AI computing power sells out immediately! Amazon (AMZN.US) announces in Q3: Capital expenditures will increase again next year, betting on long-term returns

According to Zhitong Finance APP, Amazon (AMZN.US) achieved Q3 2026 revenue of $201.9 billion, a year-on-year increase of 12%. Operating profit increased by 61% to $28.1 billion. CFO Brian Olsavsky stated that the company’s operating profit margin reached a record high of 13.9%. AWS revenue grew 39.3% year-on-year, with an annualized run rate of $184 billion and outstanding orders of $578 billion.

While AI and cloud demand have accelerated business growth, the company's Q3 cash capital expenditure reached $75.9 billion, with full-year expectations around $220 billion. Capital expenditures are anticipated to rise significantly in 2027. Management stated during the earnings call that this infrastructure investment cycle aims to expand revenue, free cash flow, and return on invested capital in the long term, though the current profit margin from AI services is lower than core cloud business and component prices remain a headwind.

AWS Accelerates Growth Driven by AI and Core Cloud Demand

CEO Andy Jassy said that performance was primarily driven by continuous acceleration of Amazon Web Services (AWS), growth in retail sales, and ongoing investments in AI infrastructure. The company also forecasts Q4 net sales of $240–247 billion and operating profit of $28.5–33.5 billion.

Specifically, AWS revenue in the quarter totaled $46 billion, up 39.3% year-on-year and an increase of $3.8 billion quarter-on-quarter. The annualized revenue run rate for this cloud segment reached $184 billion. AWS operating profit stood at $17.8 billion, with an operating profit margin of 38.7%.

Jassy noted that AWS’s growth rate had accelerated for six consecutive quarters, reaching the fastest level in nearly five years. He mentioned strong demand for both traditional cloud computing and AI services, with AWS's outstanding orders amounting to $578 billion and a triple-digit annual growth rate.

“Almost as fast as we deploy compute, customers purchase it immediately,” Jassy commented, adding that Amazon deploys an AI accelerator roughly every 14 seconds.

The company reported that Amazon Bedrock—its managed service for accessing foundational models and building applications on top of them—has become the fastest-growing multibillion-dollar service in AWS history. According to Jassy, Fortune 500 companies’ spending on Bedrock grew nearly sevenfold year-over-year, with total token processing volume up almost 17-fold. Amazon has recently added managed versions of OpenAI models as well as Kimi, GLM, and SpaceX models to the platform.

Jassy also emphasized demand for the company’s proprietary chips. He stated that revenue commitments for Graviton and Trainium chips have increased more than 20-fold year-on-year, with customers saving over $2 billion this quarter by using Amazon-designed chips. Amazon plans to offer Trainium outside AWS starting at the end of 2027.

Looking ahead, Jassy stated that Amazon expects AWS to grow over 40% per year both in 2027 and 2028, and envisions this segment could become a $1 trillion annual business in five to ten years. He said this outlook is based on visible large-scale customer compute commitments, ongoing enterprise cloud migration, and increasing production usage of AI inference.

Capital Expenditure Rises with AI Infrastructure Construction

Amazon’s Q3 cash capital expenditure totaled $75.9 billion, primarily devoted to AWS and AI infrastructure. The company maintains its forecast for full-year 2026 capex at around $220 billion, with the bulk allocated to AWS, and Jassy said spending is expected to increase significantly in 2027.

Management described this investment cycle as a long-term opportunity to expand revenue, free cash flow, and return on invested capital. Jassy stated that the payback period for infrastructure investments is being continually shortened, though returns differ between core cloud and AI infrastructure.

Olsavsky noted that AWS profitability benefits from capacity optimization, fixed cost management, proprietary chip, and network investments. However, he pointed out that AI services currently have lower margins than AWS’s core business, structuring a downward pressure on overall profit margins. Component prices also remain a negative factor.

Jassy highlighted that Amazon owns or has long-term leases on most of its datacenter capacity and, combined with using proprietary chips, this gives a structural cost advantage over cloud providers who rely more on colocation or “neocloud” arrangements.

Improved Retail Profit Margins, Faster Delivery, and Grocery Expansion

Amazon’s North America revenue grew 6% to $112.9 billion, while international revenue grew 8% to $43.1 billion on a currency-neutral basis. Global paid unit sales rose 12%.

Results were affected by the scheduling of Prime Day. This year, the event took place in most major markets during Q2, while in 2025 it will fall entirely in Q3. Olsavsky observed that removing this timing change, Q3 revenue growth would have been approximately 400 basis points higher year-on-year.

North America operating profit totaled $8.7 billion with a 7.7% margin; international operating profit was $1.6 billion with a 3.7% margin. Amazon stated that despite increases in fuel and line-haul transportation costs, the company reduced overall fulfillment costs through inventory placement optimization, shorter shipping distances, and package consolidation.

Shipping costs increased 11% year-on-year, while units shipped grew 12%. Olsavsky mentioned that, excluding fuel and line-haul inflation, shipping costs grew 8%, about 400 basis points lower than unit growth.

Jassy noted that this quarter Amazon delivered over 1.5 billion items to Prime members via same-day or next-day shipping. The company’s ultra-fast delivery service, Amazon Now, is now available in over 450 towns, with orders through the service increasing tenfold year-on-year.

Groceries and daily essentials remain a focus, with growth in these segments roughly triple the rest of Amazon’s stores, accounting for about half of this quarter’s unit sales. Management says that customers who use same-day fresh delivery buy about seven times more daily essentials than those who don’t.

Advertising, Pharmacy, and Fourth Quarter Outlook

Amazon Ads reported Q3 revenue of $19.6 billion, up 12% year-on-year. Jassy explained that growth was influenced by the timing of Prime Day moving to Q2, NBA programming on Prime Video, and a comparable base due to last year’s search engine marketing project adjustments. He said Amazon expects Q4 advertising revenue growth to recover to high teens, nearly 20%.

Jassy noted that Amazon Pharmacy customers grew over 60% year-on-year, and repeat customers grew over 50%. The company now offers three-hour prescription delivery in more than 350 cities.

Looking to Q4, Amazon expects currency exchange rates to reduce net sales growth by approximately 60 basis points year-on-year. Olsavsky also indicated that Q4 fuel and line-haul transportation inflation is expected to be higher than in Q3.

The commercial service for Amazon Leo remains on track to launch initial operations by year-end. The company expects to begin capitalizing some Leo production and launch costs in Q1 2027; thus, most Leo-related costs will continue to be expensed in Q4.

Full Transcript

(1) Moderator (Operator)

Thank you for your patience. Hello, everyone, and welcome to Amazon’s Q3 2026 earnings call. (Operator prompt) This meeting is being recorded. For opening remarks, Mr. Dave Fildes, Vice President of Investor Relations. Thank you, please proceed.

(2) Dave Fildes — VP, Investor Relations Opening Statement

Hello, everyone, and welcome to our Q3 2026 earnings call. Joining us today to answer questions are our CEO Andy Jassy and CFO Brian Olsavsky. As you listen to today’s call, we recommend you have our press release on hand, which contains this quarter’s financial results, metrics, and commentary. Note that unless otherwise indicated, all comparisons are against Q3 2025 results.

Our remarks and Q&A responses represent management’s views as of today (October 29, 2026) and include forward-looking statements. Actual results may differ materially. For more factors that could affect our financial results, please refer to today’s press release and filings with the U.S. Securities and Exchange Commission (SEC), including our most recent Form 10-K and subsequent filings.

During this call, we may discuss certain non-GAAP financial measures. In our press release, our webcast slides, and our SEC filings (all on our investor relations website), you’ll find supplemental disclosures about these non-GAAP measures, including reconciliations to comparable GAAP measures. Our guidance incorporates observed order trends to date and what we currently believe are appropriate assumptions. Our results are inherently unpredictable and may be materially impacted by many factors, including foreign exchange and energy price fluctuations, changes in global economic and geopolitical conditions, tariffs and trade policies, resource and supply issues (including memory chips), customer demand and spending (including recession fears), inflation, interest rates, regional labor market constraints, world events, internet and online commerce, cloud services and emerging technology growth rates, and other factors detailed in our SEC filings.

Our guidance assumes (among other things) we will not engage in any additional business acquisitions, restructurings, or legal settlements. It is impossible to accurately predict demand for our goods and services, so actual results may materially differ from guidance.

Now, Andy, please proceed.

(3) Andrew Jassy — President, CEO & Director Keynote Speech

Thank you, Dave. Reported revenue was $201.9 billion, up 12% year-on-year. Operating profit reached $28.1 billion, a 61% increase. It was another strong quarter for Amazon. For AWS, our plans are paying off. Despite significant capacity additions a year ago, AWS still delivered 39.3% year-on-year growth, accelerating for six straight quarters and hitting its fastest rate in nearly five years. Backlog reached $578 billion, up triple digits year-on-year with high diversification.

We’re installing an AI accelerator roughly every 14 seconds, and customers are consuming them almost as soon as they’re deployed. Core compute and storage revenue grew at their fastest pace in years. Our operating margins in the core business have long been strong, and AI services are now improving rapidly—outpacing what core service profitability was at a similar stage.

Despite higher costs for components like memory and significant capex-related depreciation, we continue to accelerate capex payback and are more confident than ever: our AWS investments will deliver revenue, free cash flow, and returns on invested capital far exceeding what we could have imagined two or three years ago.

We’re building an extremely attractive flywheel in AWS. Borrowing Jim Collins’ flywheel definition from “Good to Great”—if you do A, you almost inevitably do B; if you do B, you then do C, and so forth.

In AWS, leading functionality, security, and operational performance drive customers to run applications and store data on AWS; this drives AI demand (since customers want inference close to their other data and apps); which drives core AWS demand (as AI needs core services for reinforcement learning, post-training, and tool invocation); which in turn facilitates reinvestment in our broader AWS features.

This is different from merely hosting GPUs or short-term colocation leases. This is a long-term platform strategy to help customers innovate, stabilize operations, and build sustainable partnerships for years to come—delivering highly attractive long-term economics, regardless of supply constraints.

There are many examples of AWS progress. In core business, remember the majority (about 85%) of global IT spend is still on-premises, and we’re in a 10–20 year transition cycle toward the cloud. Traditionally, it was thought that critical workloads like mainframes or SANs would never move to the cloud. But these workloads are migrating because AWS built unique cloud alternatives and AI-driven migration tools that make this transition easier. Enterprises benefit from lower capex, lower operating costs, faster innovation, greater developer productivity, and better data accessibility, making AI easier to leverage.

Our core margins are attractive because we’ve learned how to operate efficiently over 20 years. As core grows, we continuously reinvest in core AI features. As customers generate more data and seek value from analytics and AI, data movement between storage and analytics engines becomes complex and costly. With our DuckLabs acquisition, DuckDB will become AWS’s universal SQL dialect for storage and analytics—fast, easy to use, ubiquitous, and enabling customers to write a single query to run on any AWS analytics service.

In analytics, customers can now have zero migration or portability cost between query engines for the first time—this is significant. You can imagine this will also help agents perform efficient queries.

On AI, we keep expanding inferencing services via Amazon Bedrock, which has driven significant growth—spending from Fortune 500 companies on Bedrock has grown nearly sevenfold year-on-year, and token volume nearly 17-fold. Bedrock is the fastest-growing multibillion-dollar service in AWS history. We recently brought OpenAI models to Bedrock, whose usage is also growing fast. We've just launched managed versions of Kimi, GLM, and SpaceX models on Bedrock.

It’s impossible to talk AI today without mentioning security—a recurring reason customers choose Bedrock. We designed Bedrock to keep customer data secure and private. Inputs and outputs are never shared with model providers or used to train foundational models. When a customer fine-tunes a base model, we create a private copy just for their use, with data fully encrypted in transit and at rest.

As frontier models get stronger, they discover previously unknown software vulnerabilities faster than ever. We built AWS continuum to detect, triage, verify, and fix code vulnerabilities at machine speed. Customers like Capital One, IBM, Infor, Luminate, and Robinhood are adopting it. In Q3, we announced integration with coding tools—Anthropic’s Claude Code, OpenAI’s Codex, and AWS’s Kiro. Customer interest in this new security service is the strongest we've ever seen.

Agent deployment is accelerating, with many built atop Bedrock agent core modules. Tens of thousands of customers now use AgentCore, and agents deployed in Q3 were double the total for the previous three quarters combined. In addition to building their own agents, most companies use out-of-the-box agent services from us, which are popular. Kiro (coding), Amazon Connect (contact center agents), and Amazon Transform (migration) are iterating rapidly and gaining traction. Our intelligent AI work companion, Amazon Quick, is also growing fast, with big customers like Mondelez, EchoStar, 3M, Southwest Airlines, and DXC choosing Quick for its work focus and uncompromising handling of customer data.

Finally, our chip business continues triple-digit year-on-year growth, with revenue commitments for Graviton and Trainium each up over 20 fold. AI continues to benefit Graviton as most post-training reinforcement learning and agent tool invocation run on CPUs—not AI accelerators—and Graviton is the most powerful CPU chip, offering 30–40% better price/performance than peers.

Just in Q3, customers using our custom chips saved over $2 billion. We’ve also open sourced TorchNeuron, so developers can run most existing PyTorch code on Trainium with no rewrites. Even as we launch new chips, demand for previous generations remains strong—after launching Graviton 5, commitments for the prior Graviton instances doubled sequentially. Short-term capacity for older training instances is selling at a premium compared to earlier this year.

Given explosive training demand, customers keep urging us to offer this capability outside AWS—we plan to do this by the end of 2027. Overall AWS capacity demand remains exceptionally strong. Enterprises are still at a very early stage of scaled inference; already, inference is over half our AWS AI revenue and still increasing. We have substantial training revenue as well. For 2026, we still expect $220 billion in cash capex, with most for AWS, and more investment forecast for 2027. As I said last call, our path to robust returns is clear.

AWS is an $184 billion annualized revenue business, but the opportunity is still remarkable. Given our pace of innovation, new core and AI features, enterprises wanting to run their core and AI workloads on AWS, and annual net capacity adds, we believe AWS can grow over 40% annually in 2027 and 2028 and become a $1 trillion annual business in five to ten years. We’re at the very early stage of this inflection, our strategy is working, and I expect it will keep bearing fruit into the foreseeable future.

In retail (stores), I think some underestimate the progress and current scale. As far as I know, no other retail business has reached Amazon’s scale—hundreds of billions, double-digit operating margin with ads—and continues to grow this fast. So far this year, shippable items are up nearly 15% year-on-year, on top of an enormous base of over 30 billion items shipped last year. Over the past 12 months, we added $90 billion in total sales—our growth outpaces rivals, and in the U.S., our quarterly sales adds are about twice our nearest competitors, steadily taking market share.

In the past few years, North America and International segment operating margins each improved by almost 10 percentage points, allowing us to reinvest and further enhance the customer experience. Remember, it’s still early—most retail (nearly 80%) still happens in physical stores. Over time, this will shift toward e-commerce. We’re driving this by rapidly building out an efficient, fast fulfillment network that offers a superior customer experience while delivering strong free cash flow and return on capital investment.

From faster delivery, to unit service cost, to groceries and pharmacy, to AI shopping—each priority is a growth amplifier, and we’re firing on all cylinders. Our investment in same-day delivery, Amazon Now, and Amazon Prime Air ultra-fast services continues to drive faster shipping. We just had the fastest Q3 in history, delivering over 1.5 billion items to Prime customers as same-day or next-day deliveries. Global customer enthusiasm for Amazon Now is very encouraging—it's live in 450 cities, and Amazon Now orders are up 10x year-on-year.

We’re on track to reduce unit service cost for the fourth straight year, letting us match or beat competitive prices and efficiently offer a rich assortment of fresh groceries and daily essentials. Faster speed is driving our grocery business (fresh and essentials), whose total 12-month sales are now nearly $250 billion. In Q3, grocery and essential growth was about three times that of the store’s other businesses, accounting for about half of all units sold. Customers purchasing fresh products for same-day delivery buy about seven times more daily essentials than those who don’t. Active grocery shoppers buy home, apparel, and other non-grocery products about four times more often than non-grocery shoppers, and spend over twice as much in those categories.

Amazon Pharmacy is also growing quickly. Customer count is up more than 60% year-on-year, with growth accelerating in all 2026 quarters and repeat customers up over 50%. J.D. Power recently named Amazon Pharmacy the best mail-order pharmacy for the third straight year. We keep making improvements and now offer three-hour prescription delivery in over 350 cities.

Over the past year, over 400 million customers engaged with our agent-shopping assistant Alexa for Shopping, and engagement grew 570% year-on-year. U.S. customers who use Alexa for Shopping spend 50% more per order than non-users. Alexa handles research, compares options, reviews a year’s price history, gives personalized gift suggestions, and tracks deals and price drops for customers. While we expect to work with horizontal agents, we believe Alexa for Shopping will be customers’ primary assistant for shopping on Amazon.

Browsing, finding inspiration, and shopping with a dedicated agent right at hand—able to order based on your purchase history, with your addresses and payment info ready—is a uniquely convenient experience for customers.

Looking at Amazon ads, we delivered $19.6 billion in Q3 revenue, up 12% year-on-year. Despite some timing anomalies last quarter, we’re pleased with momentum across our ad portfolio. Three key factors influenced growth: First, Prime Day moved in most countries to Q2, transferring significant Q3 2025 sales and ad revenue to Q2 2026—tailwind for Q2, headwind for Q3. For ads, this was an 850 bps sequential impact on Q3 year-on-year growth.

Second, Q2 2026 included the first NBA on Prime season (none in Q2 2025), while Q3 has no NBA (NBA resumes in Q4). Third, in Q3 2025, we adjusted our search engine marketing to drive more traffic to Amazon’s search pages. This improved customer experience and monetization at higher rates, boosting ad revenue a year ago—now, we lap that comparison. We expect Q4 growth to return to the high teens.

A natural question is whether agent shopping via external horizontal agents is significantly affecting Amazon ads. The answer: not currently. Few customers shop this way in volume, as the experience is not great, nor more convenient than going directly to retailers.

Our ad team delivered a slew of innovations in Q3 that should pay off in coming months, notably Amazon Ads Agent—a single AI-first platform integrating our full ad suite, leveraging trillions of shopping, browsing, and streaming signals to provide advertisers with insights, answers, and optimization at every step.

We also introduced full-funnel campaigns in Amazon Ads Agent, combining search, display, video, and streaming TV into a single campaign. Advertisers set initial inputs (including budget and duration), and Amazon’s AI plans, executes, and continually optimizes the rest.

Other business areas show ongoing momentum as well. To name a few: Prime Video has become a major business with an annual run rate of hundreds of billions, now profitable. Blockbuster shows and films like “Carrie,” “The Summer I Turned Pretty,” “The Boys,” “Off-Campus,” “Reacher,” “Fallout,” as well as “Project Hail Mary,” and highly appealing live sports—including our best-ever Thursday Night Football kickoff, NBA on Prime, NASCAR, Masters, and UEFA soccer—are driving strong performance.

We recently launched our most powerful device lineup yet, spanning Echo, Kindle, Alexa tablets, Fire TV, Ring, and Bee—a total of 34 new devices. It’s early days, but we’ve had a strong start and look forward to more customer adoption for the holidays. Lastly, Amazon Leo remains on schedule for initial commercial service this year.

We’re about to enter the most exciting period of the year—re:Invent and the holidays. I want to thank our teams for their relentless hard work and innovation to make customers’ lives better and more convenient. What we’re building at Amazon is unprecedented. No business has ever reached $1 trillion in revenue; Amazon is on track to have two (AWS and retail stores). That is rare, and we're excited for what's ahead.

Over to Brian.

(4) Brian Olsavsky — SVP & CFO Financial Report

Thank you, Andy. First, the overall revenue results: global revenue was $201.9 billion, up 12% year-on-year; on a currency-neutral basis, up 13%. Global operating profit was $28.1 billion, for a record 13.9% operating margin.

Regional results for North America and International: A quick reminder that this year’s Prime Day changed timing—sales happened in most large countries (including the U.S.) in Q2. In 2025, Prime Day falls entirely in Q3. Excluding the timing effect, Q3 2026 would have about 400 basis points higher year-on-year growth. We saw strong customer engagement in Q3, continuing trends from the first half. Q3 North America segment revenue was $112.9 billion, up 6%. International segment revenue was $43.1 billion, up 8% currency-neutral. Global paid units grew 12%.

Prime remains a key pillar. Hundreds of millions of Prime members respond to our combination of selection, low prices, and record shipping speeds. Over the past year, the number of free same-day grocery orders placed by Prime members doubled.

Profitability: North America operating profit was $8.7 billion, margin 7.7%. International operating profit was $1.6 billion, margin 3.7%. In fulfillment, the team managed rising shipping costs from fuel inflation and higher line-haul rates. Despite these headwinds, better inventory placement, shorter distances, and package consolidation lowered unit fulfillment cost in Q3. Shipping cost was up 11% year-on-year, while units shipped rose 12%. Excluding fuel and line-haul inflation, shipping cost rose 8%, about 400 bps below unit growth.

Robotics and AI are delivering measurable productivity and safety benefits, and we expect these to accumulate. In 2026, we doubled our fleet of robotic arms like Cardinal and Sparrow. We’re also developing baseline robot models for faster programming and teaching robots ever more complex physical tasks, aiming to deliver useful robots at scale. Going forward, we see huge opportunities to raise fulfillment productivity, lower unit cost, and deliver products faster globally. Our operating profit margin will fluctuate quarterly, but our focus remains steady long-term improvement in unit cost.

AWS segment: AWS revenue was $46 billion, up 39.3% year-on-year, driven by both core and AI services, with a sequential revenue increase of $3.8 billion. AWS is now running at an $184 billion annualized rate. Customers keep migrating to the cloud and expanding usage of core AWS services. Increasingly, customers seeking full AI benefits are accelerating migration too.

We see a strong correlation between AI spend and our core service revenue growth. As customers invest in AI, we see corresponding core consumption growth, and as more AI workloads go to production, driving incremental compute, storage, and networking demand, we expect this link to strengthen. AWS operating profit was $17.8 billion with a 38.7% margin, reflecting strong growth and our efficiency focus. Our investments in proprietary chips, custom networks, and software/process improvements help us optimize capacity and build more cost-effective infrastructure. As always, AWS margins will vary, partly based on the timing of ongoing investments. Our AI products are gaining traction, with AI revenue up triple digits year-on-year; Bedrock, in particular, is a key growth driver.

Cash capex for the quarter was $75.9 billion, mostly related to AWS and AI, as we invest to support strong customer demand. We’re making large investments, especially in AI, because we see a massive opportunity for long-term revenue, free cash flow, and return on invested capital.

Q4 outlook: We expect net sales between $240 and $247 billion. At current exchange rates, FX is about a 60bps year-on-year headwind this quarter. Q4 operating profit is expected between $28.5 and $33.5 billion. One factor affecting Q4 guidance:

Amazon Leo commercial service remains on track to start initial operations this year, with some capitalization of production and launch costs to begin next Q1—slightly later than prior guidance. Thus, most Leo-related costs will continue to be expensed in Q4. We're ready to serve customers this holiday season and support millions building on AWS.

Thanks to all Amazon teams for hard work and customer-focused dedication. Our commitment to customer experience remains the only reliable path to long-term shareholder value.

Now, Q&A.

Q&A Session

(1) Q1—BofA Securities Analyst Justin Post

Question: Andy, great to see continued acceleration in cloud, very impressive. Sequential growth has slowed somewhat—are there any customer-side factors, token costs, etc.? How do you view Q4’s capacity expansion? And Brian, what impact does fuel cost have on Q4 guidance?

Andrew Jassy response:

Sure, I’ll take the growth question first. We’re very pleased with this growth—a business with a $184 billion run rate growing 39.3% is quite rare. It's our fastest growth in nearly five years, and the sixth consecutive quarter of acceleration. We're also adding substantial capacity. If you look at the current growth drivers, it’s a combination of several factors. First on AI, we offer extraordinary end-to-end functionality—with SageMaker AI, customers can build their own models the easiest way, and with Bedrock, use leading-edge models, high security, and extremely competitive price/performance.

Whether building your own agents (we offer Strands and AgentCore for safe, scalable development/deployment), or using out-of-the-box agents (the majority of companies), adoption is rapid—Kiro for coding, Connect for contact centers, Transform for migration, Quick for AI work companionship, and so on.

We add substantial capacity each quarter, though not the same in every quarter. Looking at this year’s capacity adds, and our plans for the next ~8 quarters (we generally forecast 8–10 quarters out), there’s a lot more capacity coming. On top of that, there’s the chip impact. We’re in an advantaged position in the AI wave, as more and more customers—especially those scaling AI inference in production—are hugely motivated to save on cost for price/performance and training, and Trainium offers significant advantages in cost efficiency versus alternatives.

As AI usage broadens, core capabilities like reinforcement learning, post-training, and tool invocation are increasingly demanded, and we have the leading CPU chip, Graviton, at 30–40% better price/performance than other x86 providers. These forces are truly driving remarkable growth. While quarterly capacity adds vary, forecasting over 40% year-on-year growth in 2027 and 2028 at this scale, and a path toward $1 trillion in five to ten years, is extraordinary—we’re very excited to go after it.

Brian Olsavsky adds:

Justin, regarding fuel and line-haul inflation, I don’t have a precise number but Q4 will be higher than Q3. Recall we implemented a fuel surcharge for FBA sellers earlier this year, recouping part—but not all—of the fuel cost we’re seeing for these shipments and the remainder of retail inbound/outbound transportation. So that’s included, and is a bit higher than Q3.

(2) Q2—Morgan Stanley Analyst Brian Nowak

Question: I have two questions. Andy, your enthusiasm for AWS is palpable. You rarely talk about “$1 trillion in five to ten years”—outside your usual range. Can you walk us through how you’ll get there? Is it top-down or bottom-up? What gives you confidence in such revenue growth over the coming years?

Second, regarding robotics: looking out to 2027, what are the key milestones you’re focused on for robots in the retail business?

Andrew Jassy response:

On AWS, Brian, just look at the growth rates—the run rate acceleration, our 2027/2028 outlook. As I’ve said many times, however much capacity we provide, customers pre-commit for large blocks (hundreds of megawatts, sometimes up to a gigawatt) because you can’t leave a gigawatt sitting idle. So, we have very clear visibility into these bulk capacity needs for 2027, 2028, and some of 2029.

We always reserve a chunk of EC2/Bedrock capacity for on-demand usage and work closely with customers to ensure sufficient AWS capacity is available. But for the next few years, we have very clear visibility on what AWS will look like, and when you run the numbers, the growth rate quickly scales up the total.

I don’t think much of this usage will quickly fall off, either. Enterprise deployment of inference in production is at a very early stage—truly early. Partly due to a lack of talent who understand how to run AI, post-train models, and tune for appropriate applications; partly due to the cost of inference. If you want inference to be used as widely as we do, you must continually drive down the cost—that's a major AWS mission, and a major reason for our investments in custom chips and Trainium. Inference costs have already dropped sharply over the past two years, and I expect they’ll drop further.

Then, think about what you’re seeing. In security, for example, today’s models are so advanced they discover vulnerabilities humans never noticed. If you can use these models, you can help find, fix, and deploy solutions. Some think a large share of all tokens consumed in the next year or two could be for security—a relatively new use case, on top of growth in coding and a variety of AI companions (in both consumer and workplace, like Quick). These are just the first few major use cases.

Combine that with enterprise trends—I believe nearly all modern corporate production apps will be rewritten with agent capabilities, because if you want to move fast or mitigate security risks properly, you’ll have to. The AI wave is that big. So, while you can’t forecast precisely, especially at these growth rates, picturing this as a trillion-dollar opportunity isn't outlandish for us.

On robotics, I’d say what’s happening in robotics broadly, and particularly at Amazon, is pretty stunning. We now have over a million robots in our fulfillment network. The impact on our cost structure, productivity, and safety is significant—look at our fulfillment safety data, it’s at historic lows. Over the past four or five years, we’ve made enormous progress, and robots are a big part of it.

But it’s still early—both in terms of what we've built and deployed, and what's in development. There will always be lots of employees in our fulfillment centers, but much of the repetitive work now performed by humans will be done by robots. We’re investing in many robot forms and also have opportunities to help with the “last hundred yards” of delivery to physical stores. Lots coming soon.

For us, the robot test is: does it deliver equal or better productivity, safety, cost, and delivery time? If the economics work, we scale up in fulfillment. You’ll see that upcoming fulfillment designs—including same-day facilities extensions—will involve heavy robot investment. Robots already have a huge impact, but that's just the tip of the iceberg compared to what’s coming in five years.

(3) Q3—Evercore ISI Analyst Mark Mahaney

Question: Two things. Thanks for the AWS growth outlook. Can you also provide a margin outlook? I know it fluctuates, but over the next two years, do you see the margin structure staying around 30%+, 40%? Second, on fresh checkout—I’m very excited to see this in my community and others. How broad is your fresh checkout coverage? In other words, how many Prime households are served, and for those using it, how much does it drive household spend—20%, 50%, 100% increase? Given fresh's huge importance?

Brian Olsavsky response:

Mark, first on AWS operating margin outlook. We don’t guide specifically on AWS margins. But here’s my view. This quarter’s 38.7% margin is very strong. Some context: recent improved profitability is not random—it’s the result of disciplined efficiency improvements, capacity optimization, and tight cost controls. As Andy said, over time, custom chips build into our cost structure. Other factors are at play too. In fact, explosive AI growth puts pressure on total margin, because those margins are lower.

We’re satisfied with both core and AI margin progress. As Andy said earlier, AI margins are ahead of where core business margins were at the same stage, but still lower overall. So, there is a mix effect. We also face negative component pricing. So, it’s unpredictable, but always a mixture of factors, and we work hard on cost and efficiency, while managing new-product margins as they scale up.

Andrew Jassy adds:

On fresh checkout—Mark, I’m just as excited. We’ve been doing big grocery for some time, mainly focused on center-store items—essentials, canned foods, drugs, beauty. We own the leading organic grocer Whole Foods Market, which is growing faster than the rest of grocery, both in revenue and profit, and our new urban formats are also strong and set for expansion in coming years.

We’ve always known that to serve customers the way we and they want, we must cover mass-market products, and that starts with fresh. If you look at household spend, the biggest gap between what consumers can buy from us and what they actually do is grocery. Many grocery trips start with fresh. We’re excited by the direction we're taking—delivering fresh from same-day facilities.

Essentials already grow at about three times the rate of the rest of stores, making up about half of unit volume. Customers buying fresh with same-day delivery also buy about seven times more essentials than those not buying fresh. So, if you want to close the gap between total consumer spend and our available selection, fresh is a major factor.

We also see grocery customers buying home, apparel, and other non-grocery categories about four times more often—and spend twice as much in those areas—compared to non-grocery customers. These are large differences, and we believe progress in fresh will make us more relevant to customers in more shopping scenarios over time.

(4) Q4—Goldman Sachs Analyst Eric Sheridan

Question: Two AWS market opportunity questions. First, on Bedrock—what are your key insights on the Bedrock market opportunity, and how will it expand over the coming years? Second, on the decision to externally sell Trainium: how do you balance external sales with required compute and internal AWS business, and how do you see the market opportunity?

Andrew Jassy response:

On Bedrock—if you’ve listened over the last two years, you know my view. There’s never one tool to rule them all. Early leaders are never the only game in town—databases, analytics, etc. In AI’s early days, everyone thought TensorFlow would be the only framework, and now PyTorch is bigger.

Models are the same. Over the next 2–3 years, I believe there will be six to eight models at parity, or at least enough to cover 95%+ of workloads. Some will excel at inference, others at coding, some at vision, some with lower latency, some lower cost. Most customers will use multiple models, often multiple within a single generative AI application.

So what customers really want—which we've seen validated repeatedly—is a service with broad access to advanced models, great price/performance, and strong security/access control. They want to decide their mix and switch as performance or constraints change. That's what we see in Bedrock—tremendous excitement.

We offer closed models like Anthropic and OpenAI, plus open models like Kimi, GLM, etc. There’s strong appetite in the West for more open-source Western models, and I believe they’ll come. Bedrock’s opportunity is massive—it could rival EC2, which is huge inside AWS. So, I think it's very big.

On chip balance—how much to sell externally versus use in our own cloud—it’s a great question and something we continually evaluate. Only two companies worldwide have cloud businesses large enough to sell chips externally. Everyone uses the same foundry for advanced chips; the foundry only delivers so many per year, so everyone gets an allocation. You must decide how much to sell externally and how much to use internally.

There are trade-offs: selling externally brings immediate revenue, as chips are recognized on shipment. Alternatively, using chips inside EC2 or Bedrock drives more total revenue and free cash flow over 5–6 years of usage. So, you must think carefully. To date, we've taken a different approach on chip use than the only other cloud provider facing this question.

As noted before, external chip sales will be a significant business, launching at the end of 2027. We'll likely revisit the mix yearly, but both will matter. Our ability to deliver high price-performance, scalable chips will be a huge strategic advantage—both for our cost structure and as a standalone business. Selling chips externally is a very different business from cloud infrastructure itself.

(5) Q5—J.P. Morgan Analyst Douglas Anmuth

Question: Considering 40%+ AWS growth over the next two years and the path beyond, can you discuss the capital expenditure required to drive such growth in the coming years? Also, Andy, any updated thoughts on server, network, and data center payback cycles?

Andrew Jassy response:

Let me start with payback. As I said in my remarks, we're continually shifting payback left—shortening it. There are slight differences between core and AI, and financing differs depending on which part, but we’re making good progress. As we’ve always done at AWS, we keep extending useful life and cutting breakeven time, with solid results.

For the level of growth we’re talking about, capex will be substantial. Doug, our view is this is a once-in-a-lifetime opportunity. We believe AI is the biggest transformation of our lifetimes. If we can deliver 39% growth on a $184 billion run rate, then 40% per year on top of that for two more years, and end up as a $1 trillion+ business, we will invest to lead. We’re already infrastructure and AI leaders. If this acceleration and opportunity were spread over 25 years, you could spread out capex too, but that’s not our world or the pattern of opportunity we see.

We have the means to fund this plan. It will forever change Amazon’s scale, revenue, and free cash flow. We'll invest and pursue leadership as we do today.

(6) Q6—Barclays Analyst Ross Sandler

Question: Another follow-up on Bedrock: the Bedrock token marketplace seems to be the hottest AI service right now, and most workloads reportedly run on Trainium. As this becomes a bigger part of the AWS story, how will it affect revenue growth—especially AWS margins, as you shift from bare-metal sales toward more inference-as-a-service?

Andrew Jassy response:

Yes, you’re correct—most content on Bedrock runs on Trainium. We do this because we love the price/performance we can unlock, both for offering customers the lowest possible cost and for sustaining our own margins. Looking at cloud margins, the critical factors include data center ownership—whether you own or long-term lease, versus use colo (colocation) or neocloud (new cloud). Neoclouds’ operating margins are about 3500 basis points lower than long-term leasing.

The other consideration is the hardware. If you run commodity hardware everyone else uses, your cost structure will be like everyone else’s. If you use differentiated hardware, you have an advantage. Most of our capacity is owned or long-term leased, not heavily dependent on colo or neocloud, and this structure results in better operating margins than competitors that still depend on them. And we can run on our own chips, for both price/performance and structural cost benefit—delivering savings to customers and robust AWS economics.

(7) Dave Fildes

Thank you all for joining today’s call and for your questions. A replay will remain available on our investor relations site for at least three months. Thank you for your interest in Amazon and we look forward to connecting again next quarter.