Hello, and welcome to our Q3 2026 financial results conference call. Joining us today to answer your questions is Andy Jassy, our CEO, and Brian Olsavsky, our CFO.

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 as well as metrics and commentary on the quarter. Please note, unless otherwise stated, all comparisons in this call will be against our results for the comparable period of 2025.

Our comments and responses to your questions reflect management's views as of today, October 29th, 2026 only, and will include forward-looking statements. Actual results may differ materially.

Additional information about factors that could potentially impact our financial results is included in 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, slides accompanying this webcast, and our filings with the SEC, each of which is posted on our IR website, you will find additional disclosures regarding these non-GAAP measures, including reconciliations of these measures with comparable GAAP measures. Our guidance incorporates the order trends that we've seen to date and what we believe today to be appropriate assumptions.

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 for memory chips, and customer demand and spending, including the impact of recessionary fears, inflation, interest rates, regional labor market constraints, world events, the rate of growth of the internet, online commerce, cloud services, and new and emerging technologies, and the various factors detailed in our filings with the SEC. Our guidance assumes, among other things, that we don't conclude any additional business acquisitions, restructurings, or legal settlements.

It's not possible to accurately predict demand for our goods and services, and therefore, our actual results could differ materially from our guidance. Now, I'll turn the call over to Andy.

Thanks, Dave. We're reporting $201.9 billion in revenue, up 12% year-over-year.

Operating income was $28.1 billion, up 61% year-over-year. Q3 was another strong quarter for Amazon.

In AWS, our plan is working. Despite lapping a comparable year-over-year period where we added a lot of capacity, AWS grew 39.3% year-over-year, accelerating for the sixth straight quarter, our fastest growth in nearly 5 years.

Backlog is $578 billion, growing triple digits year-over-year and well-diversified. We're installing an AI accelerator about every 14 seconds, and almost as fast as we install capacity, customers are gobbling it up.

Our core compute and storage revenue grew at their fastest rates in several years. Our operating margins have long been strong on our core offering, but rapidly improving on the AI side too, well ahead of where they were on core at a similar time of evolution.

Even with heightened component costs like memory and significant depreciation from the CapEx we're investing, we continue to accelerate our CapEx payback, and have even more conviction than we already had that the investment we're making in AWS will yield a much larger revenue and free cash flow generating business than we would have imagined a couple years ago with strong return on invested capital. We have a compounding flywheel spinning in AWS.

Borrowing from Jim Collins' flywheel definition in Good to Great, where if you do A, you almost can't help but do B, and if you do B, you almost can't help but do C, and so on. In AWS, having the leading functionality, security, and operational performance can't help but drive customers to run their apps and store their data on AWS, which can't help but drive AWS AI demand, because customers prefer their production inference to be near the rest of their data and apps, which can't help but drive core AWS demand, because AI uses core for work like reinforcement learning, post-training, and tool usage, which can't help but drive reinvestment back into our broader AWS functionality.

This is a different model than primarily hosting GPUs or renting short-term data center capacity. This is a long-term platform strategy that broadly helps customers invent and evolve their customer experiences, operate stably, and build a reliable partnership that can grow for many years.

It also yields very attractive long-term economics for us, regardless of supply scarcity. There are many examples that illustrate our progress in AWS, but I'll share a few.

On the core side, it's important to remember that the vast majority of global IT spend, about 85%, is still on-premises, and we're in a 10-to-20-year cycle where that equation will flip to the cloud. Conventional wisdom has been that linchpin workloads like mainframes or storage array networks would never move to the cloud, but these workloads are moving, as AWS has built unique cloud substitutes and AI-driven migration capabilities to make this transition much easier.

Enterprises get the benefits of less CapEx, lower operating costs, faster innovation, better developer productivity, and data accessibility to more easily leverage AI. Our core business has attractive margins because we've learned to run it very efficiently over the last 20 years.

As our core business grows, we keep reinvesting in our core and AI functionality. For example, as customers generate increasingly more data and seek to get value from it in analytics and AI, the complexity of moving data between data stores and analytics engines is onerous and expensive.

With our acquisition of DuckLabs, DuckDB will become the universal SQL dialect of AWS storage and analytics, simplifying data processing with a modern analytics engine that is fast, simple to use, and ubiquitous, and where customers will be able to write SQL queries once and then execute using any AWS analytics service. For the first time in analytics, customers will have zero migration or portability costs between query engines.

This is a big deal, and you can imagine how it'll also help agents do efficient querying as well. On the AI side, we continue to expand the capabilities of our inference service, Amazon Bedrock, and it's leading to remarkable growth, with Fortune 500 companies spending nearly 7 times more on Bedrock year-over-year, and total Bedrock token volume growing nearly 17 times year-over-year.

Bedrock is the fastest-growing multi-billion-dollar service in AWS's history. We've recently added OpenAI models to Bedrock, and usage is growing quickly.

We also just added Bedrock-managed versions of Kimi, GLM, and SpaceX models. You can't have a conversation about AI today without talking about security, and it's another reason customers keep choosing Bedrock.

We've designed Bedrock so customers' data remains secure and private. Inputs and outputs are never shared with model providers or used to train base models.

When customers tune a foundation model, we create a private copy specifically for their use, and data is fully encrypted, both in transit and at rest. As frontier models get more powerful, they're surfacing software vulnerabilities previously not found and at unprecedented rates.

We built AWS Continuum to discover, prioritize, validate, and remediate code vulnerabilities at machine speed. It's already being used by customers like Capital One, IBM, Infor, Luminate, and Robinhood.

In Q3, we announced that we'll integrate it into the tools where code's being written: Anthropic's Claude Code, OpenAI's Codex, and AWS's Cura. The interest from customers exceeds anything we've ever seen for a new security service.

Agents continue to be deployed at accelerating rates, many of which are being built on Bedrock AgentCore's building blocks. Tens of thousands of customers currently use AgentCore, deploying 2 times more agents with it in Q3 than the prior 3 quarters combined.

In addition to building their own purpose-built agents, most companies will also use turnkey agentic services, and we offer several that customers love: Kiro for coding, Amazon Connect for contact center agents, and Amazon Transform for migrations. They're all iterating quickly and picking up customer momentum, as is Amazon Quick, our intelligent AI work companion, which is growing very rapidly as large enterprises like Mondelez, EchoStar, 3M, Southwest Airlines, and DXC are choosing Quick because it's purpose-built for work with no compromises on customer data.

Finally, our chips business continues to grow triple-digit percentages year-over-year, and revenue commitments for Graviton and Trainium have each increased more than 20 times year-over-year. AI continues to be a tailwind for Graviton because post-training, reinforcement learning, and agent tool use is mostly done on CPUs versus AI accelerators, and Graviton is the strongest CPU chip, offering up to 30% to 40% better price performance than alternatives.

Customers using our custom silicon saved more than $2 billion in Q3 alone, and we just open-sourced TorchNeuron so developers can run existing PyTorch code, which most already use, on Trainium without rewriting a thing. As we've launched new generations of chips, demand for prior generations remains strong.

After our launch of Graviton5, backlog for prior-generation Graviton instances still more than doubled quarter-over-quarter, and short-term capacity for prior-generation Trainium instances sell at a premium compared to the first half of the year. With Trainium demand so strong, customers continue encouraging us to make it available outside of AWS; we intend to do so late in 2027.

Overall demand for AWS capacity remains exceptionally strong. Enterprises are still very early in using inference at scale, and yet inference already accounts for well over half of AWS's AI revenue and growing, and we have a lot of training revenue as well.

We still expect 2026 cash CapEx of approximately $220 billion, the majority for AWS, and expect to spend meaningfully more in 2027, with a clear line of sight to attractive financial returns, as I laid out on our last call. AWS is a $184 billion annualized revenue run rate business, and yet, the opportunity remains extraordinary.

With our continued pace of innovation, new core and AI functionality, enterprises wanting to run their core and AI workloads in AWS, and the amount of capacity that we're adding each year, we believe that AWS can grow north of 40% in each of 2027 and in 2028, and be a trillion-dollar revenue business in the next 5 to 10 years. We're in the very early days of this inflection, our strategy is working, and I expect it will for the foreseeable future.

In Stores, I think some folks are sleeping a bit on how much progress that business has made and what it's become the last few years. I'm not aware of another retail business approaching Amazon's revenue size, hundreds of billions of dollars, with double-digit operating margins inclusive of ads, that's growing at this pace.

Year-to-date, units have grown nearly 15% year-over-year, and that's on an enormous base, having sold more than 30 billion items last year. In absolute dollar terms, we've added more than $90 billion in gross sales in the trailing 12 months, and we're growing meaningfully faster than other retail competitors, adding about double the gross sales per quarter than our next-largest competitor in the U.S., and consistently taking retail market segment share.

Over the past few years, our North American and International segments have each expanded operating margins by nearly 10 percentage points, which enables us to reinvest in making the customer experience even better. Remember, it's still early.

The significant majority of retail, nearly 80%, currently occurs in physical stores. That will eventually flip to e-commerce over time.

We're helping enable this shift by rapidly building out our efficient and speedy fulfillment network, providing delightful customer experiences while driving significant free cash flow and return on invested capital. Each of our priorities, from faster delivery, to cost to serve, to groceries and pharmacy, to AI shopping, are force multipliers, and we're firing on all cylinders.

Our delivery speeds keep getting faster as we invest in our same-day network, as well as our Amazon Now and Amazon Prime Air ultrafast services. We had our fastest Q3 ever globally, and we delivered more than 1.5 billion items to Prime customers same day or overnight in Q3.

We're energized by the customer response to our rapid rollout of Amazon Now around the world. It's now available in over 450 towns and cities, and the number of items ordered through Amazon Now is up 10x year-over-year.

We're on track to reduce cost to serve per unit for the fourth consecutive year, which enables us to continue meeting or beating competitors' prices and cost-effectively deliver a broad selection of fresh groceries and everyday essentials. Increasing speed is fueling our grocery business, including perishables and everyday essentials, and gross sales grew to nearly $250 billion in the trailing 12 months.

In Q3, fresh groceries and everyday essentials grew about 3 times faster than the rest of our store, representing about half of all units sold. Customers who shop perishables with same-day delivery buy nearly 7 times more everyday essential units than those who don't, and active grocery customers shop nearly 4 times more non-grocery categories, like home and apparel, and spend over 2 times more on those categories compared to non-grocery customers.

Amazon Pharmacy is also growing fast. The number of customers using it is up over 60% year-over-year, with growth accelerating each quarter through 2026, and repeat customers grew more than 50% year-over-year.

J.D. Power recently ranked Amazon Pharmacy the top mail-order pharmacy for the third year in a row, and we keep making it even better, now offering 3-hour prescription delivery in more than 350 cities.

Over 400 million customers have used Alexa for Shopping, our agentic shopping assistant, in the last 12 months, with customer interactions up 570% year-over-year. U.S.

customers using Alexa for Shopping are spending 50% more per order than those who don't. It researches products, compares options, checks a full year of price history, gets personalized gift recommendations, and tracks deals and price drops on customers' behalf.

Though we anticipate working with horizontal agents, we believe Alexa for Shopping is going to be the primary shopping agent customers use to shop Amazon. Having a shopping assistant constantly by your side as you browse, get inspired, have questions on product detail pages, and that you can simply ask to buy something based on your prior purchases that already has your history, your shipping address, and payment info is uniquely helpful and convenient for customers.

Moving to Amazon Ads, we generated $19.6 billion of revenue in Q3, up 12% year-over-year. Though there were some timing anomalies in this past quarter, we continue to be pleased with the overall momentum across our ad offerings.

I'll mention three items impacting the growth rates. First, we shifted Prime Day into Q2 for most countries, moving significant sales and ad revenue that were in Q3 '25 into Q2 '26, a tailwind to our Q2 2026 rate and a headwind to Q3 2026.

For advertising, that was an 850 basis points sequential impact to Q3 year-over-year growth. Second, Q2 '26 included the first season of NBA on Prime, which last year's Q2 didn't have, while Q3 does not include any NBA, it starts in Q4.

Third, in Q3 2025, we changed our search engine marketing programs to drive more traffic to search pages on Amazon. It's a better experience for customers, and we also monetize that traffic at a higher rate.

That helped accelerate advertising revenue growth starting a year ago, and we started lapping that in Q3. We expect year-over-year growth will return to the high teens in Q4.

A natural question is whether agentic shopping from external horizontal agents is meaningfully impacting Amazon Ads, and the answer is no. Few customers are doing much shopping this way yet, as the experience is still not great, nor more convenient than coming to a retailer directly to research or one-click order there.

Q3 was chock-full of innovation from our ads team that we expect to have impact in the coming months, headlined by Amazon Ads Agent, which brings together our full advertising offering under a single AI-first platform. Backed by trillions of shopping, browsing, and streaming signals, it supports advertisers at every step with insights, answers, and the ability to optimize and act on their behalf.

We also launched full-funnel campaigns in Amazon Ads Agent, bringing sponsored ads, display, video, and streaming TV into a single campaign. Advertisers provide initial inputs, including budget and duration, and Amazon's AI plans, executes, and continually optimizes the rest.

We're continuing to see momentum in several other areas; I'll mention just a few. Prime Video has emerged as a significant business, with a revenue run rate in the tens of billions of dollars, and now profitable in its own right.

We have momentum with hit shows like Carrie, The Summer I Turned Pretty, The Boys, Off Campus, Reacher, and Fallout, and breakout movies like Project Hail Mary, along with compelling live sports, including Thursday Night Football off to its best start ever in 2026, NBA on Prime, NASCAR, The Masters, and UEFA Champions League soccer. We recently launched our most powerful devices lineup yet, with 34 new products across Echo, Kindle, Alexa tablets, Fire TV, Ring, and Blink.

It's early, but we're off to a very strong start, and excited for more customers to get their hands on these as we head into the holidays. Finally, Amazon LEO remains on track to begin initial service this year.

We're heading into one of our most exciting stretches of the year with both re:Invent and the holidays, and I want to thank our teammates for their hard work and invention to make customers' lives better and easier every day. What we're building at Amazon is unlike anything that's ever been seen before.

There has never been a business with a trillion dollars of annual revenue. Amazon is on a path to have two of them, in AWS and Stores.

It's rare, and we're energized by what's ahead. Over to you, Brian.

Thank you, Andy. Let's start with our top-line financial results.

Worldwide revenue was $201.9 billion, a 12% increase year-over-year, and 13% increase year-over-year excluding the impact of foreign exchange. Worldwide operating income was $28.1 billion, with an operating margin of 13.9%, which is our highest operating margin ever.

Starting with our North America and International segment results, first let me remind you that Prime Day shifted this year, with the sales event occurring in Q2 for most of our largest countries, including the United States. In 2025, Prime Day was entirely in Q3.

Excluding the impact of this Prime Day time change, third quarter 2026 year-over-year growth would be approximately 400 basis points higher. In Q3, we saw strong customer engagement and a continuation of trends from the first half of the year.

In the North America segment, third quarter revenue was $112.9 billion, an increase of 6% year-over-year. International segment revenue was $43.1 billion, an increase of 8% year-over-year excluding the impact of foreign exchange.

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 our combination of vast selection, low prices, and record delivery speeds. Over the past year, monthly orders of fresh groceries placed by our Prime members for free same-day delivery have more than tripled.

Let's shift to profitability. North America segment operating income was $8.7 billion, with an operating margin of 7.7%.

International segment operating income was $1.6 billion, with an operating margin of 3.7%. In our fulfillment network, the team navigated higher transportation costs from fuel inflation and higher line-haul rates.

Even with these headwinds, we still lowered our overall cost to serve in Q3 through better inventory placement, shorter shipping distances, and package consolidation. Shipping costs increased 11% year-over-year versus a 12% increase in units.

Excluding fuel and line-haul inflation, shipping costs grew 8%, or approximately 400 basis points slower than unit growth. Robotics and AI are delivering measurable gains in productivity and employee safety, which we'll expect to compound in the future.

In 2026, we doubled our fleet of robotic arms like Cardinal and Sparrow. We're also developing robotics foundation models that enable faster programming and teach robots increasingly complex physical tasks, with the goal of delivering useful robotics at scale.

Looking ahead, we see meaningful opportunities to improve productivity across our global fulfillment network and lower our cost to serve, while continuing to get products to customers faster. Our operating margin may fluctuate quarter-to-quarter, but we're focused on sustained long-term improvement in our cost to serve per unit.

Moving to our AWS segment, revenue was $46 billion and grew 39.3% year-over-year, driven by both core and AI services. We added $3.8 billion in quarter-over-quarter revenue.

AWS now has an annualized revenue run rate of $184 billion. Customers continue to move to the cloud and scale their use of AWS core services, and increasingly, customers who want the full benefits of AI are accelerating their transition to the cloud.

We see a strong linkage between AI spend and growth in our revenue for core services. As customers invest in AI, we see a corresponding increase in core consumption, and we expect that relationship to strengthen as more AI workloads move into production and drive additional demand for compute, storage, and networking.

AWS operating income was $17.8 billion, with an operating margin of 38.7%, reflecting strong growth and our focus on driving efficiencies. Our investments in custom chips and custom networking, along with software and process improvements, are helping us optimize capacity and build a more cost-efficient infrastructure.

As we've long said, we expect AWS operating margins to fluctuate, driven in part by the level of investment we're making at any point in time. Our AI offerings are gaining traction with customers, with AI revenue growing triple digits year-over-year.

Bedrock has been a significant growth driver. Now turning to our cash CapEx, which was $75.9 billion in Q3, this primarily relates to AWS and AI as we invest to support strong customer demand.

We continue to make significant investments, especially in AI, as we believe it to be a massive opportunity with the potential to drive long-term revenue and free cash flow, as well as return on invested capital. I'll finish with our financial guidance for Q4.

Q4 net sales are expected to be between $240 billion and $247 billion. We estimate the year-over-year impact of changes in foreign exchange rates based on current rates, which we expect to be a headwind of approximately 60 basis points in the quarter.

Q4 operating income is expected to be between $28.5 billion and $33.5 billion. I'll mention one factor impacting our Q4 guidance.

Amazon LEO's commercial service is on track to begin initial service this year, and we expect to begin capitalizing certain costs in Q1 of next year, including production and launch costs. This is later than I've been predicting in earlier earnings calls.

As a result, we will continue to expense the majority of LEO-related costs in Q4. We're ready to serve customers this holiday season and to support the millions of customers building with AWS.

I want to thank our teams across Amazon for their hard work and focus on customers. Our commitment to elevating the customer experience remains the only reliable way to create lasting value for our shareholders.

With that, let's move on to your questions.

Great, thank you. Andy, continue to see the cloud business accelerate, which is really nice, but the quarter-over-quarter growth slowed, and I'm just wondering if there's any callouts there on the customer side, or token costs, or anything that kind of drove that, and how you think about the capacity adds in Q4?

Then Brian, maybe just any impact of fuel costs on your guidance in Q4? Thank you.

I'll start with the growth question. We're really pleased with the growth, 39.3% on a business that's got a $184 billion annual revenue run rate is quite unusual.

Our fastest growth in about 5 years, or sixth quarter consecutively of acceleration. We're adding quite a bit of capacity.

I think if you look at what's driving the growth right now, it's a combination of a few things. First, on the AI side, the unusual top-to-bottom stack functionality we have, whether you're building your own model with SageMaker AI, which is the easiest way to build your own model, or you want to leverage one of the leading frontier models with high security and very strong price performance, people can use Bedrock, whether you are building your own agents, where we have capabilities like Strands and AgentCore to build and to run those securely and scalably, or you're going to use turnkey agents, which most companies do, if you just look across Kiro for coding, and Connect for call center, and Transform for migrations, and Quick, which is growing incredibly fast as an AI companion at work, people are adopting those very quickly.

Then we add a lot of capacity every quarter. Now, the amount that we add every quarter is not equal, and we're adding different amounts each quarter.

If I look at how much we've added this year, and what we have planned over the next, I'll call it, 8 quarters or so, which is about as far out as we get, 8 to 10 quarters, we have a lot of capacity coming as well. Then I layer on top of that, too, just the impact of chips.

We're very well-positioned for what's happening in AI, because increasingly, what you're finding right now is every customer that uses AI at scale, but particularly enterprises now, as they're starting to deploy more inference into production, they are very eager to save on price performance. Trainium, with the meaningful advantage it has in price performance versus other alternatives, is very compelling to people.

Then as more and more people are using AI in a very broad way, it drives a lot of the core in capabilities like reinforcement learning, and post-training, and tool calling. We have the leading CPU chip in Graviton, which is about 30% to 40% more price-performant than other X86 providers.

Those things are really driving very significant growth. The amount of capacity we add each quarter will vary, but at this scale, to be projecting growing over 40% year-over-year in each of 2027 and 2028 is pretty significant.

The fact that we believe that this can be a trillion-dollar revenue business in 5 to 10 years, it's an extraordinary, unusual opportunity that we're very excited to be pursuing.

Justin, on your question about fuel and line-haul inflation, I would say, I don't have an exact number for you, but it is higher in Q4 than it was in Q3. If you remember, we instituted a fuel surcharge at the beginning of the year for FBA sellers that recovers some, but not all of the costs that we're seeing on fuel for those shipments, and then the remaining part is on the retail inbound shipments and outbound shipments.

It's in there. It's a little bit higher than Q3.

Thanks for taking my questions. I have two.

Andy, your excitement about AWS is coming through the line, and it's pretty rare you would make comments about a trillion dollars of revenue 5 to 10 years from now. It's more than you usually give.

Could you just maybe walk us through how you get to that trillion of dollars? Is it top-down, is it bottom-up?

What gives you confidence that you're going to be able to put up that type of revenue growth to get to that type of a number in the next 5 to 10 years? Then the second one, just for either of you, on the robotics side, as we look into '27, what are the key milestones you're both focused on the robotic side for the retail business?

On the AWS side, if you just look at the growth rates, Brian, look at the growth rates on the run rate, and then look at what we're projecting in '27 and '28. As I've said a few times, the big chunks of capacity that people take, the hundreds of megawatts, or sometimes up to a gigawatt of capacity, those big chunks are reserved well in advance, because you can't leave a gigawatt of capacity hanging around.

We have very good visibility into '27 and '28, and some of '29, too, with respect to what customers want in those really big chunks of capacity. We will always leave a fair bit of capacity available for on-demand on EC2 or Bedrock, and we work closely with our customers, and they'll have plenty of capacity to be able to use in AWS.

We have very good visibility into the next few years on what AWS looks like. Those types of growth rates, when you do the math on the revenue and the revenue run rate we have, gets you pretty quickly to escalating numbers.

Then I'd say that, on top of not believing that a lot of the usage we see today is going to attenuate anytime soon, we're still really early in enterprises using inference in their production applications in a pervasive way. It is just really early.

Some of that is enterprises having enough people who know how to operate in AI and post-train and fine-tune models for the right applications. Some of it is the cost of inference.

I think if we all want enterprises to be able to use inference the way we all believe they will, we have to keep working on taking the cost of inference down. That is a huge mission for AWS.

It's one of the significant reasons why we're building our own custom silicon and Trainium, is that we just really believe it's incredibly important for customers to be able to use inference in a much more cost-effective way. That has come down a lot, the cost of inference, over the last couple years, and I think it will come down meaningfully more.

Then you just think about what you're seeing. If you think about security as an example just by itself, the models today are so advanced that the vulnerabilities that they're finding, humans just haven't found.

If you've had access to these models, it leads to you're using these models to both find them and ultimately to help with patching them and deploying. There are some people who think that the total amount of tokens used over the next year or two, 20% of it, some significant percentage of it may end up being just security.

That's a relatively new use case on top of what we're seeing with coding, and how that continues to grow, on top of what we're seeing with the various, I'll call it, AI companions, both in the consumer space and in the work space, like with Quick. These are just the initial big use cases.

You layer on top of that what you're going to see with enterprises, where I believe virtually every modern enterprise production application will be rewritten to be agentically capable, because that's the way you're going to have to deploy if you want to move quickly, or you want to actually guard in the right way against some of the security risks. There is just so much AI coming.

You can't predict with any certainty 5 to 10 years from now, especially at these types of growth rates, but it's not a stretch for us to imagine that this could be over a trillion-dollar business. On the robotics side, what I would say is that it's pretty remarkable what's happening in robotics right now.

I'll say in general, but particularly at Amazon, we have over a million robots now in our fulfillment network, and while they're having very significant impact on our cost structure and on our productivity and on safety, if you look at our safety numbers in our fulfillment network, they're at record-low numbers. We've made huge progress over the last 4 to 5 years, and robotics has been a big piece of that.

Yet I would tell you, we're still so early with respect to both what we have built and deployed and what we're working on. I think that you'll see we will always have a lot of teammates in our fulfillment centers, but I think a lot of the jobs that we have that are more repetitive, that we have humans doing, will be done by robots.

I think we're investing in all sorts of additional robotics of all different form factors. I think we have the opportunity to also assist on some of the delivering the last 100 yards to customers' doors.

We have a lot coming. For us, the way we measure robotics are: Does it lead to comparable or better productivity, safety, cost structure, and time to delivery?

When we like that combination of economics, we will deploy it at scale in our fulfillment network. If I look at the new designs that we have coming in our fulfillment network, including our expansion of same-day facilities, they are very heavy on robotics.

We think it's had a big impact, and we think it's small compared to what it's going to be in the next 5 years.

Two things. One, first, thanks a lot for the forward comments on AWS growth.

Could I ask you to draw out on margins, too? I know that margins can move up and down quarter-to-quarter, but over the next 2 years, do you think the current margin structure is relatively stable?

We can kind of hold at these kind of high 30s, 40% operating margins for AWS. Then on perishables checkout, I'm super excited about this opportunity, just watching how it's rolled out in my neighborhood and in other areas.

How much coverage do you have for perishables checkout? In other words, how many Prime households do you think you've covered with perishables checkout?

For those that have used it, how much of a boost do you think that is per Prime household? Is this something that increases spend per Prime household by 20%, 50%, 100%, just given how large the fresh grocery category is?

Thanks a lot.

Hey, Mark, let me start with the operating margin outlook for AWS. We're not guiding at all on our margin for AWS, but I can give you my color on it.

We've seen very strong op margin, 38.7% this quarter. I would add that the profitability we've been seeing recently is really not a random outcome, but it's a result of a lot of disciplined efficiency gains, capacity optimization, closely managing our fixed cost, and as Andy said, the inclusion of over time more and more custom chips into our cost structure.

However, there are other things bouncing around. In fact, the outsized growth of AI puts margin compression on the business as a whole because the margins are lower there.

We are happy with the progression in both core and AI margins, and, as Andy said earlier, they're ahead of where the core was at a similar point in time, but they're still lower. There's going to be that mix issue.

There's also headwinds from component pricing that we've talked about. Cannot predict it, but there's going to be always a confluence of these multiple factors that we're working very hard on the cost side or on the efficiency side and dealing with investments and new product margins as they start up and launch and reach scale.

On the perishables checkout, Mark, I share your enthusiasm about it. As we've had a big grocery business for a while, and it's mostly been in the middle aisles of what you'd find in a grocery store: consumables, canned goods, pharmaceutical items, beauty products.

We've had this leading organic grocery store in Whole Foods Market, which also is growing faster than the rest of the grocery segment, both in the top line and the bottom line, with a new format in Daily Shop in urban environments is doing really well that we're in the process of rolling out over the next couple years. We've always known, if we wanted to be able to serve customers the way we think they want to be served, and the way we want to, we have to have a mass offering, and it starts with perishables.

If you look at the total spend per household, when we look at the difference between the total amount of spend that we think our consumers and customers are spending on items that we could serve them on versus what they spend on us, the biggest difference is grocery. A lot of grocery starts with perishables.

We really are excited that we feel like we've found something with shipping perishables out of our same-day facilities. We've grown our everyday essentials about 3 times faster than the rest of our store.

It represents about half of our units now, if I look at all the everyday essentials. What we find is customers who shop perishables through our same-day delivery, they buy about 7 times more everyday essential units than those that don't.

Again, if you're trying to close that gap between what consumers spend overall and the items that we can provide versus what they spend with us, perishables has a huge impact on that. We also see that our grocery customers shop about 4 times more non-grocery categories, like home and apparel and a bunch of other categories, and spend about 2 times more on those categories compared to those who don't use us for grocery.

Those are big differences, and we believe that the progress we're making on perishables and grocery is going to ultimately lead to being more relevant to customers for more of their shopping missions.

Thanks so much for taking the question. Maybe a two-parter on market opportunities within AWS.

First, with respect to Bedrock, what have been your key learnings about the market opportunity around Bedrock and how it might continue to scale in the years ahead? Then with respect to the decision to sell Trainium externally, how do you think about balancing the market opportunity of external sales relative to the need for compute for that opportunity relative to other opportunities broadly across both internal use cases and AWS?

Thanks so much.

Yeah. On the Bedrock side, I would tell you, you've probably heard this from us multiple times over the last couple years, but one of the lessons that we've learned, I'll say for 20-plus years in offering technology to customers, is that there is not one tool to rule the world.

People always assume that whatever is in the lead early on is going to be the only thing or things that people use, and it's never the case. It's not true in database, hasn't been true in analytics.

In the beginning of AI, everyone said the only framework people use is TensorFlow, turns out, PyTorch is much bigger now. It's not going to be true with models, either.

I'll say in the next 2 to 3 years, I think there are going to be 6 to 8 models that are about as good as each other, or at least good enough for 95%-plus of the workloads. Some will be better reasoning, and some will be better at coding, and some will be better at visual pieces, and some will be lower latency, and some will be lower cost.

Most customers are going to use multiple models, and many will use multiple models in the same generative AI application. What customers really want to do, and we see this validated over and over again when we talk to customers, is they want a service that has really broad selection of those frontier models, and that has very good price performance, and that has very strong security and access control.

They want that service so that they can kind of decide what combination of models they want to use, and to swap them back and forth as they leapfrog each other, as different constraints of their applications require them. That's what we see with Bedrock.

People are very excited about it. We have both closed models and models like Anthropic and OpenAI's models.

We have open models with Kimi and GLM and others. I think the Western world is very hungry for additional open models that are Western open models, and so I believe that will happen over time as well.

I think that the size of the opportunity in Bedrock is gigantic, and it could be a service as large as EC2, which is a very large service today in AWS. I think it's very large.

On the balance of chips in what you sell externally versus what you use inside your cloud, it's a very good question, and it's a question that we're continuing to evaluate as we speak. I would say that the number of companies who have a significant cloud business and a chip strong enough to sell externally is effectively two.

There really aren't many customers that have to grapple with this same type of problem at this type of scale. At this point, everybody is producing their advanced chips with the same provider.

There's only so many chips that provider can provide each year. Everybody gets an allocation, and you have to decide with that allocation how much of it you want to sell externally, and how much of it you want to use in your cloud.

There's a trade-off. If you sell externally, you're able to realize more revenue in the short term because you get to recognize the revenue as soon as it leaves your inventory.

On the flip side, if you decide to use it inside your cloud in the form of EC2 instances or Bedrock, you actually wind up driving more absolute revenue and free cash flow from those chips because you get to use them over 5 to 6 years, and drive revenue for a longer period of time. You have to really think through what you want to do.

To date, I would say that we've taken a different approach in how we've used our chips than the other provider who has this same type of balancing act going on. As I mentioned earlier, we believe we're going to have a very significant business selling chips externally, and we're going to start in late 2027.

We will be thinking through that balance, I think, probably every year. They'll both be very significant.

Our ability to create price-performant, scalable chips is going to be a very significant strategic advantage for us at scale, both in our own cost structure as well as in building a separate business in selling external chips. Selling external chips really is a very different business than your cloud infrastructure business.

Thanks for taking the questions. When you think about the 40%-plus AWS revenue growth each of the next 2 years and then the multi-year path beyond that, can you just talk about the CapEx required to drive that kind of revenue over the next couple years and beyond?

Then also, Andy, any updates just to how you're thinking about the duration of returns on both servers and networking, and then also data centers? Thanks.

I'll take them in reverse order. On the break-even on our infrastructure, as I mentioned in my opening comments, we're continuing to move that left, we're continuing to make that shorter.

Core is a little bit different from AI, it's a little bit different from how you finance these things, and so it varies depending on which you're talking about, but we're continuing to make progress, as we historically have been pretty good at doing, if you think about AWS, at finding ways to extend useful lives and to make the break-even sooner, and we're making good progress there, too. On the CapEx involved to grow the way we're talking about, it's meaningful, and I think that, the way we think about it, Doug, is that this is a once-in-a-lifetime opportunity.

I think what we're seeing with AI, it's the biggest transformation in our lifetime. If the idea of being able to grow 39% with a $184 billion annual run rate, and to grow 40% a year on top of that in each of the next 2 years, and ultimately end up as a trillion-dollar-plus business, we're going to invest to continue being the leader.

We are the leader today in, I'll call it, the infrastructure and AI space, and would it be more convenient if this type of acceleration and opportunity happened over 25 years, and you could spread the CapEx over 25 years? It might be more convenient, but that's not the world that we live in, and that's not the opportunity.

We can fund this plan, and I think it will forever change the size of Amazon. It will forever change our revenue, it will forever change our free cash flow, and we're going to invest and we're going to pursue it to continue to be the significant leader like we are today.

Great. I thought I'd double down on the Bedrock question.

The token marketplace on Bedrock seems to be one of the hottest services in AI right now, and our understanding is the majority of these workloads are running on Trainium. As this becomes a bigger part of the overall AWS story, like you just mentioned, Andy, how might that impact both revenue growth, but especially margins at AWS as you kind of shift from selling bare metal to selling more inference as a service?

Could you just talk about that? Thank you.

Yeah. You're right that the majority and most of Bedrock today is being run on top of Trainium, and we do that in meaningful part because we really like the price performance that we're getting, and it allows us to both keep prices as low as possible for customers, as well as we like the margins associated with it.

If you think about margins in the cloud space, and you're looking at different players and what their cost structure looks like, there are a few things to pay attention to. The first is, for their data centers, how many of them are they actually buying or long leasing versus doing co-lo or neoclouds?

Neoclouds are about 3,500 basis points worse in operating margin than if you long lease them. Then you look at what is the hardware on top of which they're running.

If you're using the hardware that everybody else is using, you'll have a comparable cost structure to what everybody is doing. If you have the ability to use hardware that's different, you'll have an advantage.

I think when you look at what we're doing, most of our capacity is either owned or long leased, which enjoys better operating margins than what you'll find when people are mostly doing co-lo or neo, or a big chunk doing co-lo or neo. The fact that we're running on top of our own chip that's providing both advantage price performance and is more cost-effective for ourselves because we own it, I think gives us some structural cost advantages that I think are going to be very helpful to customers as we pass them back to customers, and helpful to the overall economics of AWS.