---
title: "AMZN (Trans): Capex raised to $220bn"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/topics/43096142.md"
description: "Below is Dolphin Research's FY26Q2 earnings call Trans for $Amazon(AMZN.US). Key highlights from the release are as follows.1) Q3 guidance: net sales of $197.0bn–$202.0bn. Operating income of $22.5bn–$26.5bn.The sequential slowdown in revenue growth from Q2 to Q3 is mainly due to two factors. First, Prime Day timing: in 2026 the event lands in Q2 across most major countries, including the U.S., whereas in 2025 it fell entirely in Q3..."
datetime: "2026-07-30T22:39:34.000Z"
locales:
  - [en](https://longbridge.com/en/topics/43096142.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/43096142.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/43096142.md)
author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)"
---

# AMZN (Trans): Capex raised to $220bn

**Compiled by Dolphin Research:** $Amazon(AMZN.US) **FY26Q2 Earnings Call Trans**

**I. Key Takeaways**

1\. **Q3 Guidance**: **Net sales of $197.0bn–$202.0bn; OP of $22.5bn–$26.5bn.** The sequential deceleration from Q2 to Q3 mainly reflects Prime Day timing and FX. Prime Day fell in Q2 across most major markets in 2026, including the U.S., while in 2025 it was entirely in Q3; excluding Prime Day timing for both years, Q3 YoY growth would be nearly 400bps higher. At current FX, Q3 faces an ~80bps headwind.

2\. **Capex Raised**: **FY26 cash capex is now Approx. $220.0bn vs. prior ~$200.0bn, primarily on higher memory costs.** Even at this level, FY26 will not meet all demand, with FY27 similarly tight and FY28 demand already substantial. Q2 cash capex was $53.1bn, largely for AWS and Gen AI.

3\. **Q2 Financials**: Revenue of $200.6bn, +20% YoY ex-FX (including Prime Day shifting into Q2). OP of $27.5bn, +43% YoY. OP includes two net credits totaling ~$1.2bn: ~$0.6bn tariff-related refunds in North America (most of expected refunds) and ~$0.6bn FV gains under hedge accounting on energy contracts, mainly impacting AWS.

4\. **Segment Profitability**: North America revenue $116.2bn (+16% YoY); OP $9.1bn; OPM 7.9%. Intl revenue $42.2bn; +15% YoY ex-FX; OP $1.7bn; OPM 4.1%. AWS revenue $42.2bn (+36.7% YoY); OP $16.6bn; OPM 39%, +650bps YoY, or +520bps ex-derivative gains. Global paid units were up 17% YoY.

5\. **Accounting Change**: Certain power purchase agreements now apply hedge accounting due to specific terms, with FV changes recognized each period in operating expenses as unrealized gains/losses. Adjustments were immaterial in prior quarters, but may be more volatile ahead. These contracts secure power for current and future operations.

6\. **Financing**: The company issued bonds earlier this year and retains multiple funding options to support AWS build-out. It will continue to evaluate and act at appropriate times. No new arrangements disclosed on this call.

**II. Call Details**

**2.1 Management Highlights**

**1\. AWS: Five consecutive quarters of acceleration; AI and core both driving**

a. **Q2 AWS revenue rose 36.7% YoY, marking the fifth straight acceleration and the fastest growth in 18 quarters**—when AWS revenue was less than half of today. Sequential revenue added over $4.6bn, ~1.8x the largest prior single-quarter increase.

b. **Backlog reached $496.0bn, up triple digits YoY; AWS ARR is now $169.0bn**, which would rank ~24th in the Fortune 500 if standalone.

c. Chip business ARR is now above $25.0bn, up triple digits YoY; AI business ARR also climbed sharply sequentially to above $25.0bn, up triple digits YoY.

d. Customers choose AWS for breadth of capabilities, proximity of AI inference to applications and data (where AWS holds the largest installed base), and leading security and operational performance.

e. AI and core are reinforcing each other: RL post-training and agent tool calls largely run on CPUs rather than AI accelerators, and Graviton delivers 30–40% better price-performance vs. alternatives. AI data storage and vector DB needs also benefit from AWS's broad, comprehensive core infrastructure features.

f. Enterprise cloud migration: ~85% of global IT spend remains on-prem and is set to reverse over the next 10–20 years. **More enterprises are planning and executing migrations, and AWS is capturing the majority of this shift.**

**2\. AWS AI Stack: from models to agents to apps**

a. Model layer: 18 months ago, management anticipated capable enterprises would build their own foundation models—not ultra-frontier giants, but smaller models leveraging proprietary data—with SageMaker AI as the most convenient service. Inference is served by Amazon Bedrock, offering best-in-class model choice, performance, and enterprise-grade governance and security, with rapid growth.

b. Agent infrastructure: Production-grade agents need secure runtime, memory for context, identity to act on a user's behalf, connection to tools/data, and observability at scale—hard to assemble reliably. Bedrock AgentCore provides managed, out-of-the-box infrastructure, recently adding policies (deterministic control of agent behavior), payments (agents executing transactions), web search (augment external knowledge within AWS), and new harnesses (with Strands to speed agent creation).

c. Coding agents: Successful products include Claude Code, Codex, and the company's spec-driven Kiro. **Kiro delivers up to 50% better price-performance than peers, with usage tripling QoQ.**

d. Amazon Quick: An intelligent AI work assistant to search, manage, and automate digital work across email, calendar, local/cloud files, and custom workflows. It spans Slack, Salesforce, Jira, Teams, ServiceNow and other major SaaS, preserves existing enterprise access controls, and can schedule meetings, draft/send emails, update CRM records, and build dashboards. Q2 adds autonomous agents (background multi-step tasks via natural language), a personalized activity stream aggregating email/messages/calendar/tasks, and 16 new integrations (incl. Adobe, Moody's, Snowflake). Customers include 3M, Allianz, AstraZeneca, Autodesk, BMW, Exxon, FINRA, Hyundai, Intuit, Mondelez, Moody's, NBA, NFL, Life Insurance and Southwest Airlines.

e. Other fast-growing services: Amazon Connect for contact centers, and AWS Transform for automated software migration. New AWS Continuum targets code vulnerabilities end-to-end—ingesting existing backlogs, scanning with frontier models, and using agents with enterprise context to judge 'deployment status, reachability, production path relevance, and impact if exploited.' It validates in sandbox to avoid false positives and provides remediation guidance; with frontier model progress making technical security issues easier to detect, management expects rapid adoption.

**3\. In-house chips: Trainium and Graviton**

a. Beyond Anthropic and OpenAI committing multi-GW over multiple years to Trainium, more AI startups are adopting it, including unicorns like Neurobotics and Odyssey, and firms such as Twelve Labs, Descarte, Poolside, Karakuri, Neto AI, and Splash Music, as well as larger players like Uber and Pinterest.

b. Graviton is used by 98% of the top 1,000 EC2 customers; revenue commitments nearly tripled QoQ; Graviton 5 is growing at ~2x the pace of Graviton 4.

c. **Deep collaboration with NVIDIA continues, making AWS the best platform to run NVIDIA chips—customers will choose NVIDIA for the foreseeable future**, and management believes choice benefits customers, competition, and helps lower inference costs that customers care about most.

**4\. Capex return framework**

a. Two buckets with different capital cycles: Data center capex is committed ~2 years before servers are installed. Once a data center is live and servers are powered on, revenue ramps immediately and monetization lasts over 30 years without re-spending the initial build.

b. Servers and networking have shorter cycles, typically procured months before deployment, providing clear visibility to demand before spend—no demand, no spend. Servers and networking typically break even in under three years; **current server useful lives are at least five to six years, and most recent AI capacity contracts are 5+ years, implying significant FCF in the two to three years post breakeven.** AWS has a strong record of early payback and has made real progress extending useful lives without sacrificing customer experience.

Near term, simultaneous DC builds before monetization drive elevated capex and FCF pressure; as revenue growth ultimately outpaces incremental capex growth, the crossover yields highly attractive revenue, FCF, and ROIC.

d. Cloud's first era followed a similar path, though demand accumulation was more gradual and longer than AI. AI margins and returns today track core at a comparable stage, slightly ahead in fact.

e. Long-term opportunity: **Management previously saw AWS scaling to several hundred billions of revenue; now it sees at least double that, with a real chance to reach $1 trillion in annual revenue**, alongside very attractive FCF and ROIC. Enterprise production inference at scale remains very early.

**5\. Retail & Selection: breadth, low prices, and fresh**

a. Millions of new items added, including 700k+ from well-known brands. U.S. ultra-low-price channel Amazon Haul expanded selection ~20x since launch, now with 6mn+ items priced under $10. Everyday prices match or beat competitors, with deep discounts in promotions.

b. Prime Day feedback was strong; over 80% of the millions of deals hit the lowest prices of the year, with hundreds of thousands at 40%+ off.

c. Fresh and everyday essentials broke through: The company is the No. 2 grocer in the U.S., with groceries growing rapidly across fresh and non-fresh. Monthly active users in fresh grew over 50% since year start; same-day orders including fresh carry over 3x the items of other orders; fresh items held 6 of the top 20 on amazon.com best-seller lists.

d. Amazon Pharmacy more than doubled new customers in H1; same-day Rx deliveries rose nearly 5x; year-to-date, customer out-of-pocket savings reached nearly $250mn, +400% YoY.

e. Delivery speeds continued to improve, setting new records in H1. Prime Same-Day now covers millions of items, up to 40x the selection of typical big-box stores, with the same-day network expanding; global Same-Day or Next-Day units rose over 40% YoY in H1.

f. Ultra-fast delivery Amazon Now (≤30 minutes for thousands of daily essentials) added 80 U.S. towns in Q2 and several major cities in Egypt, now spanning 9 countries and 250+ towns. GMV and units grew over 80% QoQ, and customers served grew over 60% QoQ.

g. Newly launched Amazon Supply Chain Services lets any business use the same end-to-end supply chain behind Amazon's own ops for transport, warehousing, and delivery, already winning large clients such as Procter & Gamble, 3M, Land's End, and American Eagle Outfitters.

h. Retail AI: Alexa for shopping offers personalized recommendations, comparisons, price history, and automation such as price alerts and auto-ordering. Over the past 12 months, users exceeded 350mn; Q2 active users nearly doubled and interactions rose over 5x YoY. Vision search Amazon Lens added 10 countries, now reaching 21.

**6\. Fulfillment network & costs**

a. Progress in inventory placement, shorter shipping distances, fewer touch points, and higher order consolidation.

b. Expanded robotics and automation deployment, upgrading existing facilities with next-gen tech; Cardinal and Sparrow robotic arm fleets are expected to more than double in FY26.

c. Despite fuel inflation from Middle East conflicts and long-haul rate increases amid driver capacity shortages, unit service costs continue to decline. Ex-fuel and long-haul rate increases, transportation cost growth is slower than global unit growth, similar to last quarter; on OP, these cost pressures were partly offset by FBA fuel and logistics surcharges implemented in Apr. Management still sees substantial productivity headroom across the global network and will keep raising speed standards; OPM may fluctuate and progress may be non-linear, but the company maintains disciplined, long-term unit cost reduction.

d. Tariff impact limited: The team mitigated tariff costs via early purchases and forward stocking; and Amazon is not the importer of record for most items sold (vendors typically import and pay duties). Where tariffs did raise costs, the company largely absorbed rather than passed them on—third-party firm Profitero finds Amazon prices average 14% lower than other retailers. For limited, traceable pass-through cases, affected customers will be proactively contacted and refunded; remaining refunds, like other large retailers, will be reinvested in low prices.

**7\. Advertising**

a. Q2 ad revenue was $19.8bn, +26% YoY. Sponsored products remain the largest format and key growth driver.

b. More shoppers discover products via agentic and conversational experiences, including Alexa+ and Alexa for shopping; shoppers clicking sponsored prompts convert 48% higher and spend 21% more on average.

c. Engagement in Prime Video ads and sports continues to grow. In the first NBA season on Prime Video, 30+ new advertisers joined; inventory for Thursday Night Football, NBA, WNBA, and NASCAR sold out. Multi-sport advertisers achieve 2.3x de-duplicated reach vs. single-sport, with 12% higher spend and 17% more orders on Amazon.

d. AI tools: The ads agent compresses hours of setup and targeting into minutes; CPM down 8% and CAC down 6% for advertisers using it, expanded to 11 new countries this year.

**8\. Entertainment, Alexa, and Satellites**

a. The NBA's first season on Prime Video performed strongly; Game 7 of the Eastern Semifinals peaked at 6.5mn U.S. viewers, surpassing broadcast TV a year earlier. In Europe, NBA viewing on Prime Video more than doubled YoY, setting average-view records; original series 'Off Campus' reached 36mn global viewers in the first 12 days, the third-highest series debut in Prime Video history.

b. Alexa+ rolled out to Germany, Austria, France, and Brazil, with hundreds of millions of customers using the new Alexa experience. Where deployed, Alexa drives business: in the U.S., customers using Alexa for shopping spend over 40% more per order; Prime sign-ups among Alexa+ trial users are nearly 25% higher.

c. **Amazon LEO has nearly 400 satellites in orbit, sufficient to launch initial satellite internet service this year; substantial revenue commitments are in place from enterprise and Gov. customers, with 20+ partners to expand global coverage.**

**2.2 Q&A**

**Q: The market assumes AI workloads are low-margin at least near term. What drove AWS's 39% OPM in Q2, and how sustainable is it?**

A: We are pleased with AWS's combination of strong revenue growth and margin expansion in Q2, especially at this scale. As noted, despite heavy investment, AWS margins remain robust, up 650bps YoY, or 520bps ex the derivative accounting gains cited earlier.

We have said margins will fluctuate, depending on investment intensity, product mix, and the AI vs. non-AI mix. But AWS profitability is not random; it reflects disciplined efficiency gains, capacity optimization—of which we benefited materially in Q2—and tight control of fixed costs. Margins will vary, but the YoY outcome is very strong and we are comfortable with it.

Additionally, as noted at the outset, **AI margins appear to be tracking along the same trajectory as core at a similar stage, slightly ahead**, which makes us optimistic.

**Q: Bedrock customer momentum is strong, with this quarter's spend exceeding all prior quarters combined. Does Amazon need its own frontier model to deliver full-stack?**

A: AWS and Amazon can be extremely successful even without owning a frontier model. **There will not be a single model that rules them all**—we already see this, with interest beyond Anthropic and OpenAI extending to open-source models, all available in Bedrock, which is one reason for its rapid growth. If you are building critical AI applications, you want access to all viable models.

**Models will leapfrog at different times, and many will be comparable in capability. What matters is leading choices with the right price-performance and the right governance and security**, and nothing offers this as comprehensively as Bedrock today. We also use these models ourselves.

Even so, we are advancing our own frontier models for several reasons. First, it gives us extra cost control—both for our consumer apps and to lower costs for customers—having a player committed to ongoing price-performance gains benefits customers. Second, it gives us more control over model priorities for external and internal needs. Third, it gives us some control over pace. We expect at least half a dozen strong, broadly comparable models over the next few years, all in Bedrock, and one of them will be ours.

**Q: Did AWS acceleration mainly result from large capacity coming online in Q2? How many GWs will be added in H2 vs. H1, and how do you see 2027?**

A: Growth this quarter reflects multiple drivers, and we are excited by the fifth consecutive acceleration and the largest in 18 quarters. Customers choose AWS for its breadth across core and AI, leading operations and security, and the desire to co-locate inference with other workloads and data—where AWS's installed base is far larger than anywhere else.

On core, two factors stand out. Enterprises are planning migrations from on-prem to cloud—recall ~85% of global IT spend remains on-prem, set to flip over the next 10–20 years—and with AWS's advantages, we are winning most of this shift. AI is growing extremely fast and pulls core with it, as post-training, RL and tool calls run on CPUs and core, with Graviton making AWS more attractive.

So yes, we are adding substantial capacity, but growth has many other reasons. **On build, we are on the same cadence noted several quarters ago—we said we expect power capacity to double by end-2027 vs. 2025, and we remain on that track.**

**Q: For 2027, with demand coming, is it time to slow long-cycle DC spend, or will new DC builds remain necessary for the next 2–4 years?**

A: Demand is very strong. Beyond what we discussed for 2026, **most of 2027 capacity—while we are adding a lot—has been essentially pre-booked, with a meaningful portion already pre-booked for 2028.**

It helps to think about current demand and the adoption curve: AI adoption is very 'barbell-shaped' today. **On one end, AI labs consume vast compute, alongside a few breakout Gen AI apps like Claude Code and ChatGPT; on the other, enterprises realize real value in cost avoidance and productivity, such as customer support automation, process automation, and anti-fraud.** The middle is production workloads across enterprises where some use inference widely, but most do not yet.

This will change materially over time. The largest absolute opportunity will be existing enterprise production workloads and new workloads from startups. We are still early in understanding the full magnitude of AI demand; it will transform every known customer experience and create many new ones. Whether the middle grows as steeply as the lab end is unclear, but demand ahead is substantial, and we will keep investing to maintain today's leading share. As noted, AWS has the potential to reach $1 trillion in revenue, and we intend to lead.

**Q: You mentioned potentially selling Trainium to third-party DCs in the last 90 days. When could this start, and how would ROIC compare to AWS-native workloads?**

A: We are excited by the chip business, now above $25.0bn ARR. We believe we have leading price-performance in both AI and CPU chips—Trainium and Graviton.

**We have multi-year, multi-GW commitments from Anthropic and OpenAI, the two largest AI labs**, plus growing adoption of Trainium by many others as noted earlier. Demand for Trainium is immense.

Many customers are enthusiastic to use it as we offer today, while more ask for discrete training-chip supply, even off our cloud. We are actively exploring this and see a strong likelihood of doing so over time.

**Q: With Kiro, AWS Transform and productivity tools, is this a more deliberate push into the application layer, and does it strengthen the platform above infra?**

A: Building some of these apps is a meaningful opportunity for customers and AWS. Partly, customers tell us what they want; partly, we need these capabilities inside Amazon.

Our agentic coding services are one example. Amazon Quick is particularly interesting—many inside the company wanted a truly intelligent AI assistant for work. We are a very document-centric company, and people prefer summaries first, then authoring analysis and responses, with agents performing BI; Quick began for research, BI, and summarization.

Usage grew so much internally that people asked for managing email, Slack, calendars, and using these together more efficiently. That became the next-gen Quick. As noted, it spread fast inside Amazon and is already in production at many external enterprises covering large employee bases.

We see similar opportunities across enterprise needs. Amazon Connect is used by all five major airlines and many leading banks and healthcare companies, and is still growing fast; AWS Transform greatly simplifies software migration. Continuum likewise: big-company AI discussions almost always raise security, amid noise around strong new models; Continuum helps find code vulnerabilities, design fixes, and assist deployment. These early products have strong potential, and more are in the works, which we believe will help customers and our business.

**Q: Considering demand signals and capacity plans, how are you thinking about capital sources for builds over the next few years?**

A: As noted, we issued bonds this year. We have many options to continue funding AWS growth. We will keep evaluating and act appropriately, but nothing further to share today.

**Q: Reported RPO is 2.5x Q3 2025 levels (when you guided to power capacity doubling by end-2027). What does this expansion imply for future capacity, qualitatively for 2028+?**

A: Backlog is growing very substantially, as you noted, reflecting high customer enthusiasm for AWS across core and AI. We are obviously aware of this and have fully incorporated it into our capex outlook.

We expect to keep signing more deals over time. We will keep pursuing the opportunity to maintain today's leading market share.

**Q: With supply-chain inflation driving this year's capex uplift, how will AWS pricing account for future cost inflation? Do long-term contracts support stable returns?**

A: Beyond a portion of demand that is on-demand without contracts, most demand comes from signed deals and agreements. Signed deals carry the prices and terms agreed for their duration. New agreements always reflect our cost position at the time of pricing.

I think it is no secret globally—**memory, HDDs, SSDs and other components are currently priced higher.**

**Q: On quick commerce, fresh, and essentials, what are you seeing on adoption and spend by consumers? Any country or regional differences?**

A: We are optimistic and excited about the pace of expansion in essentials and fresh. Part of this is broader selection; part is much faster delivery over the past 2–3 years. When you can deliver quickly, offer broad selection and low prices, consumers consider you in more shopping occasions.

We have discussed groceries for some time—the business is very large, with GMV over $150.0bn last year, making us the No. 2 U.S. grocer. Much of this is non-fresh, i.e., center-store CPG, canned goods, beauty and OTC drugs. Whole Foods Market is a leading organic grocer; in-store regions are growing well ahead of comparable peers, with profit trajectories improving. We also found a new urban format, Daily Shop, with a strong start and fast expansion.

Separately, to serve the breadth of customers we aim to reach, we needed a way to deliver mass brands and fresh at scale. After many experiments, we have found a lever that truly changes the game—fresh via same-day facilities, now in 2,300 U.S. cities. In these cities, 9 of the top 10 best sellers are fresh items. Fresh monthly actives are up 50% since year start, and same-day orders including fresh carry 3x the items of other orders.

We see very strong momentum in essentials and groceries. By the way, we are continuing other offline grocery experiments, but fresh in same-day facilities is working and reshaping our essentials business trajectory.

<End of text\>

**Risk disclosure and statement:**[**Dolphin Research Disclaimer & General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md)

## Comments (1)

- **Jack www · 2026-07-30T22:49:44.000Z**: Here I am