Amazon conference call: AWS encounters AI power bottlenecks! Self-developed chips become the key to breakthrough, with cost performance leading by 30%-40%

On August 1st, Thursday, Amazon announced its second-quarter financial report for this year. Although the company's total revenue and operating profit exceeded Wall Street's expectations, its cloud services (AWS), which are essential for its survival, showed weak growth and a sharp contraction in profit margins, raising deep concerns in the market about its potential loss of industry leadership. In the subsequent conference call, CEO Andy Jassy unusually admitted that there are supply bottlenecks in AI computing power, further exacerbating investor worries, leading to a 7% drop in the company's stock price in after-hours trading.

For Amazon, this is a "mixed bag" of a financial report that is enough to panic the market. The report shows that the core AWS business's sales this quarter grew by 17.5% year-on-year, reaching $30.9 billion. Against the backdrop of competitors growing at rates of 30% or even higher, this figure is clearly unsatisfactory to investors.

More fatal than the slowdown in growth is the sharp deterioration in profitability. The operating profit margin of AWS plummeted from a record 39.5% in the first quarter to 32.9%. Chief Financial Officer Brian Olsavsky explained that this is mainly due to the company investing more capital in chips, data centers, and electricity to support generative AI, leading to a significant increase in depreciation expenses. This essentially confirms to the market that in order not to be left behind in the AI race, Amazon is being forced to make an expensive gamble, at the cost of its once-proud "cash cow" business bleeding.

In response to analysts' sharp questions about whether it is "falling behind," Jassy gave the most impactful answer of the conference call. He candidly stated: "Our current demand exceeds our supply capacity." He further explained that the constraints exist at multiple levels, and "the biggest limiting factor is electricity." This statement indirectly confirms market speculation about its execution capability and the speed of scale expansion potentially lagging behind competitors.

Of course, Jassy also attempted to fight back. He spent a lot of time elaborating on AWS's defensive strategies, repeatedly emphasizing its advantages in security and operational performance, and implying that "you will see very different security outcomes with other players."

At the same time, he prominently showcased Amazon's "secret weapon"—the self-developed AI chip Trainium2, claiming that its "cost-performance ratio is 30% to 40% higher than other GPU suppliers on the market," and introduced a series of new tools including the AI programming assistant Kiro, attempting to prove to the market that AWS has the long-term endurance and innovative capability to win the AI marathon.

In the shadow of AWS, the impressive performance of Amazon's other businesses seems overshadowed. The company announced that this year's Prime Day membership event is "the largest ever," and its advertising business also achieved a robust year-on-year growth of 22%. By continuously optimizing its logistics network, significant improvements have been made in delivery speed and cost control However, even this seemingly stable business faces significant external risks. When asked about tariff issues, Jassy's attitude was extremely cautious. He stated that there has been no observed weakening in demand or significant price increases in the first half of the year, but it is impossible to predict the situation in the second half, especially when the pre-purchased inventory is depleted, and it remains unknown who will bear the rising costs. He candidly remarked: “It’s hard to know how tariffs will ultimately be determined... We don’t know what will happen.”

Here are the key points extracted from the earnings call:

• Profit engine stalls: Amazon Web Services (AWS) saw a year-on-year sales growth of 17.5% in the second quarter, which the market considers insufficient compared to competitors, raising concerns about its leading position. More critically, its operating profit margin plummeted from a record 39.5% in the first quarter to 32.9%.

• CEO acknowledges AI computing power bottleneck: CEO Andy Jassy rarely admitted during the call that AWS faces a situation of "demand exceeding supply" in the generative AI field, stating that "the biggest constraint is power," confirming that the company is encountering serious physical limitations in the AI arms race.

• High costs of the expensive AI arms race: The sharp contraction in AWS's profit margin is primarily attributed to the high depreciation costs resulting from large-scale capital investments to support AI business. This indicates that chasing the AI wave is significantly eroding AWS's profitability.

• Retail advertising remains robust, but tariff clouds linger: Although the retail business set records during Prime Day and advertising revenue grew by 22% year-on-year, Jassy's statements regarding the impact of tariffs were extremely cautious, stating that the future direction is "unknowable," adding significant uncertainty to the outlook for consumer business.

• AWS defense tactics clarified: In response to questions about being "behind," Jassy elaborated on AWS's differentiated advantages, including stronger security, operational performance, and functional depth, emphasizing the significant cost-effectiveness of its self-developed AI chip Trainium2, claiming it is "30% to 40% better than other GPU suppliers."

• Release of AI application suite tools: Amazon introduced several new AI applications and tools during the meeting, including the open-source method Strands for simplifying AI Agent construction, AgentCore for secure deployment of agents, and the highly anticipated new AI programming assistant Kiro, showcasing its ambition to penetrate from underlying computing power to upper-level applications.

• AI programming agent Kiro debuts: Jassy highlighted the new AI programming assistant Kiro, stating it has "generated a huge response," with 100,000 developers using the preview version within the first five days. He explained that Kiro's uniqueness lies in its ability to allow developers to engage in "vibe coding" with AI through natural language, and it can automatically generate and update technical specification documents, making it "easier to move from prototype design to production."

• Alexa+ Future Blueprint Clear: Andy Jassy defines the generative AI-driven Alexa+ as "an actionable personal assistant," far exceeding ordinary chatbots. He outlined a clear path for its commercialization through device sales, shopping, advertising, and even "subscription elements."

• "Starlink" Project (Project Kuiper) Progress Revealed: Jassy disclosed that, although not yet commercialized, the Kuiper project has signed "a considerable number" of agreements with corporate and government clients and plans to enter the commercial testing phase "later this year or early next year."

The following is the full transcript of the earnings call, translated by AI tools:

Amazon Q2 2025 Earnings Call

Event Date: July 31, 2025

Company Name: Amazon

Event Description: Q2 2025 Earnings Call

Source: Amazon

For more event information and transcripts, please visit {AMZN US Equity EVT BB-7300379-1 }

Q2 2025 Earnings Call

Presentation Segment

Operator:

Thank you for waiting. Hello everyone, and welcome to the Amazon.com Q2 2025 financial performance earnings call. All participants are currently in listen-only mode. After the presentation, we will conduct a Q&A session. Today's meeting is being recorded. Now, I will turn the call over to Mr. Dave Fildes, Vice President of Investor Relations, for opening remarks. Thank you, sir. Please go ahead.

Dave Fildes, Vice President of Investor Relations:

Hello everyone, and welcome to our Q2 2025 financial performance earnings call. Joining us today to answer your questions are our CEO Andy Jassy and our CFO Brian Olsavsky. As you listen to today's call, we recommend that you have our press release handy, which contains the financial performance, metrics, and commentary for the quarter.

Please note that unless otherwise stated, all comparisons in this call will be based on our performance for the same period in 2024. Our comments and responses to your questions reflect management's views as of today, July 31, 2025, and will contain forward-looking statements. Actual results may differ significantly. More information about factors that could affect our financial performance is included in today's press release and in the documents we have filed with the U.S. Securities and Exchange Commission (SEC), including our most recent 10-K annual report and subsequent filings.

In this call, we may discuss certain non-GAAP financial metrics in the press release, the slides from this webcast, and the documents we have filed with the SEC, all of which have been published on our investor relations website. You can find additional disclosures regarding these non-GAAP metrics, including reconciliations of these metrics to comparable GAAP metrics

Our performance guidance includes the order trends we have observed so far and the assumptions we currently consider appropriate. Our performance is inherently unpredictable and may be significantly affected by various factors, including fluctuations in foreign exchange rates, changes in the global economy and geopolitical conditions, tariffs and trade policies, as well as customer demand and spending, including concerns about economic recession, inflation, interest rates, regional labor market constraints, world events, the internet, online commerce, cloud services, and the growth rates of emerging technologies, as well as various factors detailed in our filings with the SEC. Our guidance assumes, among other things, that we will not engage in any additional business acquisitions, restructurings, or legal settlements. Accurately predicting the demand for our goods and services is impossible, and therefore our actual results may differ significantly from our guidance. Now, I will hand the call over to Andy.

Andy Jassy, President and CEO:

Thank you, Dave. Today, we reported revenue of $167.7 billion, a 12% year-over-year increase excluding foreign exchange impacts. Operating income was $19.2 billion, a 31% year-over-year increase, and free cash flow for the past 12 months was $18.2 billion. Over the past quarter, we have seen good progress in various customer experience and business aspects. Starting with our retail store business, we feel good about both the investment and output of this business. At Amazon, we think about our business in terms of inputs and outputs. Outputs are metrics like revenue or operating profit margin, but you certainly cannot manage at the output level. It is the inputs that drive the outputs. So, almost all of our internal discussions are about setting input goals.

In our retail store business, the most important inputs for customers are product selection, low prices, and delivery speed. Over the past few months, we have taken another step forward in product selection, most notably with the highly anticipated return of Nike products to Amazon retail stores. We have added high-end brands like Away, Aveda, and Marc Jacobs fragrances, as well as brands from Saks and Amazon, such as Dolce and Gabbana, Etro, Stella McCartney, and Rosetta Getty. We have also begun to expand our very successful fresh food pilot program, where we offer fresh products at the point of purchase when customers order other items that will be delivered from our same-day fulfillment nodes. We have seen strong customer acceptance, with 75% of customers using this service this year being first-time buyers of fresh products on Amazon, and 20% of customers making repeat purchases within the first month. Our prices continue to remain low and highly competitive for customers. This is also one of the reasons why our essentials business has grown globally more than other businesses and accounts for one-third of the total items sold. This is also why a well-known research firm has concluded for eight consecutive years that Amazon has the lowest prices among all U.S. retailers

But perhaps the clearest output is the growth rate of our retail store business last quarter and the success we achieved during the recent Prime Day event. This year's Prime Day was the largest ever, with record sales, number of items sold, and Prime member registrations in the three weeks leading up to Prime Day. Customers saved billions of dollars, and independent sellers—most of whom are small and medium-sized enterprises—also achieved their best sales performance in the history of Prime Day events.

There is still a lot of noise about what impact tariffs will have on retail prices and consumption. So far, most of it has been incorrect and misreported. As we mentioned before, it is impossible to know what will happen. What level will tariffs ultimately stabilize at, especially in China? What will happen when we exhaust our pre-purchased inventory, or when our sales partners' inventory that was deployed before the tariffs take effect runs out? If costs ultimately rise, who will absorb those costs? What we can share is what we have seen so far, which is that in the first half of this year, we have not seen a weakening in demand, nor have we seen significant price increases. Additionally, there are over 2 million sellers in our marketplace, each with different strategies regarding whether to pass on higher costs to consumers. This diversity gives customers an advantage when shopping on Amazon, as they are more likely to find lower prices on the items they care about. Further improving delivery speed remains a key focus, and we are making continuous progress. We previously shared how we restructured our U.S. inbound network into a regionalized structure, allowing us to place inventory closer to customers and transfer it, thereby increasing speed and reducing costs. This work is yielding tangible results.

In the second quarter, the share of orders processed through direct channels (packages delivered directly from fulfillment centers without additional transfer stations) grew by over 40% year-on-year. We also reduced the average shipping distance of packages by 12% and decreased the average handling times per item by nearly 15%. We have also made progress in order consolidation. As more products are deployed locally, we are able to pack more items into each box and reduce the number of packages per order. This helps increase the number of items per box and improves overall service costs. In summary, these improvements are making the network faster and structurally more efficient. We also set another global speed record in the second quarter, providing Prime members with delivery services at the fastest rate ever. In the U.S., the number of items delivered the same day or the next day increased by 30% compared to the same period last year. Items that customers previously had to pick up at nearby physical stores are now often delivered to their doorsteps within hours. We are also working to further improve delivery speed, regardless of where customers live. We recently announced plans to expand our same-day and next-day delivery services to tens of millions of U.S. customers in over 4,000 smaller cities, towns, and rural communities by the end of this year. This service is already available in over 1,000 such communities in the U.S. The early response from customers in these areas has been very positive They shop more frequently and purchase household essentials at a significantly higher rate. Automation and robotics are also important contributors to improving cost efficiency and driving better customer experiences. We have deployed our one millionth robot in our global fulfillment network and launched innovative technologies at our last-mile innovation center, such as automated package sorting and a transformative technology that delivers packages directly to employees at an ergonomic height. We introduced DeepFleet, our AI that improved robot driving efficiency by 10%, which is a significant advancement at our scale.

DeepFleet acts like a traffic management system, coordinating the movement of robots to find the best paths and reduce bottlenecks. For customers, this means faster delivery times and lower costs. For our team members, our robots handle more physically demanding tasks, making our operational network safer. This combination of robotics and generative AI is just beginning. While we have made significant progress, we are still in the early stages regarding what we will launch in the coming years. Next, let's talk about Amazon Advertising, which we are pleased with its strong growth, generating $15.7 billion in revenue this quarter, a year-over-year increase of 22%.

We continue to see strong momentum in our extensive full-funnel advertising product portfolio, which helps advertisers reach an average of over 300 million ad-supported audiences through our owned media resources in the U.S. alone. These media resources include our retail marketplace, Prime Video, Twitch, and Fire TV, as well as live sports broadcasts like the NFL, NASCAR, and NBA, along with third-party websites and applications. Another area we are excited about is our Demand-Side Platform (Amazon DSP). Our DSP enables advertisers to plan, activate, and measure full-funnel investments. We combine trillions of proprietary browsing, shopping, and streaming signals with extensive supply-side relationships and our secure data clean rooms, providing advertisers with the ability to optimize ads, achieve higher precision, and drive efficient and effective advertising outcomes. In June of this year, we announced a significant partnership with Roku, allowing advertisers to exclusively reach 80 million connected TV households through Amazon DSP, the largest verified connected TV reach in the U.S. This is a huge leap for advertisers, bringing top-tier planning, audience precision, and effectiveness to TV advertising. We also announced the integration of Disney's real-time advertising trading platform with Amazon DSP. This collaboration allows advertisers to directly access Disney's premium ad inventory on platforms like Disney+, ESPN, and Hulu while leveraging insights from both companies. When advertisers partner with Amazon, they are not just purchasing ad space; they are benefiting from exceptional programming content, innovative technology, and unparalleled signal, measurement, and audience development capabilities, providing strong relevance for consumers and high ROI for brands

Next, let's talk about AWS. In the second quarter, AWS grew by 17.5% year-on-year, and its current annualized revenue run rate has exceeded $123 billion. We continue to help organizations of all sizes accelerate their transition to the cloud, signing new agreements with companies including PepsiCo, Airbnb, Peloton, NASDAQ, London Stock Exchange, Nissan, GitLab, SAP, Warner Bros. Discovery, 12 Labs, FICO, Iberia Airlines, SK Telecom, and NatWest. In the rapidly evolving world of generative AI, AWS is continuing to build a large, fast-growing (with a year-on-year growth rate reaching triple digits), multi-billion-dollar business, and currently, our demand exceeds our supply capacity.

There are a few points to note. First, in terms of hardware, our self-developed AI chip Trainium2 is being deployed at a larger scale and has impressively become the backbone of Anthropic's latest generation Claude model as well as many of our core products (such as Amazon Bedrock). We also launched Amazon EC2 instances powered by NVIDIA Grace Blackwell superchips, which are AWS's most powerful NVIDIA GPU-accelerated instances.

Second, in Bedrock, we recently added Anthropic's Claude 4, which has become the fastest-growing model in Bedrock's history. We continue to see strong adoption of our own cutting-edge model Amazon Nova, which is now the second most popular foundational model in Bedrock. Nova's new features allow customers to customize their Nova models in ways that are not achievable with other foundational models, enabling organizations to inject their unique expertise into these models while optimizing cost and speed. As excitement grows around building agents, people are realizing the lack of tools to build them. In May of this year, we released Strands, an open-source way to build agents more easily, which has been well received by a wide range of customers, garnering 2,500 stars on GitHub and over 300,000 downloads. Customers have also faced challenges in securely and scalably deploying agents into production environments. This has hindered businesses from scaling agents. To help address this issue, Bedrock has just released AgentCore. AgentCore is a set of building blocks that provides customers with the industry's first secure serverless runtime for synchronous and asynchronous execution, agent identity and boundaries, memory services, a gateway that converts services into MCP-compatible interfaces, built-in code execution and web browsing tools, and an observability service. Customers are excited about AgentCore because it allows them to start deploying agents more broadly.

Third, you are starting to see AWS release more powerful applications at the top of the AI technology stack. AWS Transform is an AWS agent that dramatically shortens the timeline for modernizing mainframes from years to months, achieving an 80-fold increase in the speed of converting VMware to EC2, and making the migration from .net Windows to .net Linux simple, reducing the licensing costs of .NET applications by up to 40%. We have also just launched Kiro, our new agent-based integrated development environment programming agent. There has been a lot of buzz around Kiro, with hundreds of thousands of developers using or requesting access in the first few weeks, and 100,000 users in the first five days of the preview. For developers, what impresses them the most is that Kiro allows them to engage in "vibe coding," where developers chat with the programming agent in natural language to build code. However, unlike other programming agents, developers do not have a real structure to rely on, and Kiro allows developers to create a specification using natural language, which is then automatically updated as they continue vibe coding or interact with Kiro. This makes it much easier to go from prototyping to production. Customers also love Kiro's event-driven agent hooks, which act like an experienced developer, capturing things that developers might overlook. When developers save a React component, the hook updates the test file. When they modify API endpoints, the hook refreshes the readme file. When they are ready to submit, the security hook scans for leaked credentials. It is still too early for Kiro, but it is clear that we are doing things that customers love, and Kiro has the potential to change the way developers build software.

I often say this, but remember that 85% to 90% of IT spending globally is still on-premises, not in the cloud. In the next 10 to 15 years, this equation will flip, and companies' enthusiasm for leveraging AI will further accelerate this process. Therefore, AWS's significantly broader capabilities, stronger security and operational performance, and deeper experience in helping enterprises modernize their infrastructure all bode well for the future of AWS business. We are also seeing momentum in some other areas at Amazon. I'll just mention a few. We are excited about the progress of Alexa+, our next-generation assistant powered by generative AI. We have begun rolling out early access to customers in the U.S., and millions of customers now have access. We are seeing very positive feedback and will continue to iterate on the experience. We recently successfully completed the third launch of Project Kuiper. We have not yet commercialized this service, but a significant number of enterprise and government customers have signed agreements to use Kuiper. In terms of Prime Video sports live streaming, our first NASCAR season attracted about 2 million viewers per race, making it the youngest audience among NASCAR broadcasters in over a decade We recently announced our upcoming star broadcasting team for the first NBA season, including Eagle, Stan Van Gundy, Kevin Harlan, Dwayne Wade, Taylor Rooks, Blake Griffin, Dirk Nowitzki, Steve Nash, and Candace Parker. We also announced that Oscar-nominated director Denis Villeneuve will direct the next James Bond film. James Bond is in the hands of one of today's greatest filmmakers, and we can't wait to start 007's next adventure. Finally, we are very pleased with the growth and resonance of Amazon Pharmacy, which has achieved a year-to-date year-on-year growth rate of 50% on an already substantial base. Many good things are happening throughout the company. That's all from me, and I'll pass the call to Brian for a financial update.

Brian T. Olsavsky, Senior Vice President and Chief Financial Officer:

Thank you, Andy. Let's start with the financial results at the top line. Global revenue was $167.7 billion, a year-on-year increase of 12% excluding foreign exchange impacts. Due to the general strengthening of foreign currencies against the dollar, foreign exchange had a favorable impact of $1.5 billion on this quarter's revenue. Just a reminder, our revenue guidance for the second quarter had anticipated a negative impact of about 10 basis points or $100 million. Operating income was $19.2 billion, exceeding the upper end of our guidance range by $1.7 billion.

In our various business segments, we continue to prioritize cost-effective innovations that create value for customers. In the North America segment, second-quarter revenue was $100.1 billion, a year-on-year increase of 11%. International segment revenue was $36.8 billion, a year-on-year increase of 11% excluding foreign exchange impacts. The number of global paid units increased by 12% year-on-year. We remain focused on the investments that matter most to customers. In the second quarter, we saw broad advantages in key performance indicators. This includes highly competitive pricing, more spot supply, and record delivery speeds for Prime members.

Our millions of global sellers continue to be significant contributors to our vast selection. This helps customers find the products they need at competitive prices. Our investments in tools, services, and SaaS delivery speeds help our selling partners reach more customers and further expand their businesses. In the second quarter, the share of global third-party seller units reached 62%, the highest ever, up 100 basis points from the second quarter of last year. We are also closely monitoring the macroeconomic environment, including the impact of tariffs. As Andy mentioned, our plans for the second quarter considered a range of assumptions, but not all assumptions have materialized. We will continue to consider a range of assumptions for the future. Turning to profitability, the operating income for the North America segment was $7.5 billion, an increase of $2.5 billion year-on-year. The North America operating margin was 7.5%, up 190 basis points year-on-year The international department's operating profit was $1.5 billion, an increase of $1.2 billion year-on-year. The international operating profit margin was 4.1%, up 320 basis points year-on-year. We are satisfied with the strong execution of the operations team and the positive experiences they provide for customers.

In the second quarter, we saw an improvement in the productivity of our transportation network, thanks to better inventory placement, strong leverage from high product volumes, and higher levels of best-selling inventory from first-party and third-party sales partners. These factors contributed to faster delivery speeds and lower costs. Outbound transportation costs increased by 6% year-on-year and continued to grow at a significantly slower rate than the growth in the number of items shipped, which, as I mentioned earlier, increased by 12% year-on-year. We are committed to taking measures to further improve our cost structure.

Strategic inventory placement has brought multiple benefits, including better spot supply, shorter delivery routes, and faster customer delivery times. As we optimize inventory locations, we can consolidate more products into each package, thereby reducing packaging materials and costs. To achieve this, we will continue to improve our inbound network, expand our same-day delivery facilities in the U.S. (including in rural areas), and implement robotics and automation technologies in our facilities. While the year-on-year increase in operating profit margin may fluctuate, we have a clear strategy to achieve sustained long-term progress.

Turning to the advertising business. Advertising revenue increased by 22% year-on-year, primarily driven by sponsored products, as we saw strong traffic in our stores. Advertising remains an important contributor to profitability in both the North American and international departments. Our full-funnel advertising approach that connects brands with customers is resonating. Next is our AWS department. Revenue was $30.9 billion, an increase of 17.5% year-on-year. AWS now has an annualized revenue run rate of over $123 billion.

In the second quarter, we continued to see growth in both our generative AI and non-generative AI businesses as the company shifted its focus to new projects, migrating more workloads to the cloud, restarting or accelerating existing migrations from on-premises to the cloud, and leveraging the power of generative AI. AWS's operating profit was $10.2 billion. We did see the profit margin for the AWS department decrease from a record 39.5% in the first quarter to 32.9% in the second quarter. The largest driver of the quarter-over-quarter decline, accounting for about half, was an increase in seasonal equity incentive expenses due to the timing of our annual compensation cycle. AWS's profit margin was also adversely affected by increased depreciation expenses and fluctuations in foreign exchange rates year-on-year. Depreciation expenses are a result of our ongoing increase in AWS capital expenditures. As we have said in the past, we expect AWS's operating profit margin to fluctuate over time, partly due to the level of investment we are making at any given point in time. We will continue to invest more capital in chips, data centers, and power to pursue this exceptionally large opportunity we have in the field of generative AI.

Now turning to our cash capital expenditures (CapEx), which were $31.4 billion in the second quarter. We expect that the capital expenditures in the second quarter can reasonably represent our quarterly capital investment rate for the second half of this year AWS remains a major driver as we invest to support the demand for our AI services, increasingly investing in self-developed chips like Trainium, as well as the technological infrastructure supporting our North American and international divisions. Additionally, we continue to invest in our fulfillment and transportation network to support business growth, improve delivery speed, and reduce our service costs through investments in same-day delivery facilities, robotics, and automation. Overall, these investments will support growth for many years to come. Next is our financial guidance for the third quarter. Just a reminder, our guidance considers a range of possibilities, taking into account the results from the second quarter, the trends we have seen so far this quarter, and expectations regarding the macroeconomic environment, including tariffs. Net sales for the third quarter are expected to be between $174 billion and $179.5 billion. Based on current exchange rates, we estimate that the year-over-year impact of foreign exchange rate changes will be a favorable impact of about 130 basis points. Just a reminder, global currencies may fluctuate during the quarter. Operating income for the third quarter is expected to be between $15.5 billion and $20.5 billion. In this dynamic environment, we will focus on the most important things, providing exceptional value to customers through a wide selection, competitive pricing, and unparalleled convenience. We remain focused on driving a better customer experience and believe that putting customers first is the only reliable way to create lasting value for shareholders. That’s all for now, let’s move into the Q&A session.

Q&A Session

Operator:

Thank you. We will now begin the question-and-answer session. (Operator instructions) Our first question comes from Doug Anmuth of JP Morgan. Please go ahead.

Doug Anmuth, Analyst:

Thank you very much for taking my question. I have two questions. First, could you help us understand in more detail how tariffs are absorbed between suppliers, Amazon, and consumers, and whether you expect any changes in the future? Second, regarding AWS, we see that the second and third players in the field are achieving significantly faster cloud growth. I completely understand that AWS has a larger base. But beyond that, do you think this output gap is more due to customer demand or infrastructure supply? Thank you.

Andy Jassy, President and CEO:

Yes, I will address both of those questions. Let me start with tariffs. I think we have said multiple times and still believe that we just don’t know what will happen in the future. It’s hard to know what level tariffs will ultimately stabilize at, especially in China. It’s difficult to predict what will happen when we exhaust the inventory that we pre-purchased for our own first-party retail business, as well as the inventory that we see third-party selling partners pre-deploying. If costs rise over time, we are currently uncertain about who will ultimately absorb these higher costs. What we can tell you is what we saw in the first half of this year, we did not see demand weaken, nor did we see any significant increases in average selling prices (ASP). So the situation in the second half may change There are many things we don't know, but this is the situation we see at the moment.

Regarding AWS, the first thing I want to say is—just as you mentioned in your question, Doug—that the year-over-year growth percentage and growth rate are always a function of the base you are operating from. Our business scale in the AWS department is significantly larger than that of other companies. I think the second player is about 65% the size of AWS. Looking at the results over the past few quarters, sometimes, as far as we know, we have grown faster than others, and sometimes others have grown faster than us. But even if you look at the second player, it is still a fairly—we still have a significant market share lead.

In any case, these are really just snapshots in time. Last week was also a snapshot in time, and what really matters is the experience customers have operating on these platforms. If you look at what is important to customers, they care a lot about operational performance, availability, durability, and the latency and throughput of various services. I think we have a significant advantage in this area. They care a lot about security. If you have important data, and for most companies, they are putting their truly important data in the cloud.

The security and privacy of this data are very important, and the security results you see in AWS are very different from what you see with other players. Yes, you just need to—just look at what has happened over the past few months, and you can almost see issues arising with these players every month. So in terms of security, I think there is a very big difference. Then I think there is also a very significant difference in functionality; not only do we have more service capabilities in our core infrastructure, but I think if you look at our end-to-end products in AWS, in AI, from the bottom of the tech stack to the top, it is very different. So I am satisfied with the investments and services we provide to customers in both AI and non-AI areas. Our demand now exceeds our supply capacity. So we could have generated more revenue and helped more customers, and we are working very hard to change this outcome and the capacity we have. But this is still—looking at this business, it is a business with an annualized revenue run rate of $123 billion, and it is still in the early stages. I mean, how often do you get a chance to face a business with an annualized revenue run rate of $123 billion and still say it is in the early stages? This is a very unusual opportunity that we are very optimistic about.

Operator:

Thank you. Our next question comes from Mark Mahaney of Evercore. Please go ahead.

Mark Mahaney, Analyst:

Okay. I will start with AWS. Can you disclose the backlog number? Also, in the past, I know you have talked about supply constraints and hoped they would resolve themselves in the second half of this year. Is that still your intention? Are there any signs that the supply constraints will be resolved sooner or later? Then there is a long-term question about Alexa+, which I have experienced for a while. Andy, when you consider its potential for increasing user engagement, as well as the potential to leverage some service revenue, advertising, or perhaps a bit more retail sales per household, You are just reducing friction. Can you talk about - from a financial perspective, how do you think this will develop, and how might we see this reflected in the numbers? Thank you very much.

Dave Fildes, Vice President of Investor Relations:

Hi, Mark, I'm Dave. Let me first give you the backlog number. As of the end of the quarter on June 30, that number is $195 billion, an increase of about 25% year-over-year.

Andy Jassy, President and CEO:

Regarding the supply constraints for AWS and what we are seeing, as I mentioned, our current demand exceeds our supply capacity. I think - and you see some constraints, they exist in multiple areas. The biggest single constraint is power. But you will also see intermittent chip constraints, and then some components, and once you have the chips to actually manufacture servers, sometimes the next generation of chips comes a bit later than expected, and sometimes you get the chips, but the yield in manufacturing servers during mass production is not what you expected. So today we are dealing with a lot of these issues. This is true across the entire industry today. I don't think we can fully resolve the capacity we need to meet our demand in a few quarters. I think it will take several quarters, but I do expect improvements each quarter, and I am optimistic about that.

I think regarding the Alexa question, I want to first say that the experience of Alexa+ is so much better than our previous Alexa experience. She is much smarter than her former self and has much stronger capabilities. And I want to say, unlike other chatbots on the market today, which are good at answering questions but can't actually take actions for you, Alexa+ can take many actions for you, which is very compelling. So I can ask Alexa+ to play music or play a video, move my music from one device to another, or if I'm listening to a song from a movie, I can ask Alexa+ to actually play that movie clip, and it will pull it up on my Fire TV on Prime Video. Or if I have guests coming over, I can say: Alexa, close the curtains, turn on the lights on the porch and driveway, raise the temperature by five degrees, and then play dinner music. She accomplishes all of this just by using natural language. So she can take many actions, which is very appealing. As for what we are currently seeing, we have started rolling out Alexa+ in the U.S., and there are already millions of customers. The rest of the users in the U.S. will gain access in the coming months, with a broader international rollout starting later this year. Customers are really enjoying this experience. They recognize how much better it is than before. The ratings are very high, the usage is much broader than what they used before, and the number of calls they make has significantly increased. I think we will see benefits in many different areas. I think first, if you build the best personal assistant in the world, it has great practicality for customers, so it will be widely used. So this means that - people are excited about being able to purchase Alexa+-enabled devices from us, leading to a lot of shopping, which is truly a delightful shopping experience, and it will only get better I believe that over time, as people engage in more multi-turn conversations, advertising will have the opportunity to play a role in helping people discover new things and become a lever for driving revenue. Moreover, I think that as we continue to add features, there may also be some subscription elements that go beyond what we have today. Currently, Prime members can access Alexa+ for free, while non-Prime members pay $19.99 per month for Alexa+. So I think it's still too early to say, but we are very encouraged by the experience we provide, and you can be sure that we will continue to iterate on it.

Operator:

Thank you. Our next question comes from Colin Sebastian of Baird. Please go ahead.

Colin Sebastian, Analyst:

Thank you for taking my question. I would like to start with the international segment and the progress in terms of revenue and margins. I hope you can provide some more detailed insights into the drivers of both and the sustainability of these market efficiency improvements.

Then, Andy, you mentioned the Kuiper project. We haven't heard much about it recently. So could you review the timeline for service launch relative to next year's launch target, and what your long-term aspirations for this satellite network might be? Thank you.

Brian T. Olsavsky, Senior Vice President and Chief Financial Officer:

Thank you, Colin. This is Brian. I'll start with the question about the international segment. Yes, the international segment performed very strongly this quarter, both in terms of revenue growth and operating margin. The operating margin increased by 320 basis points year-over-year to 4.1%. If you look at it, this is a continuation of the strong progress we've seen over the past 10 quarters. Overall, during this period, we've seen nearly 700 basis points of growth. So this is really a story about two parts or, if I may say, two regions within the international segment.

One is mature countries like the UK, Germany, and Japan, where their margin situation has become similar to that of the US. So as they continue to grow, their contribution to profits will increase over time, which is what we are seeing. In this quarter, we saw strong productivity in the transportation networks of these countries, just like we saw in the US, which brought higher units per package and faster delivery speeds, all while lowering costs. So the theme of lower service costs, faster speeds, and better selection continues to develop internationally as well.

In our emerging countries, we are pleased with the progress we are making. We have certainly entered eight countries over the past five years, all at different stages of early investment and at different points in their journeys toward profitability. But they have all made very good quarter-over-quarter improvements in selection growth, increasing Prime memberships, and expanding our customer offerings. So I think you see again the continued improvement in these areas, which feels very good. Mature countries are continuing to grow and develop, very much like the US, while emerging countries, I want to emphasize, are all at very different stages of growth right now

Andy Jassy, President and CEO:

Regarding Project Kuiper, Project Kuiper is the low Earth orbit satellite constellation we are building and launching. There are 400 million to 500 million households worldwide without broadband connectivity. This means they cannot do many things we take for granted, such as online education, online business, shopping, or entertainment. There is indeed a digital divide that needs to be addressed. This is also true for businesses and governments, as they have some assets or needs that require visibility or connectivity, which they cannot obtain today due to the lack of broadband in many parts of the world. So the demand is high.

I want to say that as our constellation goes into space, there will be two companies that I believe will have low Earth orbit satellites of modern technology. One is the existing player in the market today, and the second will be Project Kuiper. I believe we will have significant differentiation in performance here. If you look at the performance of the uplink and downlink that I expect, I believe Project Kuiper will have an advantage. I also think pricing will be very attractive for customers. Then I believe that if you consider the three key customer groups that want to use low Earth orbit satellites—consumers, businesses, and governments—given our consumer business and AWS business, we have very strong relationships with all three customer groups. Moreover, I think if you consider businesses and governments, many of the things they want to do after obtaining data from space are actually to put it in the cloud for analysis, data analytics, AI, and various operations. The fact that Project Kuiper and AWS are so seamlessly connected is very attractive to businesses and governments.

Now I am a bit surprised that we haven't launched Project Kuiper yet, but the number of agreements signed for businesses and governments to use Project Kuiper is impressive. So we are working very hard to get the satellites into space. Some of our rocket suppliers have some delays, but most of the available rocket launches in the coming years belong to us. We are very hopeful to bring this service into commercial testing later this year or early next year.

Operator:

Thank you. Our next question comes from Brian Nowak of Morgan Stanley. Please go ahead.

Brian Nowak, Analyst:

Thank you for taking my question. Andy, I have two questions for you about AWS. They are a bit tricky, but I still want to ask you. So there is a saying in the Wall Street financial circles that AWS is falling behind in generative AI, with concerns about market share being lost to peers, etc. Can you respond to this statement? What is your rebuttal to this? And tell us what the most important focus is for you and your team to ensure AWS maintains its advantage in competition with hyperscale cloud providers?

And then the second question, I know AWS is a big business, but is there any reason to believe that, given the scale of opportunity and all the generative AI capabilities that will be coming our way in the next 12 months, it shouldn't accelerate growth in the second half of this year and into 2026?

Andy Jassy, President and CEO:

Yes. Regarding the first question about AI, the first thing I want to say is that I think it's still too early in the AI space. If you look at what's really happening in this field, you'll find that it's very top-heavy. So, you have a small number of very large frontier models being trained, which are computationally expensive, some of which are trained on AWS, while others are trained elsewhere. Then you have, I would say, a relatively small number of very large-scale generative AI applications. One category is chatbots, with ChatGPT being the largest by far; another category is what I call programming agents. Companies like Cursor, Lovable, and similar ones, some of which are running significant parts on AWS. Then you have a large number of generative AI applications in pilot mode, either in pilot or under development, and many agents where people are starting to try to build and manage to scale them into production, but they are all— they are all still very early. And many existing applications— they are important, but in terms of usage, they are relatively small compared to those top-heavy applications I mentioned earlier. We have a lot of enterprises and startups running applications on AWS's AI services. And— but they all— again, to emphasize, the usage, the breadth of use cases, the extent to which people are putting them into production, and the number of agents that will exist are all still in a much earlier stage than what will be in the future. So when you think about what will be important in AI, what will be— when customers consider what infrastructure to use, what they will care about, I think you have to look at the different layers of the technology stack.

I think for those who are both building models and also if you look at where the real costs are, they will ultimately show up in inference. Today, so much of the cost is in training because customers are really training their models and trying to figure out how to put applications into production. But after scaling, 80% to 90% of the costs will be in inference because you only train periodically, but you are continuously outputting predictions and inferences. So they will be very concerned about the compute and hardware they are using. We have a very deep partnership with NVIDIA, and that will be the case for the foreseeable future, but we have seen this in the CPU space with Intel, where customers are eager for better cost-performance ratios. So just like in the CPU space, we built our own self-developed chip, Graviton, which has about a 40% better cost-performance ratio than other leading X86 processors. We have done the same thing in AI with our self-developed chip, Trainium. Our second-generation Trainium2 has really— it has become the backbone of Anthropic's next-generation models, which are being trained on it, and it has also become the backbone of Bedrock and the inferences we are doing. So I think a lot of inference has about a 30% to 40% better cost-performance ratio than other GPU suppliers currently on the market, and we are already developing our third-generation Trainium So I believe that a lot of computation and inference will ultimately run on Trainium2.

Moreover, I think that when people scale, that kind of cost-effectiveness will be very important to them. Then I want to say that the middleware of the tech stack really — it’s a combination of a series of services that customers care about in order to build models, then leverage existing leading-edge models, and then build high-quality generative AI applications for scalable inference. We see that for those building models, they continue to use SageMaker AI very broadly, and when you leverage leading-edge models, Bedrock is also growing significantly. As I mentioned in my opening remarks, the number of scaled agents is still very small in the context of what is to come, but part of the issue is that actually building agents is difficult, and deploying these agents in a secure and scalable way is also challenging. So I think the features we recently launched in Strands that make it easier to build agents, as well as the features in AgentCore that make large-scale, secure deployment easier, have received very positive feedback, and customers are excited about how this will change the possibilities for agents. Yes. Then I think — you have a very large number. I mean, remember, 85% to 90% of global IT spending is still on-premises. If you believe this equation will flip, and I believe it will, and we believe it will. You have a lot of legacy infrastructure that needs to be migrated. These are mainframes, these are instances, and when we build agents like AWS Transform, it makes migrating mainframes to the cloud much easier, migrating VMware to the cloud much easier, migrating .NET [Windows to .NET](http://Windows to .NET) Linux to save costs much easier, which is very attractive for enterprises. Or things like Kiro that allow customers to develop in a more straightforward and organized way, that’s why I think people are excited about this. So I really like the investments and services we are building in the AI space today; customers love them and resonate with them. I still think we are in a very early stage regarding AI and its applications. But one more thing I want to say is, remember, because we are now at a stage where a lot of activity is training and figuring out how to put your generative AI applications into production, people are not yet paying as close attention as they will in the future to ensure those generative AI applications run in the same place as the rest of their data and infrastructure. Remember, a lot of generative AI inference will just be another building block, like compute, storage, and databases. So people will actually want to run these applications close to where other applications are running, near where their data is, and there are far more applications and data running on AWS than anywhere else, so I am very optimistic about what will happen with AWS in AI as we scale. And I think we have a unique set of services from the ground up. I want to mention that regarding our expectations for accelerated growth, we do not provide guidance by business unit So I won't try to guide you, but I do believe that as more companies restore their modern infrastructure and migrate to the cloud, coupled with AI accelerating as more companies deploy more AI applications into production and begin to scale, along with my belief that there will be more capacity coming online in the coming months and quarters, all of this makes me optimistic about AWS's business.

Operator:

Thank you. The next question comes from Ron Josey of Citigroup. Please go ahead.

Ron Josey, Analyst:

Great. Thank you, Andy. That was really helpful. Perhaps asking the same question, but this time focusing on Amazon internally. I remember you wrote an article or blog in June discussing the ability or potential of using generative AI agents to improve internal efficiency and speed to market. So, I would love to hear your thoughts on how Amazon is adopting generative AI internally and how you might view the increased speed to market resulting from everything you just mentioned. Thank you.

Andy Jassy, President and CEO:

Yes. I believe AI is the biggest technological transformation of our lifetime, and that’s a heavy statement because even in our relatively short lives, we’ve experienced the cloud, mobile internet, and the internet itself, but I think it will be the biggest technological transformation of our lifetime. And I believe it will not only change every customer experience we know and deliver customer experiences we previously only dared to dream of, but it will also significantly change the way we work.

If you think about it, the way we code, the way we do analysis, the way we conduct research, the way we handle finance and measurement, I mean, really, every area of how we work could be meaningfully impacted by AI. And I think when you encounter such a significant shift, you have two macro choices. You either decide to embrace it, help shape it, and figure out how to build the right tools to leverage this technology; or you can hope it goes away and let it shape you. The post you mentioned, Ron, that I wrote was to make it very clear to the team that we are going to take the former approach. We are going to embrace it and work to shape it. So we have already built some tools and agents internally within the company. Things like, if you think about Kiro and the opportunity to have programming agents do a lot of coding, that’s very compelling. It will enable our team members to start from a higher level and innovate for customers faster and more broadly. If you think about services like Connect, which is our AWS service that provides call center software with a lot of built-in AI, it changes the productivity of all your customer service agents. You can imagine that in various ways, we will do a variety of things to make software easier to release, build high-quality software, conduct operations and quality assurance, and automate many of the business process coordination happening within the cloud. We will work hard in all these areas, primarily to enable ourselves to innovate for customers faster and more broadly And I also believe this will make the work of all our team members more enjoyable, as they will not have to do as much repetitive and tedious work as we are all doing now. They will be able to have more ownership of the problems they are trying to solve for customers. What we want are those who hope to have as much end-to-end ownership as possible, deep owners.

Operator:

Thank you. Our last question will come from Justin Post of Bank of America. Please go ahead.

Justin Post, Analyst:

Great. Thank you. I just have a question about the revenue guidance. The growth in the third quarter looks quite strong. I know you won't say whether AWS is expected to accelerate growth, but could you talk about some of the drivers and what kind of tariffs and other unexpected situations you are considering in that? And perhaps discuss your outlook for the fourth quarter? Thank you.

Brian T. Olsavsky, Senior Vice President and Chief Financial Officer:

Yes, hello, Justin. This is Brian. I'll take that question. Yes, our guidance is $174 billion to $175 billion—sorry, it's $179.5 billion, I misspoke. We are pleased with the growth rate we achieved in the second quarter and the acceleration in some areas, including the number of items sold, and we had a very successful Prime Day earlier this month. So, there is uncertainty about where tariffs will stabilize and the ultimate impact on consumers, as Andy mentioned earlier. But we are very confident in the key inputs.

We control pricing, selection, and convenience. We talked about improvements in delivery speed. We talked about increased selection, high inventory levels, and well-distributed inventory close to customers. So we believe all of these factors are favorable for us. So I would say we are cautiously optimistic. I won't provide any guidance for the fourth quarter right now, but we will talk about that next time.

Dave Fildes, Vice President of Investor Relations:

Thank you for joining our call today and for your questions. A replay of the meeting will be available on our Investor Relations website for at least three months. We appreciate your interest in Amazon and look forward to speaking with you again next quarter.

Operator:

Ladies and gentlemen, that concludes today's conference call. You may now disconnect. Thank you for your participation