--- title: "AMZN (Trans): AI's once-in-a-lifetime investment opportunity" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/40319821.md" description: "Below is Dolphin Research's FY26Q1 earnings call Trans for $Amazon(AMZN.US).For our earnings analysis, see 'Retail Steady, AI Breakthroughs: Is AMZN back in the top tier?'. I. Core highlights recap — 1) Guidance: Q2 net sales of $194-199bn (FX headwind ~10bps); Q2 OP of $20-24bn..." datetime: "2026-04-30T09:42:40.000Z" locales: - [en](https://longbridge.com/en/topics/40319821.md) - [zh-CN](https://longbridge.com/zh-CN/topics/40319821.md) - [zh-HK](https://longbridge.com/zh-HK/topics/40319821.md) author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)" --- # AMZN (Trans): AI's once-in-a-lifetime investment opportunity **Below is Dolphin Research's transcript of**$Amazon(AMZN.US) **FY26 Q1 earnings call. For our take, see '**[**Retail steady, AI breaking through — is Amazon back in the top tier?**](https://longbridge.com/zh-CN/topics/40319128)**'.** **I. Key takeaways** 1\. **Outlook**: Q2 net sales guided to $194–199 bn (FX headwind ~10 bps). Q2 OP guided to $20–24 bn, including seasonal step-up in SBC, approx. $1 bn YoY cost increase in Amazon Leo North America, and higher fuel-driven transport costs (partly offset by FBA surcharges). Prime Day falls in Q2 in most countries incl. the U.S. (Q3 in 2025). 2\. **Headline metrics**: Total revenue $181.5 bn, +17% YoY (+15% ex-FX). OP $23.9 bn, OPM 13.1% (record high). North America revenue $104.1 bn, +12% YoY, OP $8.3 bn, OPM 7.9%. Intl revenue $39.8 bn, +11% YoY ex-FX, OP $1.4 bn, OPM 3.6%. 3\. **AWS**: Revenue $37.6 bn, +28% YoY (fastest in 15 quarters), with a $2 bn QoQ step-up from Q4 to Q1 (largest ever). ARR \>$150 bn, OP $14.2 bn. Remaining performance obligations (RPO) at $364 bn in Q1, excluding the recently announced \>$100 bn Anthropic agreement. 4\. **CapEx**: Q1 cash CapEx $43.2 bn, concentrated in AWS and gen-AI. Management continues to view this as a once-in-a-lifetime investment opportunity and expects elevated spend for years. 5\. **Amazon Leo cost cadence**: Commercial service expected to launch in Q3. Capitalization of certain production and launch costs to begin in Q4. **II. Call details** **2.1 Management highlights** 1\. **AWS & AI** **a. AI revenue run-rate \>$15 bn, nearly 260x AWS's revenue at launch three years ago; triple-digit YoY growth.** b. Bedrock customer spend rose 170% QoQ. Tokens processed in Q1 exceeded all prior years combined. Over 125k customers use Bedrock, and nearly 80% of the Fortune 100 engage with it. c. Announced OpenAI models on Bedrock (GPT-5.4 live, 5.5 imminent). Previewed Bedrock Managed Agents (a stateful runtime co-developed with OpenAI) to accelerate agentic AI adoption. d. Strands (agent dev framework) surpassed 25 mn downloads, up 3x QoQ. AgentCore is deploying an agent every ~10 seconds on avg. e. Kiro (coding agent) developer base doubled QoQ, with near 10x enterprise usage growth. Transform saved clients over 1.56 mn hours of migration work. Quick added \>4x new customers QoQ and released desktop v1. f. New wins include OpenAI, Anthropic, Meta, NVIDIA, Uber, U.S. Bank, Fox, Southwest Airlines, U.S. Army, Bloomberg, Cerebras, AT&T, Nokia, among others. **2\. Custom silicon** a. **Chip biz grew nearly 40% QoQ in Q1, with revenue run-rate \>$20 bn and triple-digit YoY growth. On a chipmaker basis (incl. internal consumption), run-rate is ~$50 bn,** ranking among the global top three in data center silicon. **b. Trainium committed orders exceed $225 bn.** Trainium2 offers ~30% better price-performance than comparable GPUs and is essentially sold out. Trainium3 (shipping early 2026) improves price-performance by another 30–40% vs Trainium2 and is near fully booked. Trainium4 (mass supply in ~18 months) has received heavy pre-orders. c. Anthropic and OpenAI signed multi-year, multi-GW Trainium commitments, with companies like Uber also betting on Trainium. **d. Trainium is expected to save tens of billions of dollars in CapEx annually and add several hundred bps of margin leverage in inference.** **e. Meta committed to tens of millions of Graviton cores.** Graviton delivers ~40% better price-performance vs x86, with 98% of the top 1,000 EC2 customers using it. f. Continues deep collaboration with NVIDIA. 3\. **Stores** a. Order volume rose 15% YoY, the highest since the post-COVID reopening. b. Added 600+ marquee brands. **Avg. selling prices declined YoY in Q1.** c. Grocery is scaling quickly, with 2025 sales expected to exceed $150 bn, making Amazon the No. 2 U.S. grocer. d. Fresh delivery now covers 2,300+ U.S. cities. Fresh sales rose over 40x YoY, with same-day fresh baskets nearly tripling and spend per order \>80% higher. e. Whole Foods has 550+ stores, with 100 more planned over the next few years. f. Millions of SKUs available for same-day delivery. 1-hour delivery covers hundreds of cities, and 3-hour delivery covers 2,000+ cities. g. Amazon Now (30-min delivery) reaches tens of millions of customers across nine countries. India orders are growing 25% MoM. h. 3P seller sales grew strongly in Q1, particularly in the U.S., Europe, and Brazil, where seller fees have been lowered. 4\. **Fulfillment network efficiency** a. With orders up 15% YoY, outbound transport costs rose 12% YoY and fulfillment costs rose 9% YoY (both ex-FX). b. All new large U.S. fulfillment centers starting 2026 will deploy next-gen robotics and automation. Early results show safety gains, higher productivity, and lower service costs. c. Continued optimization of inventory placement to shorten delivery distances, reduce touches, and improve consolidation rates. 5\. **Advertising (Amazon Ads)** a. Revenue $17.2 bn, +22% YoY. b. Named a leader in omnichannel ad platforms by Forrester. c. Deepened partnership with Netflix (Amazon Audiences) and teamed up with Comcast Advertising to expand local ads. d. Creative Agent (AI creative agent) expanded to seven countries incl. Canada, France, and Germany. Introduced sponsored products and brand prompts in Rufus, with nearly 20% of users interacting with brand prompts continuing the conversation. 6\. **Rufus & Agentic Commerce** a. Rufus MAUs rose over 115% YoY, with interactions up nearly 400% YoY. b. Users can research products, track prices, and auto-purchase when targets are met. c. Rolled out new AI-driven seller data visualization tools. 7\. **Entertainment & Prime** **a. Paid subscriptions reached 350 mn. Prime Video is now a standalone, large, profitable business.** b. Film 'Project Hail Mary' grossed nearly $615 mn worldwide. Viewership for the NBA SoFi Play-In rose 18% YoY. c. Alexa+ expanded to Mexico, the U.K., Italy, and Spain. Conversations doubled, purchases tripled, music plays rose 25%, and smart home usage increased 50%. 8\. **Amazon Leo (satellite internet)** **a. Over 250 satellites are in orbit, with commercial service expected to launch in Q3.** b. Plans call for 20+ launches in 2026 and 30+ in 2027. c. Secured enterprise and government commitments from Delta Airlines, JetBlue, AT&T, Vodafone, NASA, and others. **d. Plans to acquire Globalstar to expand direct-to-device capability. Reached an agreement with Apple to provide satellite service for iPhone and Apple Watch.** e. Downlink performance is roughly 2x that of current alternatives, with uplink ~6x. 9\. **Zoox (autonomous driving)** a. Nearly 2 mn cumulative miles driven, carrying over 350k passengers. b. Open to the public in Las Vegas and San Francisco, with testing in eight cities. c. Announced service via the Uber app in Las Vegas and Los Angeles. **2.2 Q&A** **Q: What CapEx is needed over the next few years to fulfill AWS's current RPO, and how do custom silicon and AI infrastructure create advantage?** A: We are very pleased with AWS growth. At +28% YoY, it is the fastest in 15 quarters, and the last time we grew at this pace AWS was roughly half its current size. Delivering 28% on a $150 bn ARR base is not easy. Several forces are at work. First, customers keep choosing AWS for AI, partly due to our truly broad full-stack capabilities and partly because customers want inference close to their data and apps, and more of those sit on AWS. We also have the strongest security and operational performance. Second, our **core business is also growing significantly** — enterprise migrations from on-prem to cloud are accelerating, and the AI surge is pulling through massive core demand. All the post-training, reinforcement, agentic actions, and tool use drive core services. With our unique chip portfolio — the leading Graviton CPU and the industry-leading price-performance Trainium AI chip — we are well positioned at this inflection. **Our CapEx plan is largely unchanged. We truly see this as a once-in-a-lifetime opportunity** — every known application will be reinvented, and many we have not imagined will be built on AI, consuming large amounts of CPU and core services. I expect heavy capital investment over the next few years, benefiting customers, shareholders, and Amazon overall. **Q: What was AWS RPO in Q1, and how broad-based is it versus concentrated among a few customers?** A: Q1 RPO was $364 bn, excluding our recently announced \>$100 bn deal with Anthropic. It is diversified rather than concentrated in one or two customers. **Q: What are the key 2026 milestones for Rufus and agentic commerce?** A: We are very bullish on agentic commerce. Over the long term it will benefit customers and us. You can see the emphasis in our work on Rufus. If you have not tried Rufus in a while, it has improved a lot over the past year. We have a large and growing user base — MAUs up over 115% and interactions up over 400% YoY. While we will partner with third-party horizontal agents to improve CX, the current state reminds me of the early days when search engines tried to drive traffic to e-commerce. Third-party horizontal agent-driven traffic is only a small slice of search today because the experience is not good enough — they often lack accurate pricing or product data, personalization, or shopping history. We are talking to these companies to find solutions that work for customers and everyone involved. Over time, it will be interesting to see which agent users prefer. If you shop at a retailer you like and they offer a strong agentic shopping assistant, you will often start there — it is where you shop, they have better product data, know what similar customers buy, and you can manage your account and delivery details there. That is what we aim for with Rufus — the world's best shopping assistant. I think we are on that path. **Q: How much does listing the full OpenAI model suite on Bedrock help, and where are you prioritizing investment in Nova (in-house models)?** A: Having the full OpenAI model suite on Bedrock is a big deal and important for customers. We already run a large amount of AI workloads on Bedrock across models from Anthropic, Llama, Mistral, and others. One thing tech has shown repeatedly — in databases, analytics, models, and chips — is that customers want choice. There is no one tool to rule them all. Each model is stronger than others at certain tasks, so customers have long wanted OpenAI models in Bedrock. We enabled stateless 5.4 yesterday and expect to offer the latest 5.5 within weeks. While many model interactions are stateless today (tokens in, tokens out), the future lies in stateful models and stateful APIs. When building agents and AI apps, you do not want to start from scratch on every interaction. You need to store state, identity, and conversation and action history, and let models call different tools — which requires state. Bedrock Managed Agents, co-invented with OpenAI, is the future of agent building, and others do not have it. We will of course support other models such as Codex. This is a big deal for customers and will be good for our business. **Q: Your shareholder letter mentioned potential Trainium rack sales. How do you think about timing and opportunity given tight supply?** A: I think this is very likely. We have trade-offs — **demand for Trainium is extremely strong,** and multiple companies would take all we can produce. So we must decide how much capacity to allocate to existing demand and customers versus rack sales. For existing customers, we also distinguish Trainium with cloud infrastructure vs standalone chip sales. **But I expect rack sales will likely begin in the next few years.** **Q: What is the revenue opportunity for Amazon Leo across consumer and enterprise over the next few years, what does Globalstar add, and will you expand beyond connectivity longer term?** A: I am very optimistic about Amazon Leo. Billions of people lack broadband access, and many enterprise and government assets are not connected and thus not visible. That means they cannot access online education, commerce, shopping, or entertainment. Amazon Leo will help address this. We just completed another launch and now have over 250 satellites in orbit. When commercial service goes live, we will be one of two offerings at the leading edge of today's tech. I believe we will have clear performance advantages — roughly 2x downlink and ~6x uplink versus current alternatives — and cost advantages for customers. For enterprise and government, we have had many conversations and have signed numerous agreements even before commercial launch. The latest is Delta Airlines, committing to equip at least half its fleet starting in 2028. Another key factor for them is storing constellation data in the cloud for analytics and AI. The combination of Leo with AWS, the leading global cloud, is very compelling. Today, growth is constrained by getting the constellation to space — we plan 20+ launches in 2026 and 30+ in 2027. I think this can become a multi‑bn‑dollar revenue business. Some characteristics remind me of AWS — capital intensive upfront, with heavy early capital and cash deployed into long‑lived assets. I like the medium- to long‑term FCF and ROIC profile. On Globalstar, we increasingly see that consumers, enterprises, and governments dislike any moments of disconnection. Even in metro areas, there are always dead zones on highways. There is huge demand for direct-to-device. Globalstar has unique, scarce global spectrum, which is essential for direct-to-device. We will also gain satellite manufacturing expertise and deepen our relationship with Apple — they will use our direct-to-device services for iPhone and Apple Watch. **Q: How do rising memory and supply chain prices affect CapEx?** A: As many know, component costs — especially memory — have surged. We are in a period where supply falls well short of demand. We work closely with strategic partners. We saw this trend in mid-to-late last year and locked up significant supply with key vendors. The team has been agile, and I think we have done well to avoid being capacity‑constrained, though we are watching closely. Interestingly, changes in memory pricing and supply are further accelerating enterprise migrations from on‑prem to cloud. Vendors prioritize their largest customers, and CSPs are the largest. We see many cloud migration conversations — previously slow-moving — suddenly accelerate because we have far more supply than others. **Q: How will agentic commerce impact advertising if agents shop for users? Where are the ad opportunities?** A: I actually think this is positive for ads — good for customers and for the business. First, the tools and agents our ads team is building make advertising easier. **SMBs that used to spend weeks or months on creative and audience selection can now move much faster with agentic ad tools,** spending less time and money on creative. So as AI rises, the number of advertisers should increase materially. Second, agentic commerce tends to be multi‑turn. You are not searching once for an answer; you are asking questions, narrowing choices, and the agent is probing what you want. There are multiple opportunities in those turns to surface relevant products — many organic, some sponsored. This also creates new surfaces such as sponsored prompts. When customers ask certain questions, we generate AI‑driven suggestions that blend in sponsored prompts to help them explore deeper. This has worked well for customers in our stores. I do believe ads will perform well in an agentic commerce world. **Q: How is AI demand splitting between early adopters/mega customers and broader enterprises?** A: AI labs are spending staggering amounts on compute today, across AI and core. **Successful gen‑AI app companies are indeed spending heavily, but that is just a handful of labs.** We also see substantial enterprise adoption and usage. As I said, **the largest absolute‑value projects focus on cost avoidance and productivity — customer service automation, business process automation, fraud detection, and more.** But there is also a significant and growing set of projects around re‑inventing experiences with reasoning and AI, and more are entering production. So **we see growth in both segments.** **Q: Where are AI's biggest opportunities inside Amazon — product vs operating efficiency?** A: In short, I do not see any business or way of working that will not be profoundly affected. I have long believed that while you can incrementally add agentic and AI experiences to existing CX, ultimately — whether in three or five years, or sooner — the experiences we know will be reinvented. They will have different interfaces and interactions, and people will want to converse with them. That means you must examine every customer experience — even well‑performing ones — and allocate resources for teams to rethink from scratch. If you started today with AI and all the available tech, what would the future experience be? That is what we are doing across every experience. Adoption may take time and vary by segment, but if you are not inventing now, it is hard to lead later. All our consumer businesses — all our businesses — are doing this. On internal operations, AI will fundamentally change how we work — and it is already happening. Look at how agentic coding is changing product development. I think it will have a similar‑scale impact on DevOps, customer service, research, analytics, sales, and every function. One example from my letter: we swapped the engine of a service while it was running at full load. Normally this would take 40–50 people a year. We did it in 65 days with five very smart engineers using agentic coding tools. This is a very different operating world and where we are headed in the coming years. **Risk disclosure and disclaimer:**[**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) - [DXYZ.US](https://longbridge.com/en/quote/DXYZ.US.md) - [NOK.US](https://longbridge.com/en/quote/NOK.US.md) - [LUV.US](https://longbridge.com/en/quote/LUV.US.md) - [FOX.US](https://longbridge.com/en/quote/FOX.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) - [UBER.US](https://longbridge.com/en/quote/UBER.US.md) - [NOA3.DE](https://longbridge.com/en/quote/NOA3.DE.md) - [META.US](https://longbridge.com/en/quote/META.US.md) - [AMZU.US](https://longbridge.com/en/quote/AMZU.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [07788.HK](https://longbridge.com/en/quote/07788.HK.md) - [07388.HK](https://longbridge.com/en/quote/07388.HK.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md) - [NVDQ.US](https://longbridge.com/en/quote/NVDQ.US.md) - [FOXA.US](https://longbridge.com/en/quote/FOXA.US.md) ## Comments (1) - **cq_pIYnPr · 2026-04-30T12:59:14.000Z**: Yes, this is one of the few opportunities in this era! We must go all out to catch this tailwind! Don't get off the bus easily under any circumstances!