Amazon.com (AMZN.US) reported third-quarter 2026 revenue of $201.9 billion, up 12% year over year, with operating income rising 61% to $28.1 billion, according to the company's earnings call.
Chief Financial Officer Brian Olsavsky said the company's operating margin reached a record 13.9%. AWS revenue grew 39.3% year over year, with an annualized run rate of $184 billion and a backlog of $578 billion.
While AI and cloud demand are accelerating growth, the company's third-quarter cash capital expenditures reached $75.9 billion, with full-year spending expected at about $220 billion. Management expects spending to rise significantly in 2027.
On the earnings call, management described this infrastructure investment cycle as aimed at expanding revenue, free cash flow, and return on invested capital over the long term, while noting that AI services currently carry lower margins than the core cloud business and that component pricing remains a pressure point.
AWS accelerates on AI and core cloud demand
CEO Andy Jassy said the results were mainly driven by AWS's continued accelerating growth, rising retail unit sales, and ongoing investment in AI infrastructure. The company also forecast fourth-quarter net sales of $240 billion to $247 billion and operating income of $28.5 billion to $33.5 billion.
Specifically, AWS revenue totaled $46 billion in the quarter, up 39.3% year over year and up $3.8 billion sequentially. The cloud unit's annualized revenue run rate reached $184 billion. AWS operating income was $17.8 billion, with an operating margin of 38.7%.
Jassy said AWS growth has accelerated for a sixth consecutive quarter and reached its fastest pace in nearly five years. He cited strong demand for both traditional cloud computing and AI services, and noted that AWS's backlog reached $578 billion, with triple-digit year-over-year growth.
"Customers are buying up compute almost as soon as we deploy it," Jassy said, adding that Amazon deploys an AI accelerator roughly every 14 seconds.
The company said Amazon Bedrock, its managed service for accessing and building applications on foundation models, has become the fastest-growing billion-dollar service in AWS history. According to Jassy, Fortune 500 companies' spending on Bedrock rose nearly sevenfold year over year, while total token processing grew nearly 17-fold.
Amazon recently added OpenAI models to the platform, as well as managed versions of Kimi, GLM, and SpaceX models.
Jassy also highlighted demand for the company's in-house chips. He said revenue commitments for Graviton and Trainium grew more than 20-fold year over year, and customers using Amazon's custom chips saved more than $2 billion in the quarter. Amazon plans to make Trainium available outside AWS starting in late 2027.
Looking ahead, Jassy said Amazon expects AWS to grow more than 40% in both 2027 and 2028, and believes the business could become a trillion-dollar annual revenue operation within five to 10 years. He said the outlook reflects visibility into large-scale customer compute commitments, continued enterprise cloud migration, and growing production use of AI inference.
Capital expenditures rise with AI infrastructure buildout
Amazon's third-quarter cash capital expenditures totaled $75.9 billion, mainly for AWS and AI infrastructure. The company continues to expect full-year 2026 cash capital expenditures of about $220 billion, with most going to AWS, and Jassy said spending is expected to increase substantially in 2027.
Management said it views this investment cycle as a long-term opportunity to expand revenue, free cash flow, and return on invested capital. Jassy said the company continues to shorten the time needed to break even on infrastructure investments, though he noted a difference in returns between core cloud and AI infrastructure.
Olsavsky said AWS profitability benefits from capacity optimization, fixed-cost management, in-house chips, and networking investments. But he noted that AI services currently have lower margins than AWS's core business, creating structural pressure on overall margins. Component pricing also remains a headwind.
Jassy said Amazon owns or long-term leases most of its data center capacity, and combined with the use of in-house chips, this gives it a structural cost advantage compared with cloud providers that rely more on colocation or "neocloud" arrangements.
Retail margins, faster delivery, and grocery expansion
Amazon's North America revenue rose 6% to $112.9 billion, while international revenue rose 8% to $43.1 billion excluding currency effects. Global paid unit sales grew 12%. Results were affected by the timing of Prime Day. This year the event took place in the second quarter in most major markets, whereas in 2025 it fell entirely in the third quarter.
Olsavsky said excluding this timing shift, third-quarter revenue growth would have been about 400 basis points higher. North America operating income totaled $8.7 billion, with a 7.7% margin; international operating income was $1.6 billion, with a 3.7% margin.
Amazon said that despite higher fuel and line-haul costs, it lowered overall fulfillment costs through better inventory placement, shorter shipping distances, and package consolidation. Shipping costs rose 11% year over year, while unit sales grew 12%. Olsavsky said excluding fuel and line-haul inflation, shipping costs rose 8%, about 400 basis points below unit growth.
Jassy said Amazon delivered more than 1.5 billion items to Prime members via same-day or next-day delivery in the quarter. The company's ultra-fast delivery service, Amazon Now, is now available in more than 450 towns and cities, and the number of items ordered through the service grew tenfold year over year.
Grocery and everyday essentials remained a focus. Growth in fresh grocery and everyday essentials was about three times that of the rest of Amazon's stores, accounting for about half of unit sales in the quarter. Management said customers who use same-day fresh delivery buy about seven times more everyday essentials than those who do not.
Advertising, pharmacy, and fourth-quarter outlook
Amazon Ads third-quarter revenue was $19.6 billion, up 12% year over year. Jassy said growth was affected by the shift of Prime Day into the second quarter, the timing of NBA programming on Prime Video, and the comparison base from last year's search engine marketing changes. He said Amazon expects fourth-quarter advertising revenue growth to rebound to close to 20%.
Jassy said Amazon Pharmacy customers grew more than 60% year over year, with repeat customers up more than 50%. The company now offers three-hour prescription delivery in more than 350 cities.
For the fourth quarter, Amazon expects foreign exchange to reduce net sales growth by about 60 basis points year over year. Olsavsky also said fuel and line-haul inflation in the fourth quarter is expected to be higher than in the third quarter.
Commercial service for Amazon Leo remains on track to begin initial operations before the end of the year. The company expects to begin capitalizing some Leo production and launch costs in the first quarter of 2027; as a result, it expects to continue expensing most Leo-related costs in the fourth quarter.
Full transcript
(1) Operator
Thank you for standing by. Good day, everyone, and welcome to Amazon's third-quarter 2026 financial results teleconference. (Operator instructions) This call is being recorded. At this time, I would like to turn the call over to Dave Fildes, Vice President of Investor Relations. Thank you. Please go ahead.
(2) Dave Fildes 鈥?Vice President of Investor Relations 鈥?Opening remarks
Hello, and welcome to our third-quarter 2026 financial results teleconference. Joining us today to answer your questions are our CEO, Andy Jassy, and our CFO, Brian Olsavsky. As you listen to today's conference call, we encourage you to have our press release in front of you, which includes our financial results, metrics, and commentary for the quarter. Please note that unless otherwise stated, all comparisons in this call are against results for the same period in 2025. Our remarks and responses to your questions reflect management's views as of today, October 29, 2026, and include forward-looking statements. Actual results could differ materially. For more information about factors that could affect our financial results, please refer to today's press release and our filings with the SEC, including our most recent annual report on Form 10-K and subsequent filings. During this call, we may discuss certain non-GAAP financial measures. In our press release, the slides accompanying this webcast, and our filings with the SEC, each of which is posted on our Investor Relations website, you will find additional disclosures about these non-GAAP measures, including reconciliations to comparable GAAP measures. Our guidance incorporates the order trends we have seen to date and the assumptions we believe are appropriate at this time. Our results are inherently unpredictable and may be materially affected by many factors, including fluctuations in foreign exchange rates and energy prices, changes in global economic and geopolitical conditions, tariff and trade policies, resource and supply volatility, including memory chips, customer demand and spending, including the impact of recession fears, inflation, interest rates, regional labor market constraints, world events, the rate of growth of the internet and online commerce, cloud services, and emerging technologies, and the various factors detailed in our filings with the SEC. Our guidance assumes, among other things, that we do not conclude any additional business acquisitions, restructurings, or legal settlements. It is not possible to accurately predict demand for our goods and services, and therefore our actual results could differ materially from our guidance. With that, I will turn the call over to Andy.
(3) Andrew Jassy 鈥?President, Chief Executive Officer, and Director 鈥?Keynote remarks
Thank you, Dave. We reported revenue of $201.9 billion, up 12% year over year. Operating income was $28.1 billion, up 61%. The third quarter was another strong quarter for Amazon. On AWS, our plan is working. Despite the large amount of capacity we added in the year-ago period, AWS grew 39.3% year over year, accelerating for the sixth consecutive quarter and marking our fastest growth in nearly five years. Backlog reached $578 billion, growing triple digits year over year, and is highly diversified. We install an AI accelerator roughly every 14 seconds, and customers consume it almost as soon as we install it. Our core compute and storage revenue saw its fastest growth in several years. Our core business has long had strong operating margins, and now AI margins are improving quickly, clearly ahead of where the core business was at a similar stage of development. Despite rising component costs, including for memory, and significant depreciation from our capital expenditures, we continue to accelerate payback on our capital spending and are more confident than ever that our AWS investments will produce a much larger revenue and free cash flow business with strong returns on invested capital than we could have imagined two or three years ago. We are creating a very attractive flywheel in AWS. To borrow Jim Collins' definition of a flywheel in "Good to Great" 鈥?if you do A, you will almost certainly do B; if you do B, you will almost certainly do C, and so on. At AWS, leading features, security, and operational performance inevitably drive customers to run applications and store data on AWS; that inevitably drives AI demand for AWS, because customers want their production inference close to the rest of their data and applications; that in turn drives core AWS demand, because AI needs core services for reinforcement learning, post-training, tool use, and similar work; and that in turn inevitably drives reinvestment in our broader AWS capabilities. This is different from primarily hosting GPUs or short-term data center capacity leases. This is a long-term platform strategy that broadly helps customers invent and evolve customer experiences, operate reliably, and build durable, multiyear partnerships. It also gives us highly attractive long-term economics whether or not supply is scarce. There are many examples of AWS progress, but I will share just a few. On the core business, it is important to remember that the vast majority of global IT spending, about 85%, is still on premises, and we are in a 10-to-20-year cycle of that shifting to the cloud. Conventional wisdom held that critical workloads like mainframes or storage area networks would never move to the cloud. But these workloads are moving because AWS has built unique cloud alternatives and AI-driven migration capabilities that make the transition easier. Enterprises gain lower capital expenditures, lower operating costs, faster innovation, higher developer productivity, and better data accessibility, making it easier to take advantage of AI. Our core business margins are attractive because we have learned over the past 20 years how to operate efficiently. As the core business grows, we continue to reinvest in core AI capabilities. For example, as customers generate more data and want value from analytics and AI, moving data between data storage and analytics engines is cumbersome and expensive. Through the acquisition of DuckLabs, DuckDB will become the universal SQL dialect for AWS storage and analytics, simplifying data processing with a modern analytics engine 鈥?fast, easy to use, and ubiquitous, so customers can write a SQL query once and run it on any AWS analytics service. In analytics, for the first time, customers will have zero migration or portability cost between query engines. This is a big deal. You can also imagine this will help agents query efficiently. On AI, we continue to expand the capabilities of our inference service Amazon Bedrock, which is driving significant growth 鈥?Fortune 500 customer spending on Bedrock grew nearly sevenfold year over year, and overall Bedrock token volume grew nearly 17-fold year over year. Bedrock is the fastest-growing billion-dollar service in AWS history. We recently brought OpenAI models to Bedrock, and usage is growing rapidly. We also just launched Bedrock-hosted versions of Kimi, GLM, and SpaceX models. You cannot talk about AI today without talking about security. That is another reason customers continue to choose Bedrock. We designed Bedrock so customer data stays secure and private. Inputs and outputs are never shared with model providers and are never used to train foundation models. When customers fine-tune foundation models, we create private copies dedicated solely to them, and data is fully encrypted in transit and at rest. As frontier models become more powerful, they are discovering previously undiscovered software vulnerabilities at unprecedented speed. We built AWS Continuum to discover, prioritize, validate, and remediate code vulnerabilities at machine speed. Customers including Capital One, IBM, Infor, Luminate, and Robinhood are already adopting it. In the third quarter, we announced integration into coding tools 鈥?Anthropic's Claude Code, OpenAI's Codex, and AWS's Kiro. Customer interest in this new security service has exceeded anything we have seen. Agent deployment is proceeding at an accelerating pace, much of it built on Bedrock's agent core building blocks. Tens of thousands of customers now use AgentCore, and the number of agents deployed in the third quarter was twice the total of the prior three quarters. In addition to building their own specialized agents, most companies will use off-the-shelf agent services, and we offer several that customers love. Kiro for coding, Amazon Connect for contact center agents, Amazon Transform for migration. They are all iterating quickly and winning customer adoption. The same is true for Amazon Quick, our intelligent AI work companion, which is growing rapidly, with large enterprises such as Mondelez, EchoStar, 3M, Southwest Airlines, and DXC choosing Quick because it is built for work and uncompromising on customer data. Finally, our chip business continues to grow triple digits year over year, with revenue commitments for Graviton and Trainium each up more than 20-fold year over year. AI continues to be a tailwind for Graviton because post-training reinforcement learning and agent tool use run mainly on CPUs rather than AI accelerators, and Graviton is the most powerful CPU chip, with 30% to 40% better price-performance than comparable products. In the third quarter alone, customers using our custom chips saved more than $2 billion. We also just open-sourced TorchNeuron, enabling developers to run the PyTorch code most of them already use on Trainium without rewriting any code. As we launch new generations of chips, demand for previous generations remains strong. After launching Graviton 5, backlog for the previous generation of Graviton instances still more than doubled quarter over quarter. Compared with the first half, short-term capacity for the previous generation of training instances sold at a premium. Because training demand is so strong, customers continue to encourage us to offer this capability outside AWS, and we plan to do so by the end of 2027. Overall demand for AWS capacity remains exceptionally strong. Enterprises are still very early in using inference at scale, yet inference already accounts for more than half of AWS AI revenue and is still growing. We also have substantial training revenue. We still expect 2026 cash capital expenditures of about $220 billion, most of it for AWS, and expect to invest more in 2027. As I said on our last call, we have a clear plan for strong financial returns. AWS is a business with an $184 billion annualized revenue run rate, but the opportunity remains extraordinary. With our continued pace of innovation, new core and AI capabilities, enterprises' desire to run their core and AI workloads on AWS, and the capacity we add each year, we believe AWS can grow more than 40% in each of 2027 and 2028 and become a $1 trillion revenue business within the next five to 10 years. We are very early in this inflection point, our strategy is working, and I expect it to continue working for the foreseeable future. On the retail (stores) business, I think some underestimate the progress that business has made in recent years and the scale it has now reached. As far as I know, no other retail business has been able to grow at this pace while reaching Amazon's scale 鈥?hundreds of billions of dollars, with double-digit operating margins including advertising. Year to date, unit sales have grown nearly 15% year over year, on top of a huge base of more than 30 billion items sold last year. In absolute dollars, we have added more than $90 billion in gross sales over the past 12 months. We are growing significantly faster than other retail competitors, adding about twice as much gross sales in the U.S. each quarter as the second-largest competitor, and we continue to gain retail share. Over the past few years, our North America and international segment operating margins have each expanded by nearly 10 percentage points, which allows us to reinvest to make the customer experience better. Remember, it is still early. The vast majority of retail, nearly 80%, still happens in physical stores. Over time, this will eventually shift to e-commerce. We are driving that shift by rapidly building an efficient, fast fulfillment network that delivers a delightful customer experience while generating substantial free cash flow and return on invested capital. From faster delivery to cost to serve, to grocery and pharmacy, to AI shopping, each of our priorities is a growth multiplier, and we are firing on all cylinders. With our investments in the same-day delivery network and ultra-fast services Amazon Now and Amazon Prime Air, our delivery speeds continue to increase. We had the fastest third quarter globally in our history, delivering more than 1.5 billion items to Prime customers via same-day or next-day delivery in the third quarter. We are encouraged by customer enthusiasm for the rapid global rollout of Amazon Now, which is now available in more than 450 towns and cities, and the number of items ordered through Amazon Now grew 10-fold year over year. We are on track to reduce cost to serve for the fourth consecutive year, which allows us to continue to meet or beat competitors' prices and offer a broad selection of fresh food and everyday essentials cost-effectively. Faster speeds are driving our grocery business, including fresh and everyday essentials, with total sales over the past 12 months growing to nearly $250 billion. In the third quarter, growth in fresh food and everyday essentials was about three times the growth of the rest of the stores business, accounting for about half of total units sold. Customers who buy fresh groceries with same-day delivery buy nearly seven times more everyday essentials than those who do not. Active grocery customers buy non-grocery categories such as home and apparel nearly four times more often than non-grocery customers and spend more than twice as much in those categories. Amazon Pharmacy is also growing rapidly. Customers using it grew more than 60% year over year, accelerating each quarter in 2026, and repeat customers grew more than 50% year over year. J.D. Power recently ranked Amazon Pharmacy the best mail-order pharmacy for the third consecutive year. We continue to improve, now offering three-hour prescription delivery in more than 350 cities. Over the past 12 months, more than 400 million customers used our agentic shopping assistant Alexa for Shopping, and customer interactions grew 570% year over year. U.S. customers who use Alexa for Shopping spend 50% more per order than those who do not. It researches products, compares options, checks a full year of price history, provides personalized gift recommendations, and tracks deals and price drops for customers. While we expect to work with horizontal agents, we believe Alexa for Shopping will become the primary shopping agent customers use when shopping on Amazon. Having a shopping assistant with you as you browse, get inspiration, and view detail pages, one that can place an order directly based on past purchases because it already knows your history, shipping address, and payment information, is a uniquely convenient experience for customers. Turning to Amazon Ads. We generated $19.6 billion in revenue in the third quarter, up 12% year over year. Despite some timing anomalies last quarter, we are pleased with the overall momentum across our advertising portfolio. I will mention three factors affecting growth. First, we moved Prime Day in most countries to the second quarter, shifting substantial sales and advertising revenue that would have been in the third quarter of 2025 into the second quarter of 2026 鈥?a tailwind to our second-quarter 2026 growth and a headwind to the third quarter of 2026. For advertising, this created an 850 basis point sequential impact on third-quarter year-over-year growth. Second, the second quarter of 2026 included the first season of NBA on Prime, which was not in the year-ago second quarter, while the third quarter included no NBA, with NBA starting in the fourth quarter. Third, in the third quarter of 2025, we adjusted our search engine marketing program to direct more traffic to search pages on Amazon. This was a better customer experience, and we monetized that traffic at higher rates and began boosting advertising revenue a year ago 鈥?and now we are beginning to lap that. We expect fourth-quarter year-over-year growth to return to the high teens. A natural question is whether agentic shopping from external horizontal agents is significantly affecting Amazon advertising. The answer is no. Few customers currently shop heavily this way because the experience is not yet good enough and is not more convenient than going directly to a retailer to research or ordering with one click. Our advertising team had a very productive third quarter, and we expect positive impact in the coming months, led by Amazon Ads Agent 鈥?which brings our full advertising offering under a single AI-first platform, supported by trillions of shopping, browsing, and streaming signals, giving advertisers insights, answers, and the ability to optimize and act on their behalf at every step. We also launched full-funnel campaigns in Amazon Ads Agent, combining sponsored ads, display, video, and streaming TV into a single campaign. Advertisers provide initial inputs, including budget and flight duration, and Amazon's AI plans, executes, and continuously optimizes the rest. We continue to see momentum in several other areas. I will mention just a few. Prime Video has become a significant business, with revenue run rate in the tens of billions of dollars, and it is already profitable on its own. We have strong momentum with hit series such as "Carrie," "The Summer I Turned Pretty," "The Boys," "Off-Campus," "Reacher," and "Fallout," as well as blockbuster films such as "Project Hail Mary," plus highly attractive live sports, including the best-opening Thursday Night Football in 2026, NBA on Prime, NASCAR, the Masters, and UEFA Champions League soccer. We recently launched our most powerful device lineup ever, with 34 new products across Echo, Kindle, Alexa tablets, Fire TV, Ring, and Bee. It is still early, but we are off to a very strong start and look forward to more customers getting their hands on these products as the holiday season arrives. Finally, Amazon Leo remains on track to begin offering initial service this year. We are about to enter the most exciting stretch of the year 鈥?re:Invent and the holiday season. I want to thank our teams for their hard work and innovation day in and day out to make customers' lives better and easier. What we are building at Amazon is unprecedented. There has never been a trillion-dollar revenue business. Amazon is on a path to have two of them, AWS and retail stores. This is very rare, and we are energized by what lies ahead. Over to Brian.
(4) Brian Olsavsky 鈥?Senior Vice President and Chief Financial Officer 鈥?Financial review
Thank you, Andy. Let's start with overall revenue performance. Worldwide revenue was $201.9 billion, up 12% year over year, or up 13% excluding currency effects. Worldwide operating income was $28.1 billion, with an operating margin of 13.9%, our highest ever. Let's look at North America and international segment results. First, a reminder that Prime Day timing changed this year, with the sales event occurring in the second quarter in most of our largest countries, including the U.S. Prime Day in 2025 fell entirely in the third quarter. Excluding the impact of the Prime Day timing shift, third-quarter 2026 year-over-year growth would have been about 400 basis points higher. In the third quarter, we saw strong customer engagement and a continuation of first-half trends. North America third-quarter revenue was $112.9 billion, up 6% year over year. International segment revenue was $43.1 billion, up 8% year over year excluding currency effects. Worldwide paid units grew 12% year over year. Prime remains a key pillar of our business. Hundreds of millions of Prime members are responding to the combination of our vast selection, low prices, and record delivery speeds. Over the past year, the number of fresh grocery orders placed by Prime members each month for free same-day delivery more than tripled. Next, profitability. North America segment operating income was $8.7 billion, with a 7.7% margin. International segment operating income was $1.6 billion, with a 3.7% margin. In the fulfillment network, teams dealt with higher transportation costs from fuel inflation and higher line-haul rates. Even with these headwinds, we lowered overall cost to serve in the third quarter through better inventory placement, shorter shipping distances, and package consolidation. Transportation costs grew 11% year over year, while units grew 12%. Excluding fuel and line-haul inflation, transportation costs grew 8%, about 400 basis points below unit growth. Robotics and AI are delivering measurable gains through productivity and employee safety, and we expect those gains to compound in the future. In 2026, we doubled our fleet of robotic arms such as Cardinal and Sparrow. We are also developing robotic foundation models to enable faster programming and teach robots increasingly complex physical tasks, with the goal of delivering useful robots at scale. Looking ahead, we see significant opportunity to improve productivity and lower cost to serve across the global fulfillment network while delivering products to customers faster. Our operating margins will fluctuate quarter to quarter, but we are focused on continued long-term improvement in cost to serve. Next, the AWS segment. Revenue was $46 billion, up 39.3% year over year, driven by both core and AI services. We added $3.8 billion in revenue sequentially. AWS now has an annualized revenue run rate of $184 billion. Customers continue to migrate to the cloud and expand their use of AWS core services. And increasingly, customers who want the full benefits of AI are accelerating their migration to the cloud. We see a strong correlation between AI spending and growth in our core services revenue. As customers invest in AI, we see corresponding increases in core consumption, and we expect this relationship to strengthen as more AI workloads move into production and drive new demand for compute, storage, and networking. AWS operating income was $17.8 billion, with a 38.7% operating margin, reflecting strong growth and our focus on efficiency. Our investments in custom chips and custom networking, along with software and process improvements, are helping us optimize capacity and build more cost-effective infrastructure. As we always say, we expect AWS operating margins to fluctuate, partly based on our level of investment at any point in time. Our AI offerings are gaining customer adoption, with AI revenue growing triple digits year over year. Bedrock has been a significant growth driver. Now turning to our cash capital expenditures, which were $75.9 billion in the third quarter. This is primarily related to AWS and AI, as we invest to support strong customer demand. We continue to invest heavily, especially in AI, because we believe this is a huge opportunity that can generate long-term revenue and free cash flow as well as return on invested capital. Finally, I will close with fourth-quarter financial guidance. Fourth-quarter net sales are expected to be between $240 billion and $247 billion. Based on current exchange rates, we estimate the year-over-year impact from currency changes will be about a 60 basis point headwind in the quarter. Fourth-quarter operating income is expected to be between $28.5 billion and $33.5 billion. I will mention one factor affecting fourth-quarter guidance. Commercial service for Amazon Leo is on track to begin initial service this year, and we expect to begin capitalizing some costs, including production and launch costs, in the first quarter of next year. This is later than I previously predicted on prior earnings calls. As a result, we will continue to expense most Leo-related costs in the fourth quarter. We are ready to serve customers this holiday season and support the millions of customers building on AWS. I want to thank Amazon teams for their hard work and customer-obsessed focus. Our commitment to improving the customer experience remains the only reliable path to creating long-term shareholder value. Let's move to the Q&A portion.
Q&A session
(1) Q1 鈥?Bank of America Securities analyst Justin Post
Question: Andy, it's great to see the cloud business continue to accelerate, very strong. But sequential growth slowed, and I want to understand whether there are factors on the customer side or in token costs. And how do you think about capacity additions in the fourth quarter? Also Brian, what is the impact of fuel costs on fourth-quarter guidance?
Andrew Jassy response: Okay, I will answer the growth question first. We are very pleased with this growth 鈥?39.3% growth for a business with an $184 billion annualized revenue run rate is quite rare. This is our fastest growth in nearly five years and the sixth consecutive quarter of acceleration. We are also adding a lot of capacity. I think if you look at what is driving growth today, it is a combination of things. First, on AI, we have extraordinary full-stack capabilities 鈥?whether you are building your own models with SageMaker AI, the easiest way to build your own models, or you want access to leading frontier models with high security and highly competitive price-performance, you can use Bedrock. Whether you are building your own agents, where we offer capabilities such as Strands and AgentCore to build and run them securely and at scale, or using off-the-shelf agents, which is what most companies choose, just look at Kiro for coding, Connect for contact centers, Transform for migration, and Quick, our fast-growing AI work companion. People are adopting these very quickly. We add substantial capacity every quarter. The capacity added each quarter is not identical. But if you look at the capacity we have added this year and our plans for roughly the next eight quarters, we typically can see eight to 10 quarters out, we have a lot of capacity coming. On top of that, add the impact of chips. We are very well positioned in the AI wave because it is increasingly true that most customers 鈥?every customer using AI at scale 鈥?especially enterprises now beginning to deploy inference at scale in production, are extremely eager to save on price-performance and training, and Trainium's significant price-performance advantage over alternatives is highly attractive. As more people use AI in very broad ways, this drives substantial demand for core capabilities such as reinforcement learning, post-training, and tool use, and we have the leading CPU chip in Graviton, with about 30% to 40% better price-performance than other x86 providers. These factors are really driving very significant growth. The capacity we add each quarter will vary, but at this scale, forecasting more than 40% year-over-year growth in each of 2027 and 2028 is significant, and we believe this can become a $1 trillion revenue business within five to 10 years. This is an extraordinary and rare opportunity, and we are excited to pursue it.
Brian Olsavsky addition: Justin, on the fuel and line-haul inflation question, I do not have an exact number, but the fourth quarter will be higher than the third quarter. Remember that earlier this year we introduced a fuel surcharge for FBA sellers, recovering part, but not all, of the fuel costs we saw for those shipments and for remaining partners in retail inbound and outbound freight. So it is included and slightly higher than the third quarter.
(2) Q2 鈥?Morgan Stanley analyst Brian Nowak
Question: I have two questions. Andy, your enthusiasm for AWS has come through on the call. You rarely make comments like "$1 trillion in revenue in five to 10 years," which is beyond the usual range. Can you explain how you get to that $1 trillion? Is it top-down or bottom-up? What gives you confidence that you can achieve that kind of revenue growth and reach that number in five to 10 years? Second, on robotics, looking ahead to 2027, what key milestones are you focused on in robotics for the retail business?
Andrew Jassy response: On AWS, Brian, just look at the growth rates 鈥?look at the growth rate of the run rate, and then look at our forecasts for 2027 and 2028. As I have said many times, no matter how much capacity we provide, the large blocks of capacity customers acquire, hundreds of megawatts, sometimes up to a gigawatt, are reserved in advance, because you cannot leave a gigawatt of capacity idle. So we have very clear visibility into those large capacity needs for 2027 and 2028 and some of 2029. We will always reserve a meaningful amount of capacity for on-demand use on EC2 or Bedrock, and we work closely with customers to ensure they have ample capacity available in AWS. But we have very clear visibility into what AWS will look like over the next few years, and when we do the math on revenue and the existing run rate, those growth rates quickly push the number up. On top of that, I do not think much of the usage we see today will decay soon. Enterprises are still very early in broadly deploying inference in production, really early. Part of that is because enterprises need enough people who know how to operate AI, do post-training, and fine-tune models for the right applications; part of it is inference cost. I think if we want enterprises to use inference the way we believe they will, we must keep working to lower inference costs. This is a major mission for AWS and one of the important reasons we build our own custom chips and Trainium 鈥?we strongly believe it is extremely important to let customers use inference more economically. Inference costs have fallen substantially over the past two years, and I think they will fall substantially further. Then think about what you are seeing. Take security as an example. On its own, models are now so advanced that they find vulnerabilities humans never noticed. If you can use these models, you can use them to find vulnerabilities and ultimately help patch and deploy. Some believe that over the next year or two, a meaningful portion of total tokens consumed may ultimately be for security, which is a relatively new use case on top of the coding growth we are seeing and various AI companions in consumer and work spaces, such as Quick. These are only the first few large use cases. Add to that the enterprise trend you will see 鈥?I believe almost every modern enterprise's production applications will be rewritten to be agentic, because if you want to move quickly, or protect against some security risks in the right way, you have to deploy that way. The AI wave is simply too big. So you cannot precisely predict the next five to 10 years, especially with these kinds of growth rates, but for us it is not an exaggeration to imagine this could become a business of more than $1 trillion. On robotics, what I would say is that what is happening in robotics right now is quite remarkable 鈥?in general, and especially at Amazon. We now have more than 1 million robots in our fulfillment network. They have had a very significant impact on our cost structure, productivity, and safety 鈥?look at the safety data in our fulfillment network, which is at historic lows. We have made tremendous progress over the past four or five years, and robotics is an important part of that. But I would say we are still early, both in what we have built and deployed and in what we are developing. I think you will see that our fulfillment centers will always have a large number of employees. But I think much of the more repetitive work currently done by humans will be done by robots. We are investing in robots of various form factors. We also have an opportunity to help with the "last 100 yards" of delivery to customer stores. We have a lot coming. For us, the test for robotics is: does it deliver equal or better productivity, safety, cost structure, and delivery time? When the economics work, we will deploy at scale in the fulfillment network. If you look at the new designs coming to our fulfillment network, including the expansion of same-day facilities, they are heavily invested in robotics. So we think robotics has already had a huge impact, and this is just the tip of the iceberg compared with the next five years.
(3) Q3 鈥?Evercore ISI analyst Mark Mahaney
Question: Two things. First, thank you very much for the AWS growth outlook. Could you also give an outlook on margins? I know margins will fluctuate quarter to quarter, but over the next two years, do you think the current margin structure is relatively stable? Will we stay at AWS operating margins in the 30s or 40s? Also on fresh checkout, I am super excited to see this opportunity landing in my community and elsewhere. How broad is your fresh checkout coverage? In other words, what percentage of Prime households do you think it covers? For customers who have used it, how much do you think it increases spend per Prime household? Is it 20%, 50%, or 100%? Given how large the fresh grocery category is?
Brian Olsavsky response: Mark, I will start with the AWS operating margin outlook. We do not provide guidance on AWS margins. But I can give you my view. This quarter's operating margin of 38.7% looks very strong. I would add that the profitability you have seen recently is not a random result, but the product of a lot of disciplined efficiency improvements, capacity optimization, and tight control of fixed costs. As Andy said, over time, more custom chips enter our cost structure. However, there are other factors at play. In fact, the outsized growth of AI puts pressure on overall business margins because margins there are lower. We are pleased with the progress in both core and AI margins. As Andy said earlier, they are ahead of where the core business was at a similar stage, but still lower. So there is this mix issue. There is also the headwind from component pricing that we discussed. So it is not predictable, but there will always be multiple factors in play, and we are working very hard on the cost and efficiency side while managing new product margins as they ramp, launch, and reach scale.
Andrew Jassy addition: On fresh checkout, Mark, I am as excited as you are. As you know, we have been in the large grocery business for some time. It has mainly been centered on the middle shelves of a grocery store 鈥?household goods, canned food, drugstore items, beauty products. We own the leading organic grocery store Whole Foods Market, which is also growing faster than the rest of the grocery business, in both revenue and profit, and the daily shop format in new formats and urban environments is performing very well, and we plan to expand it in the coming years. We have always known that if we want to serve customers the way we believe they want and the way we want, we must offer mass products, and that starts with fresh. If you look at total household spend, when we compare the total set of goods we think consumers and customers could buy from us with what they actually spend with us, the biggest gap is grocery. And a lot of shopping, a lot of grocery, starts with fresh. So we are really excited because we believe we have found the right approach with fresh delivery from same-day facilities. Our everyday essentials growth is already about three times the rest of the stores business and now accounts for about half of our total units. We find that customers who buy fresh through same-day delivery buy about seven times more everyday essentials than customers who do not. So if you are trying to close the gap between total consumer spend and the products we can offer, fresh has an enormous impact on that gap. We also see that our grocery customers buy non-grocery categories such as home and apparel about four times more often and spend about twice as much in those categories than customers who do not buy groceries from us. These are huge differences, and we believe the progress we are making in fresh and grocery will ultimately make us more relevant to customers in more shopping occasions.
(4) Q4 鈥?Goldman Sachs analyst Eric Sheridan
Question: Two questions about the AWS market opportunity. First, on Bedrock, what are your key insights into the Bedrock market opportunity? How will it continue to expand in the coming years? Second, on the decision to sell Trainium externally, how do you think about balancing the market opportunity from external sales with the compute needed for that opportunity, as well as balancing it with internal use cases and other AWS opportunities?
Andrew Jassy response: On Bedrock, you may have heard us say this many times over the past two years. But more than 20 years of delivering technology to customers has taught me that no single tool rules the world. People always assume the early leader will be the only thing used, but that has never been true. It was true in databases, and it was true in analytics. In the early days of AI, everyone said the only framework was TensorFlow, but now PyTorch is much bigger. The same is true for models. Over the next two to three years, I think there will be six to eight models that are roughly comparable, or at least good enough to cover more than 95% of workloads. Some will be better at reasoning, some at coding, some at vision, some with lower latency, and some with lower cost. Most customers will use multiple models, and many will use multiple models in the same generative AI application. So what customers really want 鈥?and we have repeatedly confirmed this in conversations with customers 鈥?is a service with broad choice across these frontier models, good price-performance, and very strong security and access controls. They want this service so they can decide what mix of models to use and switch back and forth as models leapfrog one another and application constraints change. That is exactly what we see on Bedrock. People are very excited about it. We have closed models such as Anthropic and OpenAI, as well as open models such as Kimi and GLM. I think the Western world is very eager for more Western open models, and I believe that will happen over time. But I think the opportunity for Bedrock is enormous and could become a service as large as EC2, which is already a very large service within AWS. So I think it is very large. On the chip balance 鈥?how much to sell externally versus use internally in the cloud 鈥?that is a great question, and one we continue to evaluate. I would say there are really only two companies that have both a significant cloud business and enough chips to sell externally. So there are not many customers that truly have to face this kind of question at this scale. Right now, everyone manufactures their advanced chips at the same foundry. That foundry can only provide so many chips each year, and everyone gets an allocation. You have to decide how much of your allocation to sell externally and how much to use for your own cloud. There is a trade-off. If you sell externally, you can recognize more revenue in the short term because you recognize revenue as soon as the chip ships. Conversely, if you decide to use it inside the cloud as EC2 instances or Bedrock, you actually drive more absolute revenue and free cash flow from those chips because you can use them continuously for five to six years, generating revenue over a longer period. So you really have to think through what to do. To date, I would say we have taken a different approach to chip usage than the other provider facing the same balance question. As I mentioned earlier, we believe selling chips externally will be a very meaningful business, and we will start by the end of 2027. We may rethink this balance every year. But both will be very important. Our ability to build cost-effective, scalable chips will be a very significant strategic advantage at scale 鈥?both in our own cost structure and as a standalone business selling chips externally 鈥?and selling chips externally is indeed a very different business from cloud infrastructure.
(5) Q5 鈥?JPMorgan analyst Douglas Anmuth
Question: When you consider AWS revenue growth of more than 40% over the next two years, and the multiyear path after that, can you talk about the capital expenditures needed to drive that revenue growth in the coming years and beyond? Also, Andy, do you have any updated thoughts on the payback period for servers, networking, and data centers?
Andrew Jassy response: Okay, I will answer in reverse order. On infrastructure breakeven, as I mentioned in my opening remarks, we continue to move it to the left 鈥?to make it shorter. Core and AI are a bit different, and the financing approach differs depending on which part you are talking about, but we continue to make progress. As we have always been good at in AWS history, we keep finding ways to extend useful life and accelerate breakeven, and we are making good progress there. On the capital expenditures needed to grow in the way we described, it is quite substantial. I think, Doug, our view on this is that this is a once-in-a-generation opportunity. The AI we are seeing is the biggest transformation of our lifetime. If we can grow 39% on an $184 billion annual run rate and then grow another 40% on top of that each year for the next two years, ultimately becoming a business of more than $1 trillion, we will invest capital to continue leading. We are already a leader today in infrastructure and AI. If this acceleration and opportunity were spread over 25 years, you could spread capital expenditures over 25 years, would that be more convenient? Yes, maybe it would be more convenient, but that is not the world we live in, and that is not the opportunity. We are able to fund this plan. I think it will permanently change Amazon's scale, permanently change our revenue, and permanently change our free cash flow. We will invest capital and pursue it relentlessly to continue to be as important a leader as we are today.
(6) Q6 鈥?Barclays analyst Ross Sandler
Question: Okay, I want to dig a little deeper into Bedrock. The token marketplace on Bedrock seems to be one of the hottest services in AI right now. From what we understand, most of these workloads run on Trainium. As it becomes a more important part of the overall AWS story 鈥?as you just mentioned, Andy 鈥?how will this affect revenue growth, and especially AWS margins, as you move from selling bare metal to selling inference more as a service? Can you talk about that?
Andrew Jassy response: Yes. You are right that most of what is on Bedrock today runs on Trainium. One important reason we do this is that we really like the price-performance we get, which both allows us to keep prices as low as possible for customers and gives us the margin that comes with it. If you think about margins in the cloud and look at the cost structures of different players, there are several points worth noting. First is data centers, how much they buy or long-term lease rather than use colocation or neocloud. Neocloud operating margins are about 3,500 basis points worse than long-term leases. Second, look at the hardware they run on. If you use the hardware everyone else uses, your cost structure will be comparable to others. If you have the ability to use different hardware, you have an advantage. So I think look at what we are doing 鈥?most of our capacity is owned or long-term leased, which gives us better operating margins than relying mainly on colocation or neocloud, or having a significant portion in colocation or neocloud. And we run on our own chips, which both provide advantageous price-performance and are more cost-effective for us because we own them 鈥?I think this gives us some structural cost advantages, which will be very beneficial to customers when we pass savings back to them, and very beneficial to AWS overall economics.
(7) Dave Fildes
Thank you, everyone, for joining today's call and for your questions. A replay 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.
Comments