amazon disclosure | |
| Retail cost-to-serve: robotic-arm fleet to more than double in 2026; same-day perishables in 2,300 cities (MAUs +50% YTD) | UPDATED | NA margin ex-one-timers; robot deployment; Amazon Now unit economics | |
| ~$220B capex, 2027 frame, and funding mix ($67B new H1 debt; "nothing to share today" on capital sources) | UPDATED | 2027 capex number on Q3/Q4 call; further debt raises; memory-price commentary | |
| Multi-year | Own frontier model — "one of them will be ours" | NEW | re:Invent (early December) model announcement; training-cluster disclosures |
| AWS TAM "very possibly a trillion-dollar annual revenue business" | NEW | 30%+ growth sustained as run rate passes $200B (~2027); enterprise inference breadth | |
| Zoox paid robotaxi — NHTSA Part 555 exemption received, first purpose-built robotaxi cleared to charge for rides | NEW | First paid-ride city + pricing; Uber-app integration go-live |
Next confirmed catalyst: Q3 2026 earnings, 2026-10-29. Secondary windows: AWS re:Invent (early December — frontier model, Trainium4, Continuum GA), Leo commercial-launch announcement (by year-end), Anthropic funding news (episodic).
Street Q&A
| # | Analyst (Firm) | Asked | The real question | Answer quality |
|---|---|---|---|---|
| 1 | Doug Anmuth (J.P. Morgan) | Drivers/sustainability of the 39% AWS margin; does Amazon need its own frontier model? | Is the margin structural as AI mix scales? Is Amazon strategically exposed without a frontier model? | Direct + newsy — Olsavsky (his only answer): +650bps YoY, +520bps ex-derivative; Jassy: AI margins tracking "a little bit ahead" of core at the same stage, and "we are pursuing our own frontier model... one of them will be ours" |
| 2 | Justin Post (BofA) | Was the acceleration capacity-unlock driven? Gigawatts H2 vs H1; 2027 framing | Is 36.7% just a supply-gate function — making any capacity slip a revenue miss? | Partially evasive — demand-side tour (85% of IT still on-prem; AI pulls core via Graviton); no gigawatt split, only reaffirmed "double power capacity by end-2027 vs 2025" |
| 3 | Brian Nowak (Morgan Stanley) | Can long-lived data-center spend slow in 2027? Off-cloud Trainium timing and ROIC | When does the FCF bleed inflect? Is merchant Trainium real upside or a lower-return distraction? | Direct on (a) — "no": 2027 capacity "largely reserved," "quite a bit" of 2028 already reserved; new barbell framework. Half-answered on (b) — timing yes, ROIC comparison skipped |
| 4 | Colin Sebastian (Baird) | Application-layer push (Kiro, Transform, Quick); to Olsavsky: sources of capital for the build-out | With capex ~$220B and FCF negative — how much more debt? Equity? Off-balance-sheet? | Clearest dodge of the call — Jassy intercepted the CFO-directed question: "You've seen us issue debt this year. We have a lot of options... nothing to share today." App-layer answer was engaged and direct |
| 5 | Ken Gawrelski (Wells Fargo) | RPO ~2.5x Q3'25 — what does it force for 2028 capacity? Do long-term contracts protect returns from cost inflation? | Is $220B a floor? Are pre-inflation contracts locked at margins that memory costs will erode? | Deflected on 2028; candid on contracts — signed deals keep their prices for the duration; only new deals absorb memory inflation. The one answer that cut against the bull case |
| 6 | Eric Sheridan (Goldman Sachs) | Consumer signal on fast commerce / grocery / essentials; geographic differences | Is the retail flywheel (speed → frequency → essentials wallet share) actually accelerating, and does it travel? | Direct on adoption (perishables in 2,300 cities; MAUs +50% YTD; 3x units per same-day order), non-responsive on geography — US only |
- Structural tell: Jassy took nearly every question — Olsavsky answered exactly one all night, and the CEO intercepted the capital-structure question explicitly addressed to the CFO.
- Top 3 themes: (1) returns on the AI build — margin sustainability and ROIC (Anmuth, Nowak, Gawrelski); (2) capex trajectory, capacity and funding (Post, Nowak, Gawrelski, Sebastian — four of six analysts, nobody got a number beyond "2x power by YE2027"); (3) full-stack ambitions — chips, models, apps (Anmuth, Nowak, Sebastian). The street graduated from the demand/product story to underwriting capacity, funding, and returns.
- The dodge: sources of capital — "nothing to share today," at the exact moment the capex guide was raised $20B, TTM FCF runs negative, and long-term debt nearly doubled in six months.
- Most newsworthy: the own-frontier-model commitment (a directly comparable Q1 question was deflected); runner-up: 2028 capacity already partially reserved; honorable mention (bear side): fixed-price contracts do not reprice for memory inflation.
Buy-side takeaway: the company declared the demand debate over and invited a returns debate it believes it wins. Load-bearing new claims: AI margins tracking "a little ahead" of core cloud at the same stage; 39.4% AWS margins (+520bps clean); 2027 capacity largely reserved and 2028 partially reserved against a $496B backlog; and a capital-cycle framework (sub-3-year server breakevens against 5-year contracted terms) that reframes $220B of capex as pre-sold inventory rather than speculative build. The residual bear file is exactly what management declined to answer: funding ("nothing to share today") while TTM FCF runs negative, a $20B mid-year inflation bump to the capex guide, signed contracts that cannot reprice for memory costs (component inflation lands on backlog margins, not customers), and Jassy's own flag that the middle of the demand barbell — enterprise production inference, the segment the trillion-dollar AWS thesis rests on — may not ramp at the labs' "wildly steep trajectory." Net: the street left underwriting execution and funding risk, no longer demand risk — a materially better place than 90 days ago.
Contradictions
| # | Item | Then | Now | Materiality | Read |
|---|---|---|---|---|---|
| 1 | Capex guidance creep | ~$105B implied (Feb'25) → ~$125B (Oct'25) → ~$200B (Feb'26) → "I don't have an update" (Apr'26), with memory inflation already flagged as managed | ~$220B (Jul'26), the +$20B attributed to "the higher cost of memory" — the same, already-flagged inflation; H1 run rate was tracking above $200B before the guide moved | High | Serial under-guiding, externally attributed each time. Treat $220B as a floor; "this dynamic will also be true in 2027" pre-frames another raise |
| 2 | Server useful-life flip | Q4'24: cut from 6 to 5 years because AI accelerates obsolescence; $920M early-retirement charge | Q2'26: "at least five to six years" plus a "strong track record of... extending the useful life" | High | Both directions argued within six quarters, each flattering the contemporaneous story. A life extension (margin-flattering) is now pre-telegraphed — watch for it |
| 3 | Anthropic-gain disclosure standard | $3.3B (Q1'25) and $9.5B (Q3'25) gains proactively caveated on-call: "not related to Amazon's ongoing operations" | $53.4B gain got zero call airtime — no mention by either executive; PR headline leads with "$62.6B net income... $5.75 per diluted share" | High (optics) | The caveat was applied when small and abandoned the one time it had real headline consequences. Circularity: the record "profit" is a mark on AMZN's own largest AI customer |
| 4 | Prime Day adjustment — one-sided | Q3'26 drag quantified with care: growth "would have been nearly 400 basis points higher" ex-Prime-Day | The same shift boosted Q2'26 actuals (+20% headline, retail lines +300–400bps) — acknowledged but never quantified | High | Legitimate math, asymmetric use. Related: the Q3 OI guide is compared against the unadjusted $17.4B base, not the $21.7B ex-charge base management itself insisted on a year ago |
| 5 | Fixed-price contracts vs memory inflation | Prepared remarks: "clear line of sight to strong financial returns," servers break even in under 3 years | Q&A concession: signed deals keep their prices "over the duration of that contract"; only new deals price in higher costs | High | The mechanism by which the capex raise becomes a returns problem: contracted backlog margin is lower today than when signed. How much of the $496B RPO was priced pre-inflation was never quantified |
| 6 | "Operating cash flow +33%" framing | FY24Q4 headlines led with adjusted FCF ($36.2B) | PR leads with OCF +33% to $161.4B while TTM FCF sits at −$7.6B (six straight declines); OCF itself flattered by lower cash taxes and payables; ~$29B of capex sits in "acquired but not yet paid" | Medium-High | Metric selection; the cash story deteriorates further in H2 at a ~$60B/qtr capex run rate before it inflects |
Verdict: trust the numbers, discount the framing. Every arithmetically checkable claim in the print ties out against the frozen data (run rate, record QoQ add, acceleration streak, margin bps, RPO multiples), guidance has been systematically conservative, and the 2024–25 claim "we would grow faster if unconstrained" was later verified by the acceleration itself. But the framing pattern is consistently procyclical: adjustments and caveats are deployed when they excuse and dropped when they would deflate. Watch items for Q3: AWS margin ex-derivatives as fixed-price backlog meets inflated components; any server useful-life extension; whether ~$220B survives October; an organic (ex-Anthropic) RPO bridge.
Indirect read-throughs
| Sector | Tickers | Direction | The AMZN datapoint | Confidence |
|---|---|---|---|---|
| Memory / storage | MU, SK Hynix, Samsung, SNDK, STX/WDC | POSITIVE | The $20B capex raise attributed entirely to memory prices; "memory and hard drives and SSDs" named as inflated; the largest buyer paying up rather than cutting volume, with demand visible into 2028 — the cleanest supplier-pricing-power confirmation a customer ever gives | High — strongest read in the print |
| AI-capex complex | NVDA, VRT, ETN, GEV, CEG/VST/TLN | POSITIVE | ~$220B capex, supply-constrained through 2027, 2028 demand "striking"; power capacity doubling by end-2027 with 2-year shell lead times; ~$600M forward-power derivative gain evidences tightening long-dated power markets. NVDA caveat: demand floor affirmed ("best place to run Nvidia chips") but the marginal training dollar is going to Trainium, and merchant Trainium sales were floated | High (demand); Medium (NVDA mix) |
| Custom silicon | AVGO, MRVL | POSITIVE | Chips business $25B+ run rate growing triple digits; Anthropic and OpenAI multi-gigawatt Trainium commitments; "custom network gear" favors merchant switch silicon | Medium-High |
| Cloud peers | MSFT, GOOGL, ORCL | Demand-positive / share-negative | A $169B-run-rate business accelerating to 37% means the pool is expanding faster than anyone can build (85% of IT still on-prem); 39.4% AWS margin directly rebuts "AI dilutes cloud margins." But AWS is no longer the AI laggard — sentiment offset for MSFT's cloud premium; ORCL is the pure overflow beneficiary | High (demand) / Medium (share) |
| Anthropic holders | GOOGL | POSITIVE | The $53.4B mark implies a major Anthropic valuation step-up — the same logic marks up GOOGL's Anthropic stake; a concrete, dateable non-operating gain in a coming GOOGL print | High |
| Ad tech | TTD | NEGATIVE | Ads +26% at a ~$79B run rate; owned CTV inventory (TNF, NBA, WNBA, NASCAR) sold out; DSP scaling plus agentic sponsored prompts converting 48% better — retail media + closed-loop CTV squeezing the open internet's share | Medium-High |
| Parcel & grocery | UPS, FDX, KR | NEGATIVE | Amazon Supply Chain Services launched with P&G and 3M as anchors — selling Amazon's network to any business, a direct 3PL/integrator incursion; #2 US grocer with same-day perishables in 2,300 cities, perishables MAUs +50% YTD | Medium-High |
| Consumer / payments | V, MA | POSITIVE | Paid units +17% (accelerating), Prime membership growing double digits, tariffs largely absorbed with no demand damage — one of the cleaner "consumer is fine" datapoints of the season; agentic checkout scaling on existing card rails with +21–40% basket uplift | Medium-High |
Circularity risk (applies to every positive read above): the print deepens the AI funding loop — AMZN marks a $53.4B gain on Anthropic while Anthropic (and OpenAI) commit multi-gigawatts to Amazon's chips, and AMZN issues $67B of H1 debt with TTM FCF at −$7.6B to build the capacity. Every Tier-1 read-through is levered to this loop continuing to be funded.
Sources: SEC 8-K Ex-99.1 press release (Q2 2026, incl. supplemental tables), FY2026Q2 earnings-call transcript (2026-07-30) and six prior transcripts, FMP consensus and surprise history, Daloopa-cited historical fundamentals (company_id 15). Analysis: Claude (claude-fable-5). Data sourced from Daloopa.