Great. I think I'm going to leave the China questions to others, and focus on AWS and kind of AI. So, Andy, it seems like you've been bringing on a lot more P5 GPU instances since February from what it looks like to kind of support all these new AI workloads. So, how would you characterize in the first quarter and maybe here in the second quarter, the kind of supply/demand imbalance that you talked about before around AI workloads? And when do you think that AWS will be in a position to kind of capture enough AI revenue to drive acceleration? Is that something that could happen this year? Do you see that more like next year, given your capacity constraints? Thank you very much. Andy Jassy (CEO): Thanks, Ross. We've been adding many P5s, a type of NVIDIA chip instance, and are rapidly acquiring more Trainium 2 instances. Currently, our AI business is generating a multi-billion dollar annual run rate and is experiencing triple-digit growth year-over-year. As soon as we increase our capacity, it gets utilized immediately. I believe we could assist more customers and generate additional revenue if we had greater capacity. We have many Trainium 2 instances on the way, as well as the next generation of NVIDIA instances arriving soon. However, there are challenges in the supply chain, particularly with motherboards and other components, due to high demand. I am optimistic that supply chain and capacity issues will improve as the year progresses. null (Operator): And the next question comes from the line of Eric Sheridan with Goldman Sachs. Please proceed with your question.

Thanks so much for taking the question. Maybe I could ask a two-parter. First, in terms of strategy Andy, how do you think about positioning the company for the medium term, given all the levels of uncertainty out there about how the global trade environment might shift in the coming months? What do you see as the key strategic priorities that will allow the company to sort of be able to capitalize one way or another, depending on various elements of outcome? And how do you prioritize those investments in the months ahead? And then with respect to the one-quarter forward operating income guide, is there anything in there from a cost side that we should be thinking about as purely aligned with those types of investments against the trade landscape that might not repeat either later into this year or next year? Thanks so much. Andy Jassy (CEO): Thanks, Eric. It's currently difficult to predict the outcomes of tariffs and when they will stabilize. Consequently, our short and medium-term strategies align with our long-term goal of providing the widest selection of products at the lowest prices. This focus on maintaining low prices is crucial now more than ever, and we are dedicated to ensuring quick delivery and excellent customer service. Our initiatives reflect this commitment, including proactive inventory purchases as a first-party seller and encouraging our third-party sellers to stock up on their items. Our extensive selection compared to other retailers enables us to meet customer needs despite changing trends. Past experiences, such as during the pandemic, illustrate that customer preferences can shift unexpectedly, including a preference for brands that offer better pricing. Additionally, with over 2 million sellers, not every seller will respond to higher tariffs in the same way. Some may choose to absorb these costs instead of passing them on to consumers. This diversity gives our customers better access to a variety of products at competitive prices. Furthermore, over the past six years, we've strategically diversified our manufacturing locations, reducing our reliance on China for components for AWS and our devices. This diversification was a wise decision to ensure we protect our customers' interests in the short, medium, and long-term by focusing on broad selection, low prices, and fast delivery. Brian Olsavsky (CFO): Eric, I'll take your question on the guidance, and especially on operating income. I think that was your question, and the cost that might be in Q2. The thing I'd point to again is what I mentioned earlier, the stock-based comp always steps up generally in Q2 versus Q1, and then resets a rate that carries through for the next four quarters. You can look at historic trends to get an idea of that. Secondly, we do have some additional Kuiper launch costs in Q2. You saw a launch happen this week. And a reminder that we expense those launch costs until the point of commercialization, which a plan is to have that be later in this year. null (Operator): Thank you. And the next question comes from the line of Justin Post with Bank of America. Please proceed with your question.

Great. Thanks. I'll go back to AWS. I know in the past you said revenues can be lumpy. Can you explain why they might fluctuate up and down if it's beyond just capacity? And you see the competitors with some pretty good growth rates. How do you think about the difference there? Obviously, your dollar growth is very good, but how do you think about the difference there versus some of your competitors? Thank you. Andy Jassy (CEO): When we say that revenue can be inconsistent, we've noted this even before the recent developments in AI. The key factor here is the sales cycle, especially for enterprises. For startups, the goal is to have capabilities that make them want to run on your platform. Over the past 10 to 15 years, most successful startups have chosen to operate on AWS. However, it's unpredictable when these startups will achieve product-market fit and experience significant growth. This unpredictability also applies on the enterprise side, albeit in a different manner. Enterprises often take time to convince themselves to transition from on-premises solutions to cloud services and to determine if our solution is suitable for them. They select specific projects for initial experience, sometimes utilizing systems integrators, corporate services, or managing it themselves. This sets the stage for subsequent migrations, which vary in their pace; some companies move quickly while others take longer. Often, companies become excited about the cost savings and innovation speed that cloud migration offers, and what was expected to be a small next step often escalates into a larger initiative. Predicting this is challenging as it depends on how enterprises choose to allocate resources and sequence their projects. Adding AI into the mix presents its own rapid growth cycle, especially for certain use cases that evolve rapidly. Initially, we noticed a focus on AI initiatives aimed at productivity and cost management, which many AWS customers have pursued, and we are also implementing similar strategies within Amazon. Additionally, there has been significant large-scale training with many initiatives running on our Trainium 2 chip on AWS, as well as several influential chatbot developments. Recently, there has been a notable surge in coding agents, with companies like Cursor and Vercel seeing substantial growth while running largely on AWS. This level of growth was unexpected. Consequently, revenue can fluctuate significantly without prior notice. Some periods may yield sharp increases that are difficult to forecast, followed by stable growth rates that may not match the prior explosive phases. Regarding AI, it's fascinating to realize that we are just beginning to explore the various customer experiences and agents that will emerge, revolutionizing many functions. Despite our current efforts, I feel we're only at the very beginning stages. To assess relative growth rates, it's essential to understand that year-over-year growth is influenced by the percentage increase relative to the existing base. Our technology infrastructure base is considerably larger than that of others. Even achieving a 17% year-over-year growth rate on a $117 billion revenue run rate reflects significant progress. I believe we could achieve even more with greater capacity, and I anticipate that this capacity constraint will ease over the coming months. null (Operator): Thank you. Our next question comes from the line of Doug Anmuth with J.P. Morgan. Please proceed with your question.

Thanks for taking the questions. One for Brian, one for Andy; Brian, just maybe to follow-up on AWS, but more on the margin side, we've seen a lot of fluctuation over the last couple of years and now hitting almost 40%. Maybe you can just talk about what's driving the outperformance and then how we should think about normalized margins going forward? And then, Andy, your comments on Alexa about moving to more complex tasks, can you talk about that more and just with Alexa, the products have been around for a long time, they've had different use cases. How do you get users to shift their behavior more with Alexa? Thanks. Brian Olsavsky (CFO): Thanks, Doug. I'll address your first question. We had a strong quarter in AWS, particularly noted for the margin performance. This can be attributed to the robust growth we're experiencing alongside our continued investments in innovation and technology. For example, we're focusing on software and process improvements that optimize our server capacity, which reduces infrastructure costs. We've also been developing a more efficient network using our low-cost custom networking equipment. We're working to improve power usage in our existing data centers, which not only lowers costs but also allows us to use power for newer workloads. Additionally, advancements in custom silicon, such as Graviton, are contributing to lower costs for both us and our customers, enhancing price performance for them. However, margins are influenced by various factors, including our investment levels, competitive pricing, and the evolving mix of Generative AI services as they scale in the coming years. We have substantial infrastructure investments underway and planned for the second half of the year, which will begin to show results. We're pleased with the team's performance in generating cost savings, and it's a significant focus for us to expand services and features for customers. Andy Jassy (CEO): We are really excited about Alexa+. As I mentioned earlier, she is much more intelligent, capable, and able to take real action. To date, most agents have primarily been able to answer questions, which was impressive when they first launched. However, the future of agents is not just about intelligence but also about taking action. This requires a robust model and the ability to synchronize that model with executing the right APIs, or else results can be suboptimal. We have put a lot of effort into Alexa+. We began rolling it out several weeks ago and currently have over 100,000 users, with more set to be added in the coming months. So far, the feedback from customers has been extremely positive. People are excited about it and find that it offers much more functionality than previous versions of Alexa. We are fortunate to have over half a billion devices in homes, offices, and cars, providing us with a broad distribution. However, there will be some adjustment for users as they adapt to new patterns. For instance, users are no longer required to say 'Alexa' before each command. I've been using the alpha and beta versions for several months, and it took me a bit to realize I didn't need to keep saying Alexa, which is refreshing. You can do things like preparing for guests by saying, 'Alexa, please open the shades, turn on the lights in the driveway and porch, increase the temperature by five degrees, and pick some mellow dinner music,' and she just takes care of it. Experiences like that encourage users to engage more. When I was in New York for our announcement, I asked her for recommendations for great Italian restaurants or pizza places. She provided a list and even offered to make a reservation, which she did after I agreed. Once you experience these kinds of routines, they become extremely useful, almost like having a personal assistant, something most people don’t typically have. As more people get accustomed to these features, they will realize all that Alexa can do. We also plan to introduce more functionality in the coming months. null (Operator): Thank you. And our final question will come from the line of Brian Nowak with Morgan Stanley. Please proceed with your question.

Thanks for taking my questions. Hey, Andy and Brian. So, I have one for each of you guys. So, Andy, you have a very complicated retail business with a lot of moving pieces to it. I imagine you have a lot of good data on what you expect demand to be over the course of the year and the holidays and things. As you kind of step back and analyze the business and the tariff uncertainty, can you just sort of walk us through the one or two key areas operationally that you're most focused on just to ensure that Prime Day, Thanksgiving and the holidays go as smoothly as they possibly can? And then, secondly, on Brian, just to kind of go back to Eric's earlier question, as we think about the 2Q EBIT guide, are there any one-time costs or sort of tariff related costs in there similar to that $1 billion that you called out in the first quarter? Thanks. Andy Jassy (CEO): Thanks, Brian. On the retail question, I would say that the areas that maybe we're most focused on to make sure we have not just a great Prime Day, but Prime Day is just one event as you know, and so is peak. We're trying to be great all year long for customers. The obvious ones are making sure that we help our sellers however we can, because there's uncertainty for our sellers as well. So, we're trying to make sure we provide a great experience. We're trying to make sure that we have the right diversity of sellers and low prices for our customers. I think all that you have to also be very thoughtful around how much inventory you bring into your fulfillment nodes at any one time, because you can imagine scenarios where either on your own when you're the first party seller or lots of third parties want to get as much inventory in as early as possible trying to beat a deadline on what may happen. And if you end up with too much inventory in your fulfillment network, it really slows down your productivity and your ability to get things out as quickly as you want for customers at the cost structure you want. So, being able to manage that thoughtfully, we've learned that over the years, and I think the team is doing a good job of balancing that right now. Brian Olsavsky (CFO): And Brian, on your question about Q2, I guess I'd just reiterate what I had said earlier. We have stock-based comp step up, which I think you can see the normal pattern for that if you look at our history. We have additional Kuiper expenses. Specifically, you asked about tariffs. We do have tariffs that we'll be paying on retail purchases based on current tariffs. It's not large in Q2. We had done a lot of pre-buying of inventory in Q1, as I mentioned earlier. But just generally, I think with the uncertainty, we've added a bit to the range that we've given you. We generally have a wide range, but just the general uncertainty that we're seeing and uncertainty of consumer demand and everything else is causing us to increase the ranges a bit. So, we'll see. We feel an informed view of Q2 right now. As Andy mentioned earlier, we saw actually some strength in April based on what could end up being some pre-buys of a number of things, but advertising has been strong. So, we think there's a lot of positive trends, but certainly uncertainty right now for the quarter. null (Operator): Thank you. And our final question comes from the line of Brent Thill with Jefferies. Please proceed with your question.

Thanks. On AWS, I'm curious if you could give us the backlog number? And Andy, to your point about many of these core workload still yet to come to cloud, can you just update us on what you're seeing, are you seeing enterprise spring back, or are you seeing some confusion with AI clouding that transition and the timing of that? Just give some perspective on what you're seeing on migration. Thank you. Dave Fildes (Head of Investor Relations): This is Dave. I'll address the backlog, which stands at $189 billion for Q1, reflecting a 20% increase year-over-year. The weighted average lifespan of this backlog is 4.1 years. Regarding AWS, the situation with workloads that have yet to migrate reflects a previously aggressive trend toward enterprises moving away from on-premises infrastructure due to benefits like innovation speed, developer productivity, and cost savings associated with the cloud. During the pandemic, entities focused on cost optimization, including ourselves. As we came out of that phase, we witnessed a surge in generative AI, which made enterprises eager to explore workload opportunities in that space, particularly given its public interest. Over the past 16 to 18 months, companies have recognized the necessity to engage in both AI initiatives and cloud transitions. They are piloting various AI projects, some of which will succeed and others that may not, with successful ones set for scaling. However, many enterprises have additional AI initiatives still pending, either due to a lack of skills for implementation or waiting for costs to decrease, which they will. The high costs related to AI won't stabilize until we see ongoing reductions in inference costs. Meanwhile, companies are realizing that by not migrating their infrastructure to the cloud, they are missing out on easy opportunities. We're observing a renewed commitment to cloud migration plans. While transforming infrastructure from on-premises to cloud is typically a multi-year endeavor, some companies are faster than others, though they tend to proceed with caution to ensure application stability during the transition. We're experiencing significant success in dialogues with companies ready to transform their infrastructure on AWS, and we expect this trend to continue. Thank you for your time today, and for your questions. A replay will be accessible on the Investor Relations website for at least three months. We appreciate your interest in Amazon and look forward to our conversation next quarter. null (Operator): And ladies and gentlemen, that does conclude today's teleconference. You may disconnect your lines at this time. Thank you for your participation.