Retail Print Is Excellent, AWS Reacceleration Still Three to Four Quarters Out — Capex Plateau at $31B/Quarter Means FCF Compression Now Has Visibility

Results vs. Consensus

Quality of Beat/Miss

• Top-line beat is broad and clean: revenue +12% ex-FX vs. ~+9% Street; +13% reported aided by an unexpected $1.5B FX tailwind (vs. guide assumption of $100M FX headwind). Even normalizing for FX, the beat is real.
• Op-income beat of $1.7B vs. high end of guide is high quality — driven by NA margin (+190bps YoY) and International margin (+320bps YoY). No one-time items distorting the print.
• AWS top-line in-line is the cleanest message: management told us in Q1 the business was capacity-bound; Q2 prints +17.5% with backlog accelerating to +25%. The supply-side framing is corroborated.
• AWS margin compression is the soft point: 32.9% is well below Q1's 39.5% peak. Olsavsky's decomposition (~half SBC step-up, remainder depreciation + FX) is credible — the SBC half is mechanical and reverses, the depreciation half is structural and persists with capex. Underlying margin probably stabilizes 33–35% near term.
• Advertising +22% on $15.7B is best-in-class breadth: Roku (80M CTV households exclusive), Disney/ESPN/Hulu DSP integration, 300M+ U.S. ad-supported audience. This line is no longer "emerging" — it is materially additive to NA / International segment economics.

Segment Performance

AWS — Capacity-bound, backlog accelerating, margin re-baselining

AWS revenue of $30.87B was up 17.5% YoY — essentially flat with Q1's +17% pace. The story behind the number is twofold. First, backlog (RPO) accelerated from +20% YoY in Q1 to +25% YoY in Q2 ($189B → $195B), with the same 4.1-year duration. That is the cleanest leading indicator we have, and it is unambiguously bullish on demand. Second, Jassy was significantly more direct about supply-side constraints than in Q1: "the single biggest constraint is power," followed by chips and components. Critically, he reframed the timing — "I don't believe that we will have fully resolved the amount of capacity we need for the amount of demand that we have in a couple of quarters. I think it will take several quarters."

"We have more demand than we have capacity right now. So we could be doing more revenue and helping customers more, and we're working very hard on changing that outcome and how much capacity we have." — Andy Jassy, CEO

That is a meaningful pull-back from Q1's "capacity will get better as the year proceeds." It pushes the AWS reacceleration narrative from "back half of 2025" to "early-to-mid 2026" in our reading. AWS operating margin compressed from 39.5% to 32.9% — Olsavsky decomposed: roughly half is the seasonal Q2 SBC step-up (mechanical, reversing), the remainder is higher depreciation from capacity landings plus unfavorable FX YoY. We model ~33–35% as the new base while capex absorbs through D&A.

On the customer side, the wins list is enterprise-direct and broad: PepsiCo, Airbnb, Peloton, Nasdaq, LSE, Nissan, GitLab, SAP, Warner Bros Discovery, FICO, NatWest. Anthropic is now training its newest Claude generation on Trainium2 — this is the single most consequential AI partnership AWS has, and the chip is now positioned not just as a price/performance play but as the explicit backbone of a frontier-model lab. Bedrock added Claude 4 ("fastest-growing model ever in Bedrock"). Amazon Nova is the second most popular foundation model in Bedrock. AgentCore (agent runtime), Strands (open-source agent build), AWS Transform (mainframe / VMware / .NET migration agent), and Kiro (agentic IDE — 100K developers in first 5 days of preview) all launched in Q2.

Assessment: The AI stack story is the strongest it has been. The supply-side timing is the worst it has been. Both can be true; both matter. Backlog growth retires the demand-bear case; capacity timing prolongs the AWS-decel optic.

North America — Margin engine compounds

NA revenue $100.1B (+11%) crossed the $100B-quarter threshold for the first time, with operating income of $7.5B at 7.5% margin (+190bps YoY). The drivers are operationally specific: direct-lane share +40% YoY, average package distance −12%, handling touches per unit −15%, outbound shipping cost +6% on units +12%, and 30% more items same-day or next-day in the U.S. Prime Day was the biggest ever — record sales, record items sold, record Prime sign-ups in the 3-week leadup. 1 millionth robot deployed globally. DeepFleet AI improving robot travel efficiency by 10% (compounding at scale).

Selection adds in Q2 are notable: Nike returning to Amazon retail, Away, Aveda, Marc Jacobs Fragrances, Saks luxury portfolio (Dolce & Gabbana, Etro, Stella McCartney, La Prairie). Perishables Pilot expansion: 75% of viewers are first-time perishables shoppers; 20% return within first month. Same-day / next-day expansion to 4,000+ smaller cities and rural areas by year-end (1,000+ already live).

Assessment: NA is in a structural margin-expansion regime. The combination of inbound network productivity + advertising leverage + 3P unit-mix accretion (62%, record) is a 3-engine flywheel. This is the part of the print that, on its own, would justify a higher rating; the AWS / capex offsets are what hold us at Hold.

International — 10 straight quarters of margin progress

International revenue $36.8B (+11% ex-FX) at 4.1% operating margin (+320bps YoY). Olsavsky framed it as "10 quarters of progressive margin gains, ~700bps cumulative." The split: established countries (UK, Germany, Japan) at margin profiles approaching the U.S. and contributing more profit as they grow; emerging countries (8 launched in last 5 years) at varying stages of upfront investment but all showing quarter-over-quarter improvement.

Assessment: This trajectory is the longest-running and most under-appreciated structural story at Amazon. International earned a 7-segment in our view of the model.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Confident on retail and advertising, more granular and notably more honest on AWS supply-side timing, and increasingly enthusiastic about Trainium2 as a structural advantage rather than a price/performance bet. Jassy spent more airtime on Kiro and AgentCore than on AWS revenue — a signal that the AI-products narrative is the new center of gravity in the AWS story. Olsavsky's Q3 op-income range ($5B wide, vs. Q2's $4.5B) is the second consecutive widening of the guide — a measured acknowledgment that visibility on tariffs and AWS timing remains low.

AWS Capacity Timing — From "Back Half" to "Several Quarters"

The most important call-shift between Q1 and Q2: in Q1 Jassy said "supply chain issues and the capacity issues will continue to get better as the year proceeds." In Q2, he said it will take "several quarters" to fully match capacity to demand. That is not a downgrade of demand — backlog grew $6B sequentially and accelerated 5pp to +25% YoY. It is a downgrade of how quickly the supply-side bottleneck (now explicitly identified as power, then chips, then components) can be relieved.

"I don't believe that we will have fully resolved the amount of capacity we need for the amount of demand that we have in a couple of quarters. I think it will take several quarters. But I do expect that it's going to get better each quarter, and I'm optimistic about that." — Andy Jassy, CEO

Assessment: This pushes the AWS reacceleration print into Q4 2025 / Q1 2026 in our base case. Backlog accumulation in the meantime is stored revenue — every $10B of backlog growth is ~3pp of forward-revenue tailwind once capacity lands. The delay is timing, not impairment.

Capex Plateau at $31B/Quarter — H2 "Reasonably Representative"

Olsavsky: "We expect Q2 CapEx to be reasonably representative of our quarterly capital investment rate for the back half of this year." That implies $120B+ FY25 cash capex — well above the $100–105B baseline carried by most models pre-print. AWS is the primary driver (chips, data centers, power), with custom silicon (Trainium) ramping aggressively. Fulfillment / transport investment continues alongside.

Assessment: The capex curve is now visible. FCF compression continues — TTM FCF $18.2B, down $7.7B sequentially from Q1. The investment is rational if AWS revenue ramps with capacity in 2026, but the model now needs to absorb 6–9 months of capex-led FCF compression before that happens.

The "Output Gap" — Hyperscaler Peer Comparison

Doug Anmuth's question on AWS growing slower than peers got the most direct response of the call. Jassy's framing: (a) base-rate effect — "the second player is about 65% of the size of AWS"; (b) it's a moment-in-time comparison with quarters where each leads; (c) what really matters to enterprise customers is operational performance, security, functional breadth, and end-to-end stack — and on those, Jassy claims AWS leads. The implicit acknowledgment: in any given quarter, Azure or GCP may print higher growth, and Amazon is willing to live with the optic so long as the structural advantages compound.

"We have a meaningfully larger business in the AWS segment than others. The second player is about 65% of the size of AWS… these are all really just moments in time." — Andy Jassy, CEO

Assessment: Defensible answer; the bear narrative on AWS losing AI share won't get retired by an answer, only by a print. The +17.5% / +25% backlog combination is the most credible defense. We are still in wait-and-see mode on this.

Trainium2 as Anthropic's Backbone — A Structural Advantage

The single most important new disclosure: Trainium2 is "the backbone for Anthropic's newest generation Claude models" and "the backbone of Bedrock and the inference that we do." This is a step-function from Q1's "Anthropic is training on Trainium2." Jassy framed it as a structural NVIDIA-vs-custom-silicon parallel to Graviton vs. Intel: "we saw this movie in the CPU space with Intel, where customers anchor for better price performance." Trainium2 claimed at 30–40% better price/performance than competing GPU instances; Trainium3 already in development. With Anthropic anchored as the lab customer and Bedrock as the inference fabric, the chip story is increasingly self-reinforcing.

Assessment: The Anthropic / Trainium2 / Bedrock combination is the cleanest moat narrative in the AWS story. If Trainium3 lands on schedule and Anthropic's Claude generations continue to anchor on it, AWS's AI economics improve materially over the next 18–24 months — independent of any peer-growth-rate optic.

Advertising — From Big to Structural

Q2 advertising revenue $15.7B (+22%) accelerated from Q1's +19%. Roku DSP partnership (80M authenticated CTV households, exclusive) and Disney real-time ad exchange / Amazon DSP integration (Disney+, ESPN, Hulu) are step-function moats. 300M+ U.S. ad-supported audience (vs. 275M in Q1).

"When advertisers work with Amazon, they're not just buying ad space — they're benefiting from exceptional programming, innovative technology and unrivaled signals, measurement and audience development." — Andy Jassy, CEO

Assessment: The Roku and Disney DSP deals are competitive moats vs. Google and Meta in CTV. Advertising is now a $60B+ annual run rate compounding at +20%+ — if rate of growth holds, it crosses $80B by 2027.

Tariffs — Still No Signal in the Numbers

Jassy: "Through the first half of the year, we just haven't seen diminished demand. And we haven't seen any kind of broad-scale ASP increases. And so that could change in the second half." 1H consumer signal benign. The 2M+ marketplace seller diversity argument repeats. Q3 op-income range widened explicitly to absorb tariff scenario uncertainty.

Assessment: Two consecutive quarters of "no demand attenuation, no ASP move" is a meaningful real-time data point for the broader retail tariff debate. Q3 will be the first quarter where post-summer-tariff-finalization data shows up; we will know more when Q3 prints.

Project Kuiper — Commercial Service "Later This Year or Early Next"

Jassy reframed Kuiper around 400–500M un-broadbanded households globally, plus enterprise / government customer demand. Three successful production-satellite launches done. "Some delays with rocket providers" but most of the available rocket cadence over the next couple years secured. Commercial service "later this year or early next year." Critically, he positioned Kuiper as the natural pair to AWS for enterprise / government data flows.

Assessment: Optionality. The AWS-pairing argument is real and differentiating. Still immaterial to FY25/26 financials.

Guidance & Outlook

The Q3 op-income range of $15.5B–$20.5B is unusually wide ($5B span; Q2's was $4.5B). Olsavsky was explicit: "uncertainty on where tariffs will settle and maybe the ultimate impact on consumers." But: revenue guide is above Street, with both NA / International momentum and a $1.5B+ FX tailwind. The shape suggests management has high confidence in revenue and lower confidence in how to model the tariff-cost flow-through to margin.

Implied Q3 op-income trajectory: midpoint $18.0B is +25% YoY (Q3 2024 was $14.4B). High end $20.5B is +44%. Low end $15.5B is +9%. Wide bias toward the high end of the range given Q2 over-delivery.

Street at: ~$19.4B pre-print; will recalibrate to ~$18–18.5B post-guide.

Guidance style: Conservative-with-flexibility. Amazon historically beats the high end of its op-income range; the wider-than-usual band is the company asking for room to absorb uncertain tariff cost.

Analyst Q&A Highlights

AWS Output Gap and Reacceleration Timing

• Doug Anmuth, JPMorgan: Asked whether AWS's slower growth vs. Azure/GCP is demand or supply. Jassy: "We have more demand than we have capacity right now." Reframed peer comparison as moments-in-time and emphasized base-rate effect ("second player is about 65% of the size of AWS").
  Assessment: Direct supply-side answer. The base-rate point is mathematically valid but not what bears want to hear.
• Mark Mahaney, Evercore: Asked whether supply constraints will resolve in H2 as previously suggested. Jassy: "I don't believe we will have fully resolved capacity in a couple of quarters. I think it will take several quarters." Assessment: This is the timing pull-back. Most important downgrade in management's narrative this quarter.
• Brian Nowak, Morgan Stanley: Asked directly about the "AWS falling behind in GenAI" narrative and whether AWS should accelerate H2 / 2026. Jassy gave a 5-minute layered answer (top of stack, middle of stack, bottom of stack) — strong on Trainium2, AgentCore, Strands, Kiro — but declined to commit to acceleration timing: "we don't give guidance by segment." Assessment: The most thorough articulation of the AI strategy yet, but no commitment on timing. Backlog continues to be the primary leading-indicator answer.

Tariff Pass-Through

• Doug Anmuth, JPMorgan: Asked about tariff absorption across suppliers, Amazon, and consumers. Jassy: 1H showed no demand attenuation and no broad ASP moves; 2H is uncertain. Three layers of absorption (suppliers via 1P forward-buy, Amazon via inventory cost, consumers via potential ASP) — distribution still being determined.
  Assessment: Honest answer. The lack of 1H signal is itself a data point.

International Margin Sustainability

• Colin Sebastian, Baird: Asked about sustainability of International margin progression. Olsavsky: 10 quarters of progressive gains, ~700bps cumulative, two-tier mix (UK/Germany/Japan approaching U.S. profile; 8 emerging countries with varying upfront investment maturities).
  Assessment: Most concrete framework yet for the International margin story. Very bullish.

Project Kuiper Status

• Colin Sebastian, Baird: Asked about Kuiper service rollout timing and long-term ambitions. Jassy: 3 production launches done; commercial later this year or early next; some rocket-provider delays; AWS-pairing as differentiation; impressive enterprise/government commitments already.
  Assessment: Slight slippage on rocket cadence. AWS-pairing argument is structurally important; we will track.

Alexa+ Monetization Pathways

• Mark Mahaney, Evercore: Asked about how Alexa+ might play into engagement / retail / ad / subscription. Jassy: rolled out to "millions of customers"; very high ratings, expansive usage, more calls than prior Alexa. Future levers include shopping, advertising in multi-turn dialogs, expanded subscription beyond Prime.
  Assessment: Most explicit articulation of Alexa+ monetization thinking. Still optionality, not yet a model line item.

Q3 Guide Composition

• Justin Post, Bank of America: Asked about Q3 revenue drivers and tariff contingencies, plus Q4 framing. Olsavsky: cautiously optimistic on Q3 — Prime Day strength, delivery speed, selection, in-stock all working; tariff uncertainty drives the wide op-income range. No Q4 guide given.
  Assessment: Confirms revenue confidence; margin uncertainty drives the range width.

What They're NOT Saying

1. FY25 capex number explicit: Olsavsky implied $120B+ via the "Q2 reasonably representative for H2" framing but did not commit to a specific full-year number. Conspicuous omission given the magnitude.
2. AWS reacceleration quarter: Brian Nowak asked directly. Jassy declined to commit. Pattern from Q1 continues.
3. Trainium2 revenue mix in AWS AI run-rate: Anthropic anchor disclosed; relative size of Trainium2 vs. NVIDIA inference still opaque.
4. Q4 guide framing: Olsavsky declined to give Q4 color despite Justin Post asking. Holiday-quarter visibility appears low.
5. Amazon Pharmacy revenue base: "+50% YoY YTD on already-significant base" — but no absolute number. Implied $4–6B run rate.
6. FTC antitrust trial timeline: Still no mention. Trial proceedings continue in background; no surface in earnings.
7. Anthropic equity stake updated valuation: Q1 included $3.3B mark-up; Q2 silent. Standard accounting cycle but the silence is noted.

Market Reaction

• After-hours move (July 31): Initially positive on the $19.2B op-income beat (+$1.7B vs. high end), then turned modestly negative as the call progressed. The call-driven swing reflects (a) AWS reacceleration timing pull-back and (b) wide Q3 op-income range. Estimated final after-hours print: ~−1% to −3%.
• Next-day (Aug 1) early indications: Stock weaker pre-market; Street recalibrates Q3 op-income to ~$18.0–$18.5B midpoint and pushes AWS reacceleration into Q4/Q1 2026.
• Volume: Elevated; earnings sessions consistently print high volume.
• Analyst reactions (within first 12 hours):
  • Most desks maintained ratings; price targets nudged in a tighter band reflecting offsetting beats (retail) and timing slips (AWS)
  • Bull camp emphasized $195B backlog (+25%), Anthropic / Trainium2 anchor, retail margin expansion, Roku / Disney DSP
  • Bear camp emphasized AWS margin compression (39.5% → 32.9%), capex run rate at $31B/quarter, AWS reacceleration timing slip

The reaction is a clean read on the print: a strong retail / ads quarter is being penalized for the AWS timing pull-back and capex visibility step-up. The market wants a Q3 or Q4 print where AWS reaccelerates and capex stops compounding before re-rating up.

Street Perspective

Debate: AWS Reacceleration — When?

Bull view: Backlog +25% YoY at $195B is the highest growth rate in over a year. Trainium2 anchoring Anthropic + Bedrock inference. Capacity is landing each quarter, even if total resolution takes "several quarters." A late-Q4 or early-Q1 2026 reacceleration print is the path of least resistance.

Bear view: Two quarters of "supply-side capacity-bound" framing without acceleration. Azure and GCP continue printing higher growth. The risk that "several quarters" becomes "many quarters" is real, especially with power as the binding constraint.

Our take: Backlog accumulation is the highest-quality data point and supports the bull. Power as the named bottleneck is harder to relieve than chips — that argues for the bear's "several quarters might be longer" worry. Our base case: Q4 2025 prints +18–19%, Q1 2026 prints +20%. If that lands, the bull thesis is correct.

Debate: Capex — Investment or Excess?

Bull view: $31B/quarter capex is matched by $30B+ of incremental backlog accumulation per quarter. AWS LTV per dollar of capacity is the highest in the company. This is rational growth investment, not excess.

Bear view: TTM FCF compressed from $25.9B (Q1) to $18.2B (Q2). At $120B+ FY25 capex pace, FY25 FCF will land closer to $25–35B vs. consensus $50–60B. Multi-year FCF compression at this magnitude historically gets re-rated by the market.

Our take: Right framework remains AWS-revenue-per-AWS-capex. Q2's backlog growth supports the bull view operationally. Whether the multiple holds while FCF compresses is a separate question — one we expect to see tested in H2.

Debate: Retail Margin — Sustainable or Cyclical?

Bull view: NA at 7.5% (+190bps), International at 4.1% (+320bps, 10-quarter streak). Inbound network productivity, robotics (1M+ deployed, DeepFleet AI +10% efficiency), 3P unit mix at record 62%, ad leverage. Multiple compounding engines.

Bear view: H2 will see incremental tariff cost pressure (1H showed no signal but that's the lagged data point). Holiday-period operational complexity always pressures margin. NA could give back 50–100bps in Q3/Q4.

Our take: The structural drivers (network productivity, 3P mix, ad leverage) compound. Cyclical pressure (tariffs, holiday) is real but second-order. We expect NA margin to hold above 6% for the year.

Model Update Needed

Valuation impact: Net target moves modestly higher despite FCF compression — the retail / ads upgrade plus AWS-LTV-per-capex framework offsets. Still not enough to re-rate to Outperform without an AWS reacceleration print.

Thesis Scorecard Post-Earnings

Overall: Thesis modestly strengthened on retail / ads / Trainium2-Anthropic axis; modestly weakened on AWS reacceleration timing and capex visibility. Net unchanged at Hold pending AWS print.

Action: Hold maintained. Initiation rating from Q1 unchanged. Re-rate higher on a single print where AWS growth ticks above 18% with capex showing first signs of plateau.