Amazon Earnings Call: AWS Quarterly Revenue of $46 Billion, Capital Expenditures to Continue Rising in 2027

Cailian Press Financial News Agency

Open the APP

Amazon earnings call: Soaring AI spending scares off the market; even AWS's trillion-dollar "pie in the sky" loses its appeal?

Cailian Press Liu Rui | Editor: Lin Qi | 2026-10-30 09:01

① Amazon AWS quarterly revenue was $46 billion, with $578 billion in backlog orders. Amazon executives expect AWS to maintain growth of over 40% in 2027-2028 and to reach a trillion-dollar revenue scale within 5-10 years; ③ However, after hearing the company's expectation that "capital expenditure will also increase significantly in 2027," the market's concerns about the outlook for the company's AI investment further intensified.

Cailian Press, October 30 (Editor: Liu Rui) — After the U.S. market close on Thursday, Amazon released its third-quarter 2026 financial report, with both overall revenue and operating profit rising, while the AWS cloud business grew 39.3% year over year, marking its sixth consecutive quarter of accelerating growth.

On the earnings call, company management set out extremely optimistic targets for future growth: AWS to maintain growth of over 40% in 2027-2028 and to reach a trillion-dollar revenue scale within 5-10 years.

In addition, company executives also responded to core questions including the logic behind AWS's trillion-dollar revenue, margin pressure, the balance of internal and external supply of Trainium chips, the outlook for Bedrock, robot deployment, growth in fresh grocery e-commerce, and the payback cycle for capital expenditure.

However, after hearing the company's expectation that "capital expenditure will also increase significantly in 2027," the market's concerns about the outlook for the company's AI investment inevitably further intensified, and the company's stock fell 7.3% after hours.

I. Key points of the core financial report

1. Overall performance: Total revenue of $201.9 billion (+12% year over year), operating profit of $28.1 billion, and a record-high profit margin of 13.9%. Q4 guidance is revenue of $240-247 billion and operating profit of $28.5-33.5 billion.

2. AWS: Revenue of $46 billion, annualized at $184 billion, backlog orders of $578 billion, and a profit margin of 38.7%; AI inference revenue accounts for more than half of AWS AI revenue.

Bedrock explodes: Spending by Fortune 500 companies increased 7-fold year over year, token volume increased 17-fold year over year, and it connects to multiple mainstream large models.

Orders for in-house chips Graviton/Trainium increased 20-fold year over year, and Trainium is planned to be sold externally by the end of 2027.

3. Retail & advertising: Retail sales volume grew 15% year over year on a high base; fresh groceries and daily necessities grew at 3 times the rate of other categories, significantly driving cross-category spending. Advertising revenue was $19.6 billion, up 12% year over year; affected by scheduling disruptions, Q4 growth is expected to recover to nearly 20%; external AI agents have not yet hit the advertising business.

II. Core Q&A (condensed)

1. AWS capacity and retail costs: Quarterly capacity deployment naturally fluctuates, and a large amount of capacity will still come online in the future; in Q4, fuel and line-haul freight pressure is higher than in Q3, and seller surcharges can only cover part of the cost.

2. The logic behind AWS's trillion-dollar revenue & robots

Trillion-dollar logic: Large computing orders are locked in in advance; enterprise AI inference is still in its early stages, and falling inference costs will unleash demand; new scenarios such as security and AI Agents continue to open up incremental growth, and a large number of applications will be transformed into Agent architectures in the future.

Robots: There are currently more than 1 million warehouse robots, with improved safety; repetitive positions are gradually being replaced by machines; new warehouses deeply integrate robots, which will be deployed at scale only after cost efficiency is met, while last-mile delivery is also being explored.

3. AWS profit margin & fresh grocery business

Fresh groceries: Same-day fresh grocery users purchase 7 times as many daily necessities as ordinary users, and non-fresh spending doubles, making it the core gap in retail growth.

4. Bedrock and the balance of internal and external chip sales

Bedrock (a fully managed generative AI product): Multiple models will coexist in the future, positioned to aggregate various large models, with the goal of growing to the same scale as EC2 (Amazon Elastic Compute Cloud product).

Chips: Constrained by wafer capacity quotas; external sales generate short-term cash, while internal use in cloud instances can generate cash flow continuously for years; Trainium external sales will begin at the end of 2027, with the ratio of internal to external supply adjusted dynamically each year.

5. Capital expenditure and payback cycle: 2026 capital expenditure is about $220 billion, and 2027 spending will continue to rise; the infrastructure payback cycle will continue to be shortened.

6. The impact of the inference service model on profit margin: Bedrock mainly runs on in-house Trainium chips; owned/long-term leased data centers + in-house hardware bring structural cost advantages, which is good for profitability.

III. Core management outlook

1. AWS: Growth of >40% in 2027-2028, moving toward trillion-dollar revenue

2. Retail: Focus on fresh perishables to drive overall spending; robots will reduce costs and improve efficiency over the long term.

3. Advertising: Q4 growth recovers, and external general-purpose AI agents pose no substantial threat for now. Full transcript of Amazon's third-quarter 2026 earnings call (AI-assisted translation, some content abridged)

Dave Fildes (Vice President of Investor Relations, Amazon): Hello everyone, and welcome to Amazon's Q3 2026 earnings call. Joining us today to answer questions are CEO Andy Jassy and CFO Brian Olsavsky. I'll hand it over to Andy.

Andy Jassy (CEO, Amazon): Thanks, Dave. This quarter we delivered $201.9 billion in revenue, up 12% year over year, and operating income of $28.1 billion, up 61% year over year. Amazon delivered another strong quarter.

On AWS, our strategy is playing out. Despite a tough comparison against last year's large-scale capacity expansion, AWS still grew 39.3% year over year, its sixth consecutive quarter of accelerating growth and its fastest growth rate in nearly five years.

Our backlog stands at $578 billion, with triple-digit year-over-year growth and a highly diversified customer base. We deploy an AI accelerator roughly every 14 seconds, and capacity is snapped up by customers as soon as it comes online. Revenue in our core compute and storage businesses grew at its fastest rate in years.

Our core business has long maintained excellent operating margins, and AI segment margins are improving rapidly as well—at the same stage of development, they are already higher than our core business's margins were in its early years. We achieved this even in the face of rising component costs, including memory, and the high depreciation from large-scale capital expenditures.

Capital expenditure payback periods continue to shorten, and we are more confident than ever that our investments in AWS will generate far more revenue and free cash flow in the future than we expected a few years ago, with very attractive returns on investment. A powerful flywheel has already formed inside AWS. To borrow Jim Collins's flywheel logic from Good to Great: completing A naturally drives B; completing B drives C, and so on in a continuous loop.

In the AWS business: leading features, security, and performance drive customers to deploy their applications and data on AWS; production inference workloads tend to be deployed alongside the rest of a customer's data and applications, which in turn drives AI demand for AWS; AI reinforcement learning, model post-training, and tool calling consume large amounts of core cloud resources, driving growth in our foundational business; and growth in the foundational business in turn pushes us to keep iterating and enriching AWS's overall product capabilities. This is completely different from a business model that simply provides GPU hosting or short-term data center leasing. This is a long-term platform strategy that helps customers achieve technological innovation, ensure stable business operations, and build reliable partnerships that can grow together over many years. Whether or not hardware supply is tight, this model delivers highly attractive long-term economics.

Let me give a few examples of AWS's business progress. On the foundational business side, about 85% of global IT spending is still going to on-premises data centers, and we are in the middle of a 10-20 year cycle in which cloud computing will rewrite this landscape. The market used to believe that critical core workloads like mainframes and storage area networks would never move to the cloud. But now, with dedicated cloud replacement solutions built on AWS, plus AI-driven migration tools that greatly reduce migration difficulty, these workloads are migrating to the cloud at scale. Enterprise customers can reduce capital expenditures, lower operating costs, accelerate innovation cycles, improve developer productivity, and more easily access data to put AI to work. Over the past 20 years we have honed an efficient operating system, and our core business margins are strong.

As the core business continues to grow, we keep reinvesting in foundational capabilities and AI features. For example: customers are generating ever-larger volumes of data and need analytics and AI to extract value from it, but moving data across storage and analytics engines is cumbersome and costly. Through the acquisition of DuckLabs, DuckDB will become the universal SQL dialect for AWS's storage and analytics services. This modern analytics engine is fast, easy to use, and universal. Customers can write a SQL query once and run it directly on any AWS analytics service. For the first time in data analytics, there is zero migration and zero adaptation cost between query engines. This is significant, and it also helps AI agents complete query tasks efficiently.

On the AI side, our inference service Amazon Bedrock continues to iterate and its business is growing explosively. Fortune 500 companies' spending on Bedrock nearly 7x'd year over year, and total token calls grew nearly 17x year over year. Bedrock has become the fastest-growing billion-dollar business in AWS history. We recently brought OpenAI models into Bedrock, and usage is climbing rapidly; we also added Bedrock-hosted Kimi, GLM, and SpaceXAI models. You can't talk about AI today without talking about security, and that is an important reason customers keep choosing Bedrock. Bedrock is designed to protect customer data security and privacy: input and output data are not shared with model providers and are not used to train foundation models.

When customers fine-tune foundation models, the platform generates a dedicated private copy for the customer's exclusive use; data is encrypted in transit and at rest throughout. Frontier models are becoming increasingly capable, and software vulnerabilities that were previously hard to find are now being exposed at unprecedented speed. We launched AWS Continuum to discover, prioritize, validate, and remediate vulnerabilities at machine speed. Enterprises including Capital One, IBM, Infor, Luminate, and Robinhood have already begun using it. In Q3 we announced that we will embed this capability into coding tools, including Anthropic's Claude Code, OpenAI's Codex, and AWS's own Kiro. This new security service has received far more customer interest than any comparable product in our history. AI agent deployment continues to accelerate, and a large number of agents are built on AgentCore components.

Tens of thousands of customers currently use AgentCore; in Q3, the number of agents deployed on the platform exceeded the total of the previous three quarters combined. In addition to customers building their own specialized agents, the vast majority of enterprises also use out-of-the-box agent services, and several of our products are widely popular with customers: Kiro for code development, the Amazon Connect call center agent, and the AWS Transform migration tool—all iterating rapidly with a continuously expanding customer base. The AI office assistant Amazon Quick is also growing rapidly. Large enterprises such as Mondelez, EchoStar, 3M, Southwest Airlines, and DXC Technology have chosen this product, which is purpose-built for office scenarios and never compromises customer data security.

Our custom chip business maintained triple-digit year-over-year growth; revenue commitments for Graviton and Trainium each grew more than 20x year over year. The AI business continues to benefit Graviton chips: model post-training, reinforcement learning, and agent tool calls mostly run on CPUs. Graviton is a top-performing CPU with 30–40% better price-performance than competing products. In Q3 alone, customers using our custom chips saved more than $2 billion in total costs. We have just open-sourced TorchNeuron, so developers can run the PyTorch code most people are already using directly on Trainium chips without rewriting code. Even as new-generation chips launch, demand for the previous generation remains strong. After the launch of Graviton5, pending orders for previous-generation Graviton instances still doubled quarter over quarter. After the first half of the year, spot prices for short-term capacity of older Trainium instances carried a premium.

Demand for Trainium is extremely strong, and many customers have urged us to sell this chip externally. We plan to begin external supply in late 2027.

Overall AWS capacity demand is stronger than ever. Enterprise large-scale deployment of inference workloads is still in its early stages, but inference revenue already accounts for more than half of AWS AI revenue and its share continues to rise, while model training also contributes substantial revenue. Capital expenditure in 2026 is expected to be about $220 billion, the vast majority going to AWS; capital expenditure in 2027 will rise significantly further, and at the same time we can clearly see substantial returns on investment, as stated on our previous earnings call. AWS annualized revenue has reached $184 billion, but there is still enormous room for growth.

Backed by continuous innovation and new foundation and AI products, enterprises are willing to deploy their core and AI workloads on AWS, combined with large-scale new capacity added each year. We expect AWS year-over-year growth to exceed 40% in both 2027 and 2028, and to grow into a trillion-dollar revenue business over the next 5–10 years. The industry inflection point has only just begun. Our strategy is working and we expect it to continue working. On the retail business, I believe some market participants underestimate the progress this business has made in recent years. It is hard to find another hundred-billion-scale retail business that, including advertising, achieves double-digit operating margins while maintaining its current growth rate.

Year to date, unit sales grew nearly 15% year over year on top of a very high base, after exceeding 30 billion units for all of last year. Over the past twelve months, gross merchandise volume increased by more than $90 billion net. Growth significantly outpaces retail peers, with quarterly incremental transaction volume about twice that of the second-largest U.S. competitor, continuing to gain retail market share. Over the past several years, North America and International segment operating margins each improved by nearly 10 percentage points, allowing us to reinvest resources in improving customer experience. Nearly 80% of retail transactions still occur in physical stores, and over the long term e-commerce's share will continue to rise.

We are building a highly efficient, ultra-fast fulfillment network to support this shift, delivering an excellent customer experience while generating substantial free cash flow and return on investment. Faster delivery, fulfillment cost optimization, grocery and pharmacy, and AI shopping—multiple key initiatives are working in concert and delivering results across the board. We are investing more in our same-day delivery network, Amazon Now, and Prime Air ultra-fast delivery, and delivery speeds continue to improve. In Q3, global delivery performance hit an all-time high, with more than 1.5 billion same-day/next-day orders delivered to Prime members. Amazon Now is expanding rapidly around the world with enthusiastic market response, now covering more than 450 towns and cities, with order volume on the platform up 10x year over year.

Fulfillment cost per unit is expected to decline for a fourth consecutive year. We can continue to match or even undercut competitor pricing while economically and efficiently supplying fresh groceries and everyday essentials. Faster delivery times are driving the grocery business, including perishables and everyday essentials, which reached nearly $250 billion in gross transaction value over the past twelve months. In Q3, grocery and everyday essentials grew three times as fast as the rest of our retail categories, accounting for about half of all units sold. Customers who use same-day grocery delivery buy 7x more everyday essentials than typical customers; active grocery customers place orders in non-grocery categories such as home and apparel about four times as often as typical customers, with corresponding spend more than double.

Amazon Pharmacy is also growing rapidly. Its user base grew more than 60% year over year, with growth accelerating every quarter in 2026; repeat customers grew more than 50% year over year. In the J.D. Power rankings, Amazon Pharmacy was ranked the No. 1 mail-order pharmacy for the third consecutive year, and the product experience continues to iterate, with prescription medications now delivered within three hours in more than 350 cities. Over the past twelve months, more than 400 million customers used the AI shopping assistant Alexa Shopping, with interactions surging 570% year over year. In the U.S., customers using the feature spend 50% more per order. The tool can research products, compare prices side by side, review a full year of price history, generate personalized gift recommendations, and monitor promotions and price drops for users.

While we will also partner with general-purpose external intelligent agents, we believe Alexa Shopping will become the preferred agent for customers shopping on Amazon. When browsing products and having questions, a shopping assistant is right there, and can place purchase orders directly using your order history, shipping address, and payment information—extremely convenient for customers. Next, Amazon Advertising: Q3 advertising revenue was $19.6 billion, up 12% year over year.

There were some timing-related distortions this quarter, but overall business momentum remains strong. Three factors affected growth this quarter:

First, most countries moved the Prime member major sales event to Q2, shifting a large amount of transactions and advertising revenue that would have occurred in Q3 2025 into Q2 2026. This benefits Q2 and weighs on Q3 year-over-year growth, creating an 850-basis-point quarterly distortion to advertising year-over-year growth.

Second, in Q2 2026 we added Prime's first season of NBA broadcasts, which did not exist in last year's second quarter; the new NBA season begins in Q4, so there was no related revenue in Q3.

Third, in Q3 2025 we adjusted our search engine marketing approach to drive more traffic to Amazon's on-site search, improving customer experience while increasing traffic monetization efficiency. That policy began driving high advertising growth in last year's third quarter, and this year's Q3 marks the start of a high-base comparison period. We expect Q4 advertising year-over-year growth to return to a range close to 20%.

The market will naturally wonder whether external general-purpose AI agents will significantly impact Amazon Advertising. The answer is no. At this stage, very few users complete shopping through external agents, and the experience is inferior to visiting a retailer's platform directly to search or make a one-click purchase. The advertising team completed several innovations in Q3, and we expect the effects to be gradually released over the coming months. The core product, Amazon Ads Agent, integrates the entire suite of advertising capabilities into an AI-first unified platform. Leveraging trillions of shopping, browsing, and streaming behavior signals, it delivers insights and answers questions across the full advertiser workflow, automatically completing optimization and execution. We have also launched full-funnel advertising campaigns in this product, integrating sponsored ads, display ads, video, and streaming TV ads into a single campaign.

Advertisers input basic information such as budget and duration, and Amazon AI automatically plans, executes, and continuously optimizes. Several other business lines are also gaining momentum. Prime Video has grown into an important business segment with tens of billions in annualized revenue and is now independently profitable. Hit series include The Summer I Turned Pretty, The Boys, Gen V, Reacher, and Fallout; the film Project Hail Mary had a strong box office; sports content includes Thursday Night Football (best start in history in 2026), Prime's NBA broadcasts, NASCAR, the Masters, and UEFA Champions League soccer. We just launched the strongest hardware product lineup in our history, with 34 new products across Echo, Kindle, Alexa tablets, Fire TV, Ring, and Bee.

The business is still in its early stages, but it has had a strong start, and we look forward to more users trying it during the holiday season. Project Amazon Leo remains on track to launch initial service this year. We are about to enter the most important period of the year: the re:Invent developer conference plus the year-end shopping peak. Thank you to all employees for your relentless innovation and for continuing to improve customers' lives. Amazon is building a business landscape unlike any before. No company in history has ever achieved a trillion dollars in annual revenue, and Amazon is poised to have two trillion-dollar businesses at the same time: AWS and online retail. Such an opportunity is extremely rare, and we are full of fighting spirit about the prospects. I'll now hand it over to CFO Brian.

Brian Olsavsky (Chief Financial Officer): Thank you, Andy. Let's start with the overall financial data. Worldwide revenue was $201.9 billion, up 12% year over year; excluding foreign exchange effects, it grew 13% year over year. Worldwide operating income was $28.1 billion, with an operating margin of 13.9%, a record high for the company. First, let's look at the North America and International segments. A reminder: this year most major countries held Prime Big Deal Days in Q2, whereas in 2025 all of them were in Q3. If we exclude the impact of the timing shift of the big sales event, the year-over-year growth rate for Q3 2026 would be 400 basis points higher. Customer engagement remained high in Q3, continuing the trend from the first half of the year.

North America segment revenue was $112.9 billion, up 6% year over year. Excluding foreign exchange effects, International segment revenue was $43.1 billion, up 8% year over year. Worldwide paid unit sales grew 12% year over year. Prime membership remains the cornerstone of the business. Hundreds of millions of Prime members enjoy a broad selection, low prices, and the fastest delivery ever, and feedback has been positive. Over the past year, monthly orders for Prime members' same-day grocery delivery have more than tripled. Now let's look at profitability. North America segment operating income was $8.7 billion, with a margin of 7.7%; International segment operating income was $1.6 billion, with a margin of 3.7%. At the fulfillment network level, the team faced transportation cost pressure from fuel inflation and higher line-haul freight rates.

Despite these headwinds, thanks to inventory optimization, shorter shipping distances, and package consolidation, overall fulfillment cost per unit still declined this quarter. Freight costs rose 11% year over year, while unit sales rose 12% year over year. Excluding fuel and line-haul inflation, freight costs grew only 8%, 400 basis points lower than unit sales growth. Robotics and AI continue to deliver quantifiable productivity gains and improved employee safety, and the effects will continue to amplify in the future. In 2026, deployment of robotic arm equipment such as Cardinal and Sparrow will double. We are also developing foundation models for robots to simplify programming and teach robots complex physical tasks, with the goal of deploying practical robots at scale. Looking ahead, there remains enormous room to improve productivity across the global fulfillment network and reduce fulfillment cost per unit, while continuing to increase delivery speed.

Quarterly operating margins will fluctuate; our core goal is long-term, continuous improvement in fulfillment cost per unit.

Now let me cover the AWS segment: revenue was $46 billion, up 39.3% year over year, driven by both the core business and AI services, with $3.8 billion in sequential quarterly revenue growth. AWS annualized revenue reached $184 billion. Customers continue to move to the cloud and expand their use of core cloud services; enterprises that want to fully unlock the value of AI are accelerating their migration to the cloud. AI investment and core cloud consumption are strongly correlated. As customers increase AI investment, consumption of corresponding core compute, storage, and networking resources rises in tandem; as more AI workloads go into production, this linkage will strengthen further.

AWS operating income was $17.8 billion, with a margin of 38.7%, benefiting from high business growth and efficiency optimization. Investment in in-house chips and in-house networking hardware, combined with software and process optimization, helps us optimize capacity and build lower-cost infrastructure. As we have repeatedly noted, AWS operating margins will fluctuate due to the scale of phased investments. Customer adoption of AI products continues to rise, AI business revenue achieved triple-digit year-over-year growth, and Bedrock is the core growth engine. On capital expenditures, Q3 cash capex was $75.9 billion, the vast majority directed to AWS and AI, matching strong customer demand. We continue to invest at scale in AI, which we believe is a huge opportunity that can drive long-term revenue, free cash flow, and improve return on invested capital.

Next, Q4 guidance. Q4 net sales are expected to be in the range of $240-247 billion. Based on current exchange rates, foreign exchange movements will create a year-over-year headwind of about 60 basis points. Q4 operating income is expected to be in the range of $28.5-33.5 billion. One factor affecting guidance: Amazon Leo's commercial service plans to begin its initial launch this year; capitalization of related production and launch costs will be deferred to the first quarter of next year, later than I expected at the previous earnings call. Therefore, most Leo-related spending in Q4 will still be expensed in the current period. We are ready to welcome holiday-season customers while serving the thousands of businesses that build their businesses on AWS.

Andy Jassy (Amazon Chief Executive Officer): Thank you to the global Amazon team for your hard work and unwavering customer focus. Continuously improving the customer experience is the only reliable path to creating long-term value for shareholders. Let's move to the Q&A session.

Moderator: Thank you. We will now open the line for questions. Our first question comes from an analyst at Bank of America.

Analyst: Thank you. Andy, the accelerating growth in the cloud business is very impressive, but the sequential revenue increase has slowed somewhat. I'd like to understand what factors on the customer side, token costs, or anything else are causing this phenomenon. Also, how do you view the capacity deployment in Q4? Brian, how much impact will fuel costs have on the Q4 guidance? Thank you.

Andy Jassy (Amazon Chief Executive Officer): Let me start with the growth question. We are very pleased with the 39.3% growth rate; achieving that rate on an annualized revenue base of $184 billion is very rare, with six consecutive quarters of acceleration, close to the highest in five years. We are adding capacity at scale every quarter, but the amount of new capacity added each quarter is not uniform. Looking back at the capacity added this year and the plan for roughly the next 8-10 quarters, there is still a large amount of capacity coming online.

Growth is driven by multiple factors working together. On AI, our complete full-stack product capabilities are key: you can build models from scratch with SageMaker AI; or use Bedrock to securely and cost-effectively call various top frontier large models.

Customers can use AgentCore and Strands to build their own AI agents and run them securely at scale; or they can directly use out-of-the-box agent services, which is the choice of most enterprises: the coding tool Kiro, the call center Amazon Connect, the migration tool AWS Transform, and the fast-growing office AI assistant Amazon Quick. Customers are adopting quickly. The amount of new capacity added each quarter will fluctuate.

Even at our current large scale, we are still guiding for more than 40% growth per year in 2027 and 2028; reaching a trillion dollars in revenue in 5-10 years is a rare major opportunity, and we are fully committed to seizing it. Our custom chips also help: enterprises deploying inference at scale place great importance on price-performance. Trainium has a significant price-performance advantage over competitors and is extremely attractive. The broad adoption of AI is driving reinforcement learning, post-training, and tool calling, which pulls demand for foundational compute; Graviton CPUs have 30-40% better price-performance than other x86 products. All these factors together drive high growth.

Brian Olsavsky (Chief Financial Officer): Regarding fuel and line-haul freight, I can't give an exact number; the pressure in Q4 is higher than in Q3. Early in the year we introduced a fuel surcharge for FBA third-party sellers, which covers part but not all of the fuel cost; the rest of the pressure shows up in retail inbound and outbound logistics costs, with the overall level higher than in the third quarter.

Moderator: Our next question comes from an analyst at Morgan Stanley.

Analyst: Thank you for taking my question. Two questions. Andy, you are very optimistic about AWS's prospects; you rarely give such a long-term target as a trillion dollars in revenue over 5-10 years. Is this trillion-dollar target calculated top-down or derived bottom-up? Where does your confidence that this revenue scale can be achieved come from? Second question, for the retail fulfillment business, what is the key robotics milestone for 2027?

Andy Jassy (Amazon Chief Executive Officer): On AWS, looking at the current annualized revenue growth rate, combined with our forecasts for 2027 and 2028, a simple calculation shows revenue scaling rapidly. Customers pre-book large-scale megawatt and even gigawatt capacity, so it won't sit idle. For these large orders, we have good visibility into demand for 2027, 2028, and even part of 2029. At the same time, we still reserve a large amount of elastic on-demand instance capacity (EC2, Bedrock) to ensure customers can use it flexibly.

Enterprises deploying inference at scale in production environments is still very early. On one hand, enterprises still lack enough in-house talent that understands AI fine-tuning and post-training; on the other hand, inference cost remains a barrier. Continuously lowering inference cost is a major mission for AWS, which is an important reason we developed our own Trainium chips. Inference costs have already fallen substantially over the past two years and will continue to decline significantly in the future.

At the same time, entirely new AI use cases keep emerging. In security, for example, new models can find vulnerabilities that are very hard for humans to find; some market views hold that in the next one to two years, security-related token consumption will account for a considerable share; add to that code development and consumer and office AI assistants (Quick), and these are only the first wave of large-scale use cases. In the future, almost all enterprise production applications will be re-architected to support intelligent agents, enabling rapid iteration while better defending against security risks. AI has enormous room to grow. Although it is hard to predict precisely over 5-10 years, the trillion-dollar revenue target is not an unrealistic imagination.

Turning to robotics. The number of robots in Amazon's fulfillment network has exceeded 1 million, and they have genuinely improved cost, productivity, and safety metrics, with warehouse incident rates at an all-time low. Progress over the past 4-5 years has been enormous, but both our deployed products and our in-development solutions are still early. Fulfillment centers will still have many employees in the future, but a large number of repetitive manual roles will be handed to robots. We are investing in multiple forms of robotics. In the future we will also explore last-hundred-feet delivery solutions. We will evaluate comprehensively on productivity, safety, cost, and delivery speed, and deploy at scale only after meeting economic metrics. Newly built same-day delivery warehouses are designed with deep robotics integration. What robotics has achieved now is just the tip of the iceberg compared with its potential over the next five years.

Host: The next question comes from an Evercore ISI analyst.

Analyst: Two questions. First, thank you for sharing the long-term AWS growth outlook. I'd like to ask about margins. There will be quarterly volatility; over the next two years, can AWS margins stay at current levels, hold firm at 30%+ or even move toward 40%? Second, on the fresh perishables fulfillment business, I see it advancing very quickly around me. How many Prime households does it cover now? Roughly how much does it lift customer spending—is it 20%, 50%, or 100%?

Brian Olsavsky (Chief Financial Officer): We don't provide guidance on AWS margins, but we can comment. This quarter's 38.7% operating margin was strong. This result is not accidental; it comes from continuous efficiency optimization, capacity optimization, strict control of fixed costs, and, as Andy mentioned, the rising share of our own chips.

But there are also disruptive factors: the AI business is expanding rapidly, and AI segment margins are lower than the traditional core business, so the shift in business mix will create a blended effect. Although AI margins at a comparable stage of development are already better than the core cloud business in its early years, they are still lower in absolute terms. Component prices will also cause disruption. With multiple factors intertwined, we continue to work on cost, efficiency, and scaling new products.

On the fresh grocery business, we are likewise very optimistic. Amazon's retail grocery has long been dominated by shelf-stable goods, canned goods, personal care, and pharmacy; Whole Foods, as a premium fresh format, is growing revenue and profit faster than the rest of the grocery segment on the back of the new small-format urban Daily Shop model, and will continue expanding over the next two years. We have always believed that to grow grocery, we must build up perishables. Comparing total consumer spending with actual spending on Amazon, the largest gap is in grocery, and perishables are the entry point to grocery spending. Everyday essentials are growing unit sales at three times the rate of the rest of retail and account for half of total units. Customers who use same-day delivery for perishables buy seven times as many everyday essentials as ordinary customers; grocery customers place about four times as many orders in non-grocery categories such as home and apparel, and their spending doubles. The perishable grocery business will significantly close the gap between us and customers' full shopping needs.

Host: The next question comes from a Goldman Sachs analyst.

Analyst: Two AWS-related questions. First, what key lessons has Bedrock yielded so far, and how will you continue to scale it in the future? Second, with the plan to sell Trainium externally, how do you balance the commercial opportunity of external sales against internal demand for compute within the cloud?

Andy Jassy (Amazon Chief Executive Officer): Bedrock: looking back on more than twenty years of building technology products, there is no single tool that dominates everything. The database and analytics engine fields confirm this; in the early days of AI, many thought TensorFlow would reign supreme, but now PyTorch is larger. The same is true in large models. Over the next 2-3 years there will be 6-8 mainstream models of comparable strength, each excelling in different areas such as reasoning, code, images, low latency, and low cost. The vast majority of enterprises will use multiple models at once, and a single generative AI application will also call multiple models internally.

What customers really want: a service that brings together many frontier models, offers excellent price-performance, and has robust security and permission controls. Customers can freely combine and switch models to adapt to business changes. That is Bedrock's positioning. It includes closed models from Anthropic and OpenAI as well as open models such as Kimi and GLM. Western markets will also expect more local open large models, which will gradually arrive in the future. Bedrock's market opportunity is huge, with the chance to grow to the same order of magnitude as EC2.

Balancing in-house use and external sales of our own chips: worldwide, there are really only two companies that simultaneously have a large-scale cloud business and chips strong enough to sell externally. All high-end chips come from the same foundry, and annual capacity is subject to quota limits. Once you get quota, you must weigh how much to keep for internal cloud use and how much to sell externally. External sales can quickly recognize near-term revenue; but running EC2 and Bedrock instances inside the cloud lets chips generate revenue and free cash flow continuously over a 5-6 year cycle, yielding higher total returns over the long run, so careful trade-offs are needed.

Our chip usage strategy differs from that of the other peer. We plan to begin external sales of Trainium at the end of 2027 and will reassess the balance each year. The external chip business and the cloud infrastructure business are themselves completely different businesses; the price-performance advantage from our own chips is a huge strategic advantage, whether used to lower internal cloud costs or to develop an external sales business.

Host: The next question comes from a JPMorgan analyst.

Analyst: For AWS to grow 40%+ over the next two years, how large does capital expenditure need to be? Also, what is the latest on payback periods for server, network, and data center infrastructure investments?

Andy Jassy (Amazon Chief Executive Officer): First, payback periods. We continue to push to shorten infrastructure payback periods. AI and traditional core hardware models differ, and payback periods vary by category. AWS has historically been good at extending hardware useful life and shortening payback time, and we are still making steady progress.

The scale of capital expenditure will be very substantial. AI is the biggest technological transformation of our lifetime. Continuing to grow 40% on top of an annualized $184 billion and marching toward the trillion-dollar goal means we must keep investing to consolidate our leadership position. If the transformation could be spread over 25 years, the capital expenditure pressure would be smaller, but reality is not like that. We have the ability to support this set of investments. This investment will permanently reshape Amazon's revenue and free cash flow scale; we are firmly committed to investing and maintaining our position as the industry leader.

Host: The last question comes from a Barclays analyst.

Analyst: I'd like to follow up on Bedrock. The vast majority of Bedrock workloads run on Trainium chips. As the business shifts more from selling bare metal to inference-as-a-service, how will that affect AWS revenue, especially margins?

Andy Jassy (Amazon Chief Executive Officer): That's right, the vast majority of Bedrock currently runs on Trainium. We do this largely because we are very satisfied with the price-performance it delivers, which both allows us to lower prices for customers as much as possible and lets us achieve our desired margins.

If you consider margins in the cloud computing space and look at the cost structures of different vendors, you'll find a few things to note.

First is their data centers: how much of theirs is purchased or long-term leased, and how much is colocation or neocloud? The operating margin of neoclouds is about 3,500 basis points lower than that of long-term leased data centers.

Then you look at what hardware they use to run their servers. If you use the same hardware as everyone else, your cost structure is similar to everyone else's. If you can use different hardware, you have an advantage.

I think if you look at our company, most of our capacity is owned or long-term leased, which gives us a much higher operating margin than companies that primarily use colocation or new-style data center models, or that use colocation or new-style data centers for a large portion of their capacity.

We use our own chips, which not only provides a price-performance advantage but is also more cost-effective for us because we own it. I think this gives us some structural cost advantages, and when we pass these advantages back to customers, I think it is very helpful to customers and also helps AWS's overall economics.

Host: Thank you for joining our conference call today and asking questions. A replay of this call will be available on our investor relations website for at least three months. We thank you for your interest in Amazon and look forward to speaking with you again next quarter.

Global Market Intelligence

US Stock Market News

Cailian Press statement: The content of this article is for reference only and does not constitute investment advice. Investors who act on it do so at their own risk.

Business Cooperation

Popular Unlocks

[Institutional Research] This memory chip supplier has completed a small-computing-power layout in the smart vision SoC field

This memory chip company has become a major supplier to leading AI server manufacturers and has completed a small-computing-power layout in the smart vision SoC field.

[Research Selection • Margin Trading Data] Robots + Pharmaceuticals + PCB + Chips + AI Applications: Net margin purchases in five major hot spots over the past week (table attached)

Robots + Pharmaceuticals + PCB + Chips + AI Applications: Net margin purchases in five major hot spots over the past week (table attached).

[Research Selection • Industry Data] Food and beverage holdings have bottomed out and are recovering; analysts judge that baijiu is clearing in an orderly manner and bottom signals are clear

Food and beverage holdings have bottomed out and are recovering; analysts judge that baijiu is clearing in an orderly manner and bottom signals are clear (table attached).

Amazon Earnings Call: AWS Quarterly Revenue of $46 Billion, Capital Expenditures to Continue Rising in 2027

Why read this

Topics

Tags

Body · AI translation