Amazon Earnings Call: AWS Expected to Grow Over 40% in 2027 and 2028, 2026 Capex Around $220 Billion

Driven by strong demand for AI computing power, Amazon's AWS cloud business achieved its fastest growth rate in nearly five years in the third quarter, and management even issued a rare guidance of "AWS growth exceeding 40% over the next two years, moving toward a trillion-dollar revenue."

As mentioned by Wall Street Insights, after the U.S. market closed on October 29, Amazon disclosed that Q3 revenue grew 12% year-over-year, slightly below analyst expectations, while EPS fell 3% year-over-year instead of rising, and the operating margin approached 14%, hitting a record high.

The biggest highlight of the earnings report remained the AWS cloud business. In the third quarter, AWS net sales grew 39.3% year-over-year to $45.98 billion, 1.3% above analyst expectations, marking the highest single-quarter growth rate in nearly five years, with an annualized revenue run rate of $184 billion at quarter-end, up 8.9% from the end of the second quarter.

Later on the earnings call, Amazon CEO Andy Jassy described the current AWS growth logic as a self-reinforcing "flywheel": leading features and security attract customers to migrate to the cloud, and the data and applications concentrated in the cloud in turn drive AI inference demand, which in turn drives consumption of core computing, storage and networking, thereby pushing AWS to continue reinvesting in foundational features.

Jassy expects AWS growth to exceed 40% in both 2027 and 2028, and is expected to achieve annual revenue exceeding $1 trillion within the next five to ten years. At the same time, the company's backlog reached $578 billion, with year-over-year growth in the triple digits.

Management also emphasized that AI demand will continue to push up capital expenditures. Jassy called AI "the opportunity of a lifetime" and said the company has the ability to fund this plan. Amazon expects cash capital expenditures of approximately $220 billion in 2026, most of which will be for AWS, and expects spending to "increase significantly" in 2027.

"One AI accelerator installed every 14 seconds": AWS cloud business fully erupts

As Amazon's profit engine, AWS's performance this quarter dispelled market concerns that it was "too large to sustain high growth."

AWS Q3 revenue reached $46 billion, up 39.3% year-over-year, achieving accelerated growth for the sixth consecutive quarter and the fastest growth rate in nearly five years. The current annualized revenue run rate has reached as high as $184 billion.

CEO Andy Jassy used highly vivid original words at the meeting to describe the current frenzied demand to investors:

We install an AI accelerator roughly every 14 seconds, and customers snap them up almost as soon as we install the capacity.

Not only is the current boom intense, but the future order pool is also unfathomably deep. According to disclosures, AWS's current backlog is as high as $578 billion, showing triple-digit year-over-year growth.

For the future, Jassy gave guidance with enormous imaginative space:

On such a massive base, we expect AWS to grow more than 40% annually in both 2027 and 2028. We believe AWS will become a business with more than $1 trillion in revenue within the next five to ten years. This is absolutely an extremely rare and extraordinary opportunity.

Amazon listed Bedrock as a key service for AWS AI growth.

The company said spending by Fortune 500 companies on Bedrock nearly septupled year-over-year, and total Bedrock tokens grew nearly 17-fold year-over-year, making it the fastest-growing multi-billion-dollar service in AWS history.

Jassy positioned Bedrock as a multi-model aggregation platform, believing that within the next two to three years there will be six to eight comparable frontier models coexisting, and most enterprises will use multiple models simultaneously, even mixing calls within the same generative AI application.

He compared Bedrock's long-term potential to EC2, believing its scale "could be as large as EC2."

Responding to massive capital expenditures: "This is a once-in-a-lifetime opportunity"

Amazon's third-quarter capital expenditure was $75.9 billion, mainly for AWS and AI infrastructure construction to support strong customer demand. Company management firmly believes that AI is a huge opportunity expected to drive revenue and free cash flow growth over the long term, as well as excellent returns on invested capital.

For the trailing 12 months ended September 30, Amazon's operating cash flow reached $168.3 billion, up 29% year-over-year. Free cash flow was negative $42.4 billion, compared with positive $14.8 billion in the same period last year.

Amazon maintained its full-year cash capital expenditure expectation of approximately $220 billion, without further raising it in this earnings report, but expects this spending to "increase significantly" in 2027.

Faced with this pace of "burning money," Jassy's position is exceptionally firm. He admitted that this is a strategy to build a long-term platform, not short-term data center leasing:

AI is our once-in-a-lifetime opportunity; it is the biggest technological transformation in our lifetime. If we can keep a business with $184 billion in annual revenue growing at more than 40% for two consecutive years and ultimately move toward a trillion dollars, in order to continue to maintain leadership, we must invest.

Jassy summed up the significance of this huge bet in one sentence:

We have the ability to fund this plan, and I think it will forever change Amazon's scale, and forever change our revenue and free cash flow.

In addition to buying computing power, the benefits of Amazon's self-developed chips are emerging.

In Q3 alone, customers using Amazon's custom chips (Graviton and Trainium) saved more than $2 billion. Revenue commitments for Trainium and Graviton increased more than 20-fold year over year. Jassy revealed:

Because demand for Trainium is so strong, customers keep encouraging us to offer it outside AWS, and we intend to do so by the end of 2027.

"People have somewhat underestimated the retail business": Groceries become a new engine, with over a million robots deployed

On the e-commerce side, third-quarter sales grew only 2% year over year, the slowest single-quarter growth since early 2023. Amazon's guidance for holiday-season and fourth-quarter sales, at the midpoint, also fell short of analysts' expectations, suggesting its online business may struggle to recover quickly in the near term.

But management stressed that excluding the impact of the Prime Day timing mismatch, North America and international businesses still maintained growth of 6% and 8%.

This year, the timing of Amazon's Prime Day sales events was adjusted: most major markets, including the United States, moved the event to the second quarter, while all of 2025's Prime Day was scheduled in the third quarter.

Excluding the impact of the Prime Day timing shift, year-over-year growth in the third quarter of 2026 would be about 4 percentage points higher. Management believes customer engagement in the third quarter was strong, continuing the growth trend from the first half. Jassy specifically reminded the market not to underestimate the progress of its physical stores and retail business:

I think some people have somewhat overlooked how much progress this business has made. I don't know of any other retailer whose revenue can approach Amazon's scale of hundreds of billions of dollars and still grow at this pace while maintaining double-digit operating margins (including advertising).

Among these, fresh groceries and everyday essentials have become a huge traffic driver, growing three times as fast as other retail categories.

Data shows that customers who buy fresh groceries and use same-day delivery services purchase seven times as many everyday essentials as customers who do not use the service; at the same time, these grocery customers also spend more than twice as much in non-grocery categories (such as home and apparel) as other customers.

In terms of cost reduction and efficiency gains, robotics has played a major role. CFO Brian Olsavsky noted that more than 1 million robots are now in use across Amazon's fulfillment network.

In 2026, Amazon doubled the size of its robotic-arm fleet (such as Cardinal and Sparrow), which enabled the company to lower per-unit fulfillment costs for the fourth consecutive year despite headwinds from rising fuel and transportation costs.

On the advertising business, Q3 advertising revenue reached $19.6 billion, with the slowdown mainly due to short-term factors such as the Prime Day promotion timing adjustment, which dragged on growth by about 8.5 percentage points. Management expects Q4 advertising growth to rebound sharply to a high level close to 20%.

For the upcoming fourth quarter, Amazon provided strong guidance: it expects Q4 net sales of between $240 billion and $247 billion, and operating income of between $28.5 billion and $33.5 billion.

Full transcript of Amazon's Q3 earnings call (AI-assisted translation):

Welcome, everyone, to Amazon's third-quarter 2026 financial results conference call. All participants are in listen-only mode, and a question-and-answer session will follow the presentations. This call is being recorded.

Remarks by Dave Fildes, Vice President of Investor Relations

Welcome to our third-quarter 2026 financial results conference call. Joining us today to answer your questions are our Chief Executive Officer, Andy Jassy, and our Chief Financial Officer, Brian Olsavsky.

While you listen to this call, we encourage you to refer to our press release, which contains the financial results, metrics and related explanations for the quarter. Unless otherwise stated, all comparisons on this call are against results for the same period in 2025. Our comments and responses today reflect management's views only as of October 29, 2026, and include forward-looking statements, and actual results may differ materially.

Potential factors affecting our financial results are described in today's press release and in our filings with the U.S. Securities and Exchange Commission (SEC), including our most recent 10-K annual report and subsequent filings. On this call, we may discuss certain non-GAAP financial measures, and the relevant supplemental disclosures and reconciliations to comparable GAAP measures are available on our investor relations website.

Our guidance incorporates the order trends observed to date and the assumptions we consider reasonable, but actual results are inherently unpredictable and may be materially affected by a variety of factors, including: fluctuations in foreign exchange rates and energy prices; changes in global economic and geopolitical conditions; tariff and trade policies; fluctuations in resources and supply (including memory chips); changes in customer demand and consumer spending (including the effects of recession concerns, inflation, interest rates and regional labor market constraints); global events; the growth rate of the internet and online commerce; the development of cloud services and emerging technologies; and the various other factors detailed in our filings with the SEC.

Our guidance assumes that we will not complete any additional business acquisitions, restructurings or legal settlements. Because we cannot accurately predict demand for our goods and services, actual results may differ materially from guidance.

Revenue for the quarter was $201.9 billion, up 12% year over year; operating income was $28.1 billion, up 61% year over year. The third quarter was another strong quarter for Amazon.

AWS Business

AWS's strategy is progressing effectively. Despite facing a high base effect from last year's large-scale capacity expansion, AWS's year-over-year growth rate still reached 39.3%, accelerating for the sixth consecutive quarter and marking its fastest growth rate in nearly five years. Contract backlog reached $578 billion, achieving triple-digit year-over-year growth with a diversified structure. We install an AI accelerator roughly every 14 seconds, and capacity is snapped up by customers almost as soon as it comes online. Revenue from our core compute and storage businesses achieved their fastest growth rates in recent years. Our core business's operating margin has long been at high levels, and the AI business's margin is also improving rapidly, progressing far faster than the core business did at a similar stage of development—even amid rising costs for key components such as memory and substantial depreciation pressure from capital expenditures. We continue to accelerate the payback period on capital expenditures and are more convinced than ever that our investment in AWS will create a business far larger than we envisioned two years ago, delivering higher revenue, free cash flow and return on invested capital.

AWS is forming a powerful flywheel effect—borrowing Jim Collins's description of the flywheel in Good to Great: leading functionality, security and operational performance drive customers to migrate their applications and data to AWS; customers migrating to AWS in turn drives growth in AI demand, because customers want to deploy production inference adjacent to their data and applications; growth in AI demand then drives demand for core AWS, because AI needs to use core services for reinforcement learning, post-training, tool calling and other work; and all of this in turn drives us to continuously increase reinvestment in AWS's overall capabilities. This is fundamentally different from simply providing GPU hosting or short-term data center capacity leasing; it is a long-term platform strategy—broadly helping customers innovate and iterate on their user experiences, operate reliably, and build reliable multi-year partnerships. This strategy also delivers highly attractive long-term economics for us, without relying on supply scarcity.

Examples of AWS Progress

On the core business side, it is worth emphasizing that the vast majority of global IT spending—about 85%—remains on-premises. Over the next 10 to 20 years, this landscape will gradually shift to the cloud. The industry previously widely believed that critical workloads such as mainframes or storage area networks would never migrate to the cloud, but with AWS's unique cloud alternatives and AI-driven migration capabilities, such workloads have begun accelerating their migration to the cloud. Enterprise customers benefit: reduced capital expenditures, lower operating costs, faster innovation, higher developer productivity, and enhanced data accessibility, making it easier to leverage AI. Our core business has highly attractive margins, thanks to the deep experience we have accumulated over the past 20 years of lean operations. As the core business continues to grow, we continuously reinvest resources into building core AI capabilities.

For example, as customers generate ever-larger volumes of data and seek to extract value from it through data analytics and AI, moving data between different data storage and analytics engines has become cumbersome and expensive. Through the acquisition of DuckLabs, DuckDB will become AWS's universal SQL dialect for storage analytics, simplifying data processing with a modern analytics engine—one that is fast, easy to use and broad in coverage. Customers need only write a SQL query once and can execute it through any AWS analytics service. This will be the first breakthrough in analytics, with customers bearing no migration or porting costs between different query engines—a significant development that will also greatly benefit efficient querying by future agents.

On the AI business side, we continue to expand the capabilities of our inference service Amazon Bedrock, driving significant business growth: Fortune 500 companies' spending on Bedrock grew nearly 7x year over year, and total token usage on Bedrock grew nearly 17x year over year. Bedrock is the fastest-growing billion-dollar-scale service in AWS history. We recently added OpenAI models to Bedrock, and usage is growing rapidly; we also added Bedrock-hosted versions of Kimi, GLM and SpaceX models.

Today, no discussion of AI can avoid security, which is also one of the important reasons customers continue to choose Bedrock. We ensure the security and privacy of customer data by design: inputs and outputs are never shared with model providers and are never used to train foundation models; when customers fine-tune a foundation model, we create a separate private copy for their exclusive use, and data is encrypted end-to-end in transit and at rest.

As frontier model capabilities continue to grow, they are discovering previously undetected software vulnerabilities at an unprecedented pace. To address this, we built AWS Continuum to discover, prioritize, validate and remediate code vulnerabilities at machine speed, and it is already being adopted by enterprise customers including Capital One, IBM, Infor, Luminate and Robinhood. In Q3, we announced its integration into code-writing tools, including Anthropic's Claude Code, OpenAI's Codex and AWS's Cura. Customer interest has exceeded anything we have seen for any security service we have ever launched.

Agent deployment continues to accelerate, with many built on Bedrock Agent Core. Tens of thousands of customers now use Agent Core, and the number of agents deployed in Q3 was 2x the total of the previous three quarters combined. Beyond building their own specialized agents, most enterprises also use out-of-the-box agent services, and we offer several that customers love: Kiro for code writing, Amazon Connect for contact center agents, Amazon Transform for migrations, and Amazon Q as an intelligent AI work assistant—all iterating rapidly and winning broad customer recognition. Amazon Q in particular is growing extremely fast, as large enterprises such as Mondelez, EchoStar, 3M, Southwest Airlines and DXC choose it as an AI companion purpose-built for work with zero compromise on customer data.

Our chip business continued to grow triple digits year over year, with revenue commitments for both Graviton and Trainium growing more than 20x year over year. AI is giving Graviton additional growth momentum, because post-training reinforcement learning and agent tool calls run mainly on CPUs rather than AI accelerators, and Graviton is the strongest CPU chip, with a 30% to 40% better price-performance advantage than alternatives. Customers saved more than $2 billion in Q3 simply by using our custom chips. We also just open-sourced TorchNeuron, enabling developers to run their existing PyTorch code directly on Trainium without rewriting any code (the vast majority of developers already use PyTorch).

With the launch of new-generation chips, demand for the previous generation remains strong: after the Graviton 5 launch, backlog for previous-generation Graviton instances still doubled quarter over quarter; short-term capacity for previous-generation Trainium instances also sold at a premium to the first half of the year. Given such strong Trainium demand, customers keep encouraging us to make it available outside AWS. We plan to begin external sales by the end of 2027.

Overall, AWS capacity demand remains extremely strong. Customers are still at a very early stage with large-scale inference applications, but inference already accounts for more than half of AWS AI revenue and that share keeps rising; at the same time, our training business revenue is also substantial. We maintain our expectation of about $220 billion in capital expenditure for 2026, most of it for AWS; and we expect capital expenditure to increase significantly further in 2027, with a clearly visible path to attractive financial returns, as I explained on our last earnings call.

AWS's current annualized revenue run rate has reached $184 billion, but the opportunity remains enormous. With our continued pace of innovation, our continuously improving core and AI capabilities, enterprises' strong willingness to migrate their core and AI workloads to AWS, and the capacity we keep adding every year, we believe AWS can achieve growth of more than 40% in both 2027 and 2028, and become a business with more than $1 trillion in annual revenue within the next 5 to 10 years. We are at the very early stage of this inflection point, our strategy is working, and I believe this momentum will continue for the foreseeable future.

Retail business

I think some people have not yet given sufficient attention to the progress this business has made in recent years and the scale it has reached. I don't know of any other retailer that can reach Amazon's size (hundreds of billions of dollars) while achieving double-digit operating margins including advertising revenue, and growing at such a fast pace.

Year to date, unit sales have grown nearly 15% year over year, and that is on an extremely large base—more than 30 billion items sold last year. In absolute dollar terms, we added more than $90 billion in gross merchandise sales (GMV) over the past 12 months, growing clearly faster than other retail competitors, with quarterly GMV additions about twice those of the second-largest U.S. competitor, and we continue to gain share across retail segments.

In the past few years, both our North America and International segments have improved operating margins by nearly 10 percentage points, which has allowed us to continue reinvesting resources into improving the customer experience. Of course, all of this is still in the early stages—the vast majority of retail sales (nearly 80%) still occur in physical stores, and this landscape will gradually shift toward e-commerce in the future. We are actively driving this shift by rapidly building an efficient, fast fulfillment network that delivers an exceptional shopping experience for customers while generating substantial free cash flow and returns on invested capital.

Our strategic priorities—faster delivery, lower cost to serve, grocery and pharmacy, and AI shopping—reinforce one another, and all are currently in full swing. As we continue to invest in building our same-day delivery network, along with the rollout of Amazon Now and Amazon Prime Air ultra-fast delivery services, delivery speeds continue to set records. This quarter was our fastest third quarter ever worldwide, with more than 1.5 billion items delivered to Prime members on the same day or next day in the third quarter alone.

The rapid global expansion of Amazon Now is very encouraging, now covering more than 450 towns and cities, with the number of items ordered through Amazon Now up 10x year over year. We are on track to lower our cost to serve for the fourth consecutive year, which enables us to continue offering a wide selection of fresh groceries and everyday essentials at competitive prices.

Faster delivery is driving rapid growth in our grocery business (including fresh groceries and everyday essentials), with merchandise sales growing to nearly $250 billion over the past 12 months. In the third quarter, fresh groceries and everyday essentials grew about 3x as fast as our other categories, accounting for about half of all units sold. Customers who receive same-day delivery of fresh groceries purchase nearly 7x as many everyday essentials as customers who do not buy fresh groceries; customers who buy fresh groceries shop non-grocery categories such as home and apparel about 4x as often as non-grocery customers, and spend more than 2x as much in those categories.

Amazon Pharmacy is also growing rapidly: active customers are up more than 60% year over year, with growth accelerating in each quarter of 2026; repeat customers are up more than 50% year over year. JD Power has ranked Amazon Pharmacy the No. 1 mail-order pharmacy for the third consecutive year, and we continue to expand our service capabilities, now offering three-hour prescription delivery in more than 350 cities.

In AI shopping, over the past 12 months, more than 400 million customers have used the shopping agent assistant Alexa, customer interactions with Alexa have grown 570% year over year, and U.S. customers who shop using Alexa spend 50% more per order than those who do not. Alexa helps users research products, compare options, view year-round price history, get personalized gift recommendations, and track deals and price drops on their behalf. We expect that, although horizontal agents will work alongside ours in the future, Alexa for Shopping will remain the primary shopping agent customers use when shopping on Amazon—always by your side, helping you browse, find inspiration, answer detailed questions on product pages, and, with knowledge of your purchase history, shipping address, and payment information, complete an order with just a word from you, which is extremely convenient and thoughtful for customers.

Advertising Business

Advertising revenue was $19.6 billion in the third quarter, up 12% year over year. Growth this quarter was affected by several timing factors:

First, we moved Prime Day in most countries to the second quarter, which means a large amount of sales and advertising revenue that would have occurred in the third quarter of 2025 shifted to the second quarter of 2026, creating a tailwind to Q2 year-over-year growth and a headwind to Q3, with an approximately 850 basis point sequential impact on the advertising business's Q3 year-over-year growth.

Second, the second quarter of 2026 included the first season of NBA on Prime, whereas last year's Q2 did not, and there is no NBA in the third quarter either (the NBA will begin airing in the fourth quarter).

Third, in the third quarter of 2025, we adjusted our search engine marketing program to direct more traffic to Amazon search pages, which provides customers with a better experience while allowing us to monetize that traffic at higher rates. This began driving faster advertising revenue growth a year ago, and in the third quarter we began facing tougher comparisons against that effect.

We expect advertising year-over-year growth to recover to the high-teens percentage range in the fourth quarter.

As for whether external horizontal agent shopping will materially impact Amazon's advertising business, the answer is no. Currently, very few users shop this way, the experience is still immature, and it is far less convenient than going directly to a retailer to research or placing an order with one click.

In the third quarter, the advertising team launched a large number of innovative products, the most important of which is Amazon Ads Agent, which integrates our complete advertising product suite on an AI-centric platform, leveraging trillions of shopping, browsing, and streaming signals to support advertisers at every step—from generating insights and answering questions to providing optimization recommendations—and to execute the corresponding actions on advertisers' behalf. We also launched full-funnel campaign capabilities, integrating Sponsored Ads, Display Ads, Video Ads, and Streaming TV Ads into a single campaign. Advertisers only need to provide an initial budget and flight duration, and Amazon AI will handle the planning, execution, and continuous optimization of everything else.

Other Business Highlights

Prime Video has become a significant business, with annual revenue run rate in the tens of billions of dollars, and it is already profitable on its own. We have launched hit series such as *Carrie*, *The Summer I Turned Pretty*, *The Boys*, *Off Campus*, *Doctor Who*, and *Fallout*, as well as breakthrough theatrical films such as *Project Hail Mary*, and we also have highly competitive live sports content, including Thursday Night Football (whose 2026 season opener was the best ever), NBA on Prime, NASCAR, the Celebrity Game, and UEFA Champions League soccer matches.

We have just launched our strongest hardware lineup to date, spanning Echo, Kindle, Alexa tablets, Fire TV, Ring, and Blink, with 34 new products in total. Although we are still in the early stages of the launch, momentum is very strong, and we look forward to more customers experiencing these products during the holiday season.

Finally, Amazon Leo commercial service remains on track to begin initial service this year. With re:Invent and the holiday season approaching, we are entering one of the most anticipated phases of the year. Thank you to all Amazonians for your hard work and relentless innovation, which make customers' lives better and easier every day.

Amazon is building a business landscape unlike any other. Never in history has there been a company with annual revenue of one trillion dollars, and Amazon is on a path to having two trillion-dollar businesses at the same time—AWS and retail. This is extremely rare, and we are excited about the future.

I will now hand it over to Brian.

Remarks by Brian Olsavsky, Senior Vice President and Chief Financial Officer

Revenue Overview

Worldwide revenue was $201.9 billion, up 12% year over year, or up 13% excluding the impact of foreign exchange. Worldwide operating income was $28.1 billion, with an operating margin of 13.9%, the highest in history.

North America and International Segments

First, a reminder that the timing of this year's Prime Day sales events changed: most major markets, including the U.S., moved the event to the second quarter, while all of 2025 Prime Day was in the third quarter. Excluding the impact of the Prime Day timing shift, year-over-year growth in the third quarter of 2026 would be about 400 basis points higher. In the third quarter, customer engagement was strong, continuing the growth trend from the first half.

North America: Third-quarter revenue was $112.9 billion, up 6% year over year. International: Third-quarter revenue was $43.1 billion, up 8% year over year excluding the impact of foreign exchange. Worldwide paid units: up 12% year over year. The Prime membership program remains a key pillar of our business, and hundreds of millions of Prime members are responding positively to our broad selection, low prices, and record delivery speeds. Over the past year, Prime members' monthly orders of fresh groceries through Same-Day Delivery have grown more than threefold.

Profitability

North America: Operating income was $8.7 billion, with an operating margin of 7.7%. International: Operating income was $1.6 billion, with an operating margin of 3.7%. In our fulfillment network, the team overcame transportation cost pressures from fuel inflation and higher line-haul freight rates. Despite these headwinds, in the third quarter we still reduced our overall cost to serve per unit through measures such as optimizing inventory placement, shortening delivery distances, and consolidating packages. Delivery costs increased 11% year over year, while units grew 12%; excluding the impact of fuel and line-haul freight inflation, delivery costs grew 8%, about 400 basis points lower than unit growth.

Robotics and AI technologies are delivering quantifiable gains in productivity and employee safety, and we expect them to compound over time. In 2026, we will double the number of robotic arms, represented by Cardinal and Sparrow, and we are developing robotic foundation models to accelerate robot programming and progressively teach robots to perform increasingly complex physical tasks, with the goal of deploying practical robots at scale. Looking ahead, we see significant opportunities to substantially increase productivity across our global fulfillment network, continue reducing cost to serve per unit, and keep accelerating delivery speeds. While our operating margin may fluctuate from quarter to quarter, we are committed to driving sustained long-term improvement in cost to serve per unit.

AWS Segment

AWS third-quarter revenue was $46 billion, up 39.3% year over year, with contributions from both core services and AI services. Sequentially, revenue increased by $3.8 billion, and the annualized revenue run rate has reached $184 billion. Customers continue to migrate to the cloud and expand their use of AWS core services; at the same time, customers seeking to fully benefit from AI capabilities are accelerating their migration to the cloud. We observe a significant positive correlation between AI spending and core services revenue growth—as customers increase their AI investments, consumption of core services rises accordingly. As more AI workloads move into production, this correlation effect is expected to strengthen further, driving continued expansion of demand for compute, storage, and networking.

AWS operating income was $17.8 billion, with an operating margin of 38.7%, reflecting strong growth momentum and the results of our ongoing efforts to improve operational efficiency. Investments in custom chips and custom networking, along with software and process improvements, are helping us optimize capacity allocation and build more cost-effective infrastructure. As we have long said, AWS operating margin will fluctuate with the level of investment in different periods. AI products are gaining increasing customer recognition, and AI revenue achieved triple-digit year-over-year growth.

Capital Expenditures

Third-quarter capital expenditures were $75.9 billion, primarily for AWS and AI infrastructure to support strong customer demand. We firmly believe that AI is a massive opportunity that is expected to drive long-term growth in revenue and free cash flow, as well as excellent returns on invested capital.

Fourth-Quarter Financial Guidance

Net sales: Expected to be between $240 billion and $247 billion. Foreign exchange impact: Based on current exchange rates, expected to create an approximately 60 basis point headwind to revenue growth. Operating income: Expected to be between $28.5 billion and $33.5 billion. One point requiring special note regarding fourth-quarter guidance: Amazon Leo commercial service is scheduled to launch initial service this year, and we expect to begin capitalizing some related costs, including production and launch costs, starting in the first quarter of next year. This timing is later than previously forecast on our earnings calls, so most Leo-related costs will continue to be expensed in the fourth quarter.

We are well prepared for this holiday shopping season and to support the millions of customers building their businesses on AWS. Thank you to the entire Amazon team for their hard work and intense focus on customers. We firmly believe that continuously improving the customer experience is the only reliable path to creating lasting value for shareholders.

We will now move to the Q&A session.

Q&A Session

Question 1: Justin Post (Bank of America) Q: The cloud business acceleration is encouraging, but sequential growth has slowed somewhat—what factors are behind this? Also, Brian, how does fuel cost affect the fourth-quarter guidance?

Andy Jassy's answer (AWS growth):

We are very pleased with the 39.3% growth rate—this is a business with an annualized revenue run rate of $184 billion, and achieving this growth rate is truly remarkable; it is the fastest in nearly five years and the sixth consecutive quarter of acceleration. The factors driving growth are multifaceted.

On AI, we have a complete technology stack from the bottom to the top: if you want to build your own models, SageMaker AI is the easiest way; if you want to call leading frontier models, Bedrock offers high security and highly competitive price-performance; if you want to build your own agents, we have Strands and Agent Core; if you want to use out-of-the-box agent services, products such as Kiro (code writing), Connect (call center), Transform (migration), and Q (AI work assistant) are all rapidly gaining customer adoption. In addition, we are also well positioned in chips. For enterprise customers using AI at scale, especially after deploying more inference into production environments, they are eager to optimize price-performance, and Trainium has a significant advantage in price-performance compared with other alternatives. At the same time, as AI applications roll out broadly, workloads such as reinforcement learning, post-training, and tool calling are generating increasingly strong demand for core compute, and Graviton, as a leading CPU chip, offers about 30% to 40% better price-performance than other x86 solutions. Together, these factors are driving significant business growth.

The capacity added each quarter varies, but on an annualized run rate of $184 billion, achieving year-over-year growth of more than 40% in 2027 and 2028 is quite substantial. We believe AWS has the potential to become a trillion-dollar business within 5 to 10 years—an extremely rare opportunity.

Brian Olsavsky's answer (fuel costs):

Regarding the impact of fuel and line-haul transportation inflation, I do not have a precise figure, but this cost will be slightly higher in the fourth quarter than in the third quarter. Earlier this year, we added a fuel surcharge to shipping fees for FBA sellers, which covers part of the cost, with the remainder reflected in retail inbound and outbound transportation. This cost will be slightly higher in the fourth quarter than in the third quarter.

Question 2: Brian Nowak (Morgan Stanley) Q: What is the basis for the judgment that AWS can reach a trillion dollars in revenue? Also, what are the key milestones for robotics technology in 2027?

Andy Jassy's answer (AWS path to a trillion dollars):

Look at the growth rate and revenue run rate, and then combine that with our forecasts for 2027 and 2028—we have good visibility into large customers (large-capacity reservations of hundreds of megawatts and even millions of watts), and such customers must lock in capacity in advance, so we have a relatively clear grasp of demand for the next eight to ten quarters. Layer these growth rates on top of the current revenue scale, and the numbers climb rapidly.

Nor do I think current usage will fade anytime soon—the process of large-scale adoption of inference in production applications by enterprise customers has only just begun. One factor is inference cost; for enterprise customers to be able to use inference in the way we envision, costs must continue to come down, which is one of the core reasons we built Trainium custom chips. Over the past two years, inference costs have fallen substantially, and I expect they will continue to fall in the future.

Security will be a huge application scenario—the vulnerabilities currently discovered by frontier models are ones humans had previously failed to find, and some argue that a considerable share of token consumption in the next one to two years may be used solely for security. This is a major emerging use case on top of already mature application scenarios such as code writing. Add to that the continued growth of AI assistants (whether in consumer scenarios or enterprise work scenarios, such as Amazon Q), and my view that almost every modern enterprise production application will eventually be rewritten as an agentic version—because that is the inevitable path to moving fast and effectively guarding against security risks. Looking ahead, there are simply too many imaginable AI application scenarios. So, while it is impossible to predict precisely what things will look like in 5 to 10 years, we believe it is not impossible for AWS to become a trillion-dollar-plus business; rather, that is the direction we genuinely see.

Andy Jassy's answer (robotics milestone):

Amazon currently has deployed more than 1 million robots in its fulfillment network, and they have had a significant impact on the cost structure, productivity, and employee safety—safety metrics in our fulfillment network are at historic lows, and we have made tremendous progress over the past four or five years, with robotics playing a major role. Nevertheless, whether in systems already deployed or in technologies under development, we are still at a very early stage.

I believe that in the future many repetitive tasks in our fulfillment centers will be handled by robots, and we are also investing in robots of various forms, including robots expected to help complete "the last 100 yards" of delivery to customers' homes. The criterion by which we evaluate robotics investments is: whether it can deliver comparable or better productivity, safety levels, cost structure, and delivery speed—when the economics work, we will deploy at scale in the fulfillment network. In the design of our next-generation fulfillment centers, including the expansion of same-day delivery centers, robots account for a very high proportion. We believe robotics has played an important role to date, but compared with its impact over the next five years, this is merely the tip of the iceberg.

Question 3: Mark Mahaney (Evercore ISI) Q: Will AWS's margin structure be relatively stable over the next two years? How is grocery coverage progressing, and what is its pull-through effect on Prime member spending?

Brian Olsavsky's answer (AWS margin outlook):

We do not provide separate guidance on AWS margins, but we can offer some context. The high margin of 38.7% this quarter is the combined result of years of effort in operational discipline, efficiency gains, capacity optimization, strict control of fixed costs, and a continuously rising share of custom chips—it is not accidental. However, there are multiple volatile factors: AI's hypergrowth puts pressure on overall margins, because the AI business currently has lower margins than the overall business; there are also cost headwinds from component pricing. Although AI business margins are improving (as Andy said, already ahead of where the core business was at a similar stage of development), they are still below the core business. Overall, margins will always be determined by multiple factors together, and we will continue to focus on cost and efficiency and manage the margin dilution during the investment phase of new products.

Andy Jassy's answer (grocery business):

We are no less excited about grocery than you are. We have long operated a large grocery business, but previously it mainly focused on products in the middle of the food aisles—consumables, canned foods, pharmaceuticals, and beauty products. We own Whole Foods Market, a leading organic grocery retailer whose growth rate ranks among the top in the grocery category, performing well on both revenue and profit, and it is advancing new formats and an urban "everyday shopping" model, with a good response.

We have always known that to serve customers in the way we envision, we must have mainstream supply in mass categories, and the starting point for all of this is grocery. From the perspective of overall household consumption, the biggest gap between the product categories we can offer and the amount customers actually spend with us is precisely in grocery, and a large share of grocery spending starts with fresh food.

That is precisely why we are excited about the progress we have made in delivering fresh groceries through our same-day warehouse system. The everyday essentials business is growing at about three times the rate of our other categories and now accounts for roughly half of all units sold. Same-day customers who buy fresh groceries purchase about 7 times as many everyday essentials units as customers who do not buy fresh groceries. This is highly significant for closing the gap between "total customer spending potential" and "actual spending on Amazon." At the same time, grocery customers buy non-grocery categories such as home and apparel about 4 times as often as non-grocery customers, and spend about 2 times as much in those categories. These differences are significant, and we believe that continued progress in fresh groceries will ultimately enable us to win customers' first-choice position across more shopping occasions.

Question 4: Eric Sheridan (Goldman Sachs) asked: What is the market opportunity for Bedrock? How does the decision to sell Trainium externally balance internal use and external sales?

Andy Jassy's answer (Bedrock market opportunity):

Our experience serving technology customers over the past 20-plus years tells us that there is no "single tool that rules everything." Whether in databases or analytics, the early leaders were never the only option. The same is true in AI—initially everyone thought TensorFlow was the only framework, but PyTorch is now more mainstream. The model space will be no exception.

I believe that in the next two to three years there will be six to eight comparable mainstream models coexisting, each with its own strengths in dimensions such as reasoning, code, vision, latency, and cost, and each able to meet more than 95% of workload needs. Most customers will use multiple models at the same time, and many will even mix multiple models within the same generative AI application.

This is exactly what customers really need: a service that supports broad model choice, offers excellent price-performance, and has strong security and access controls, so they can switch flexibly as the capabilities of each model rise and fall. This is exactly what Bedrock provides. We have both closed-source models such as Anthropic and OpenAI and open-source models such as Kimi and GLM, and I believe the Western world also has strong demand for more Western open-source models, a space that will continue to expand in the future. Bedrock's market opportunity is extremely large, and it has the potential to become an AWS pillar business comparable in scale to EC2.

Andy Jassy's answer (Trainium internal-external balance):

This is a question we are still continuously evaluating. Currently, there are really only two companies that have both a significant cloud business and chips strong enough to sell externally, and the scale and complexity of managing this balance are something most companies will never encounter.

All advanced chips are currently produced by the same foundry, capacity is limited, and each company has an allocation quota and needs to decide the proportion allocated to external sales versus internal cloud deployment.

The logic of this trade-off is: external sales allow revenue to be recognized more quickly in the short term (revenue is recognized when chips ship); internal deployment (in the form of EC2 instances or Bedrock), although the revenue recognition cycle is longer, can generate revenue continuously over five or six years, ultimately bringing more absolute revenue and free cash flow. So this requires careful weighing.

Our current approach differs from that of the other company facing the same choice. As mentioned earlier, we plan to launch external sales of Trainium by the end of 2027, and we may re-examine this balance every year in the future. Both channels will be important parts of the business. Our ability to build cost-effective, scalable custom chips will become a highly strategic competitive advantage—both optimizing our own cost structure and building an independent external chip sales business; and external chip sales and cloud infrastructure are essentially two very different business models.

Question 5: Doug Anmuth (JPMorgan) asked: What scale of capital expenditure is needed to support growth of more than 40% over the next two years and beyond? How are the payback periods for AWS servers, networking, and data centers changing?

Andy Jassy's answer:

First on payback periods—we are indeed continuing to shorten them. Different categories of assets (AI versus traditional computing, different financing methods, etc.) differ, but overall we have consistently made good progress in extending useful lives and shortening payback periods, and we are continuing to advance this.

As for the scale of capital expenditure needed to support the above growth, the answer is: quite substantial. But we believe this is a historic opportunity that comes once in a generation. On top of an annualized revenue run rate that has already reached $184 billion, growing at 39% and then continuing to grow at 40% per year for the next two years, ultimately becoming a business of more than a trillion dollars—in order to maintain and consolidate our leadership position, we will invest firmly. If such an acceleration opportunity could be spread over 25 years, the capital expenditure pressure would be smaller, but reality is not like that; the opportunity is right in front of us. We have the ability to support this plan, and it will permanently change Amazon's scale—not only revenue scale but also free cash flow, and we will use this to continue to maintain our lead.

Question 6: Ross Sandler (Barclays) asked: As the token market on Bedrock (inference as a service) expands and most workloads run on Trainium, how will this business transition affect AWS's revenue growth, especially margins?

Andy Jassy's answer:

You are right that most workloads on Bedrock do indeed run on Trainium, largely because we are very pleased with Trainium's price-performance—it enables us to keep prices as low as possible for customers while maintaining the profit levels we expect.

Looking at the cost structure of the cloud computing industry from a broader perspective, there are several key factors worth noting.

First is how data centers are held: long-term purchase or long-term lease offers operating margins about 3,500 basis points better than using colocation or neocloud providers.

Second is the hardware level: if you use the same hardware as everyone else, your cost structure is comparable to the market average; if you can use differentiated hardware, you gain a cost advantage.

Most of our capacity is owned or long-term leased, which gives us better operating margins than competitors operating on a colocation or neocloud model. Combined with the fact that we run on our own custom chips, this brings a structural cost advantage both in price-performance for customers and in our own economics. I believe these advantages will continue to benefit customers and help improve AWS's overall economics.

Thank you all for participating in today's conference call. A replay of this call will be available on our investor relations website for at least three months. Thank you for your interest in Amazon, and we look forward to speaking with you again next quarter. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investors bear responsibility for their own investment decisions based on this.