Amazon conference call: AI is a "once-in-a-lifetime" business opportunity, capital expenditures may reach $100 billion by 2025, AWS growth constrained by supply chain

Amazon announced its earnings report after the market closed on Thursday, showing that despite exceeding expectations for the holiday season in the fourth quarter, the outlook for the first quarter is poor, leading to a more than 7% drop in the company's stock price after hours.

At the same time, Amazon's capital expenditures in the fourth quarter were significantly higher than analysts' expectations and set a record. The company's capital expenditures are expected to reach $100 billion in 2025, indicating that Amazon is increasing spending to support the development of its AI business.

Summary of key points from the earnings call:

1. Capital expenditures may reach $100 billion in 2025, with the majority allocated to AI and cloud services AWS: The capital expenditure level of $27.8 billion in the fourth quarter will be "relatively representative" in 2025. Based on this trend, capital expenditures in 2025 are expected to be around $100 billion. The "vast majority" of this capital expenditure will be used for AI and cloud services AWS. Similar to 2024, most of the spending will be aimed at meeting the demand for AI services and supporting the technological infrastructure for North American and international businesses.

2. Cost efficiency will drive AI demand: The cost of inference will decrease, which is good news for tech companies. When the costs of technological components like AI inference decrease, it does not mean that companies will reduce their overall spending on technology. On the contrary, there is excitement about building new things that were previously expensive, leading to increased overall spending on technology. This trend is expected to boost overall AI demand.

3. Planning such large-scale expenditures only after seeing demand: CEO Andy Jassy stated that they will not make purchases without seeing significant demand signals. When AWS expands its capital expenditures, especially in once-in-a-lifetime business opportunities like AI, I believe this is a very good signal for the medium to long-term development of the AWS business. In fact, I think it is wise to invest this capital to pursue this opportunity because we believe that nearly all applications we know today will be reinvented with AI integrated, and inference will become a core building block like computing, storage, and databases. AI is undoubtedly the biggest opportunity since cloud computing and may represent the largest technological transformation and opportunity in business since the internet.

4. AWS growth is constrained by supply chain issues, which will gradually ease in the second half of 2025: The constraints are primarily reflected in several aspects: First, the speed of chip supply from our third-party partners has slowed down compared to before, and many midstream changes take time to produce the healthy and high-quality servers we expect. Second, regarding our own hardware and the launch of the Trainium 2 chip, we just introduced it to the general market at re:Invent, but most of the shipments will actually come in the next few quarters and months. Additionally, there are power constraints, as the world is still limited in power supply. These constraints are expected to gradually ease in the second half of 2025, and although our current growth is already quite good, I believe we could grow even faster

5. Depreciation will affect profits: Reducing the lifespan of some servers and network equipment will impact profits in 2025. Increasing the lifespan of some fulfillment center heavy equipment will add approximately $900 million to profits in 2025. Starting from January 2025, the lifespan of some servers and network equipment will be shortened from 6 years to 5 years. We expect this will reduce operating profits for the fiscal year 2025 by approximately $700 million.

6. Comprehensive application of AI: Artificial intelligence is being used across various business areas, including customer service, product listing, inventory management, and robotics. New experiences: Providing a new shopping experience through features like Rufus (AI shopping assistant) and Amazon Lens. We are using it in our retail business, and every business we are in is doing the same. We have built or are building approximately 1,000 different generative AI applications.

Below is the full transcript of the conference call (translated by AI):

Host: Thank you for waiting. Hello everyone, welcome to Amazon's fourth quarter 2024 financial performance conference call. Currently, all participants are in mute mode. After the presentation, we will have a Q&A session. Today's conference call is being recorded. Next, I will hand the meeting over to Mr. Dave Fildes, Vice President of Investor Relations, for his opening remarks. Thank you, Mr. Fildes, please go ahead.

Dave Fildes, Vice President of Investor Relations: Hello everyone, welcome to our fourth quarter 2024 financial performance conference call. Today, our CEO Mr. Andy Jassy and our CFO Mr. Brian Olsavsky will answer your questions. As you listen to today's meeting, we recommend you have our press release ready, which contains our financial performance as well as metrics and comments for this quarter. Please note that unless otherwise stated, all comparisons in this meeting are against the results for the same period in 2023. Our comments today and responses to your questions reflect management's views as of February 6, 2025, and contain forward-looking statements. Actual results may differ significantly. Additional information regarding factors that may affect our financial performance has been 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 annual report (Form 10-K) and subsequent filings.

During this meeting, we may discuss certain non-GAAP financial metrics. In our press release, the slides for this webcast, and the documents we have filed with the SEC (all of which have been published on our Investor Relations website), you will find additional disclosure information regarding these non-GAAP metrics, including reconciliations of these metrics to comparable GAAP metrics. Our guidance incorporates the order trends we have seen to date and assumptions we believe are appropriate today. Our performance is inherently unpredictable and may be significantly affected by many factors, including fluctuations in exchange rates, changes in global economic and geopolitical conditions, customer demand and spending (including recession concerns, inflation, interest rates, regional labor market constraints, world events, the internet, online commerce, cloud services, and the growth rates of emerging technologies and new technologies), as well as various factors detailed in the documents we have filed with the SEC Our guidance assumes that, among other matters, we will not complete any additional business acquisitions, restructurings, or legal settlements. It is not possible to accurately predict the demand for our goods and services; therefore, our actual performance may differ significantly from our guidance. Now, I will hand the meeting over to Andy.

Andy Jassy, President and CEO: Thank you, Dave. Today, we reported revenue of $187.8 billion, a year-over-year increase of 10%. Due to the continued strength of the dollar this quarter, we faced greater foreign exchange headwinds than expected, amounting to $700 million. Without this headwind, revenue would have grown 11% year-over-year, exceeding the high end of our guidance. Operating income was $21.2 billion, a year-over-year increase of 61%; adjusted free cash flow after equipment financing leases was $36.2 billion, an increase of $700 million year-over-year. We are pleased with our innovations, enhancements to customer experience, and achievements in 2024, and we have more plans for 2025. Let me first talk about our store business. In North America, our revenue grew 10% year-over-year, and internationally, excluding currency effects, revenue grew 9% year-over-year. We continue to focus on expanding our selection of goods, lowering prices, and enhancing convenience, driving strong unit sales growth that even outpaced our revenue growth. We continue to expand our selection of goods, offering customers choices at various price points. In 2024, we welcomed several well-known brands into our stores, including Clinique, Estée Lauder, Ora Rings, and Armani Beauty. We also continue to increase products from our selling partners, which accounted for 61% of our sold goods in 2024, the highest annual share of third-party seller units in our history. In the fourth quarter, we launched Amazon Hall for U.S. customers, providing a fun shopping experience that integrates ultra-low-priced goods into a convenient destination. It has had a very strong start. Customers consistently want Amazon to be their trusted low-price shopping place.

In the fourth quarter, consumers saved over $15 billion through our low-priced everyday items and record-breaking events during October's Prime Day, Thanksgiving Black Friday, and Cyber Monday. Additionally, according to the Federal Annual Price Study, entering the holiday shopping season, Amazon had the lowest online prices among major U.S. retailers for the eighth consecutive year, with average prices 14% lower than other leading retailers. Our delivery speed continued to accelerate in 2024, marking a record year for Prime membership. We increased the number of same-day delivery sites by over 60% in 2024, currently serving over 140 metropolitan areas. Overall, we achieved over 9 billion units delivered same-day or next-day globally. Our relentless pursuit of better selection, prices, and delivery speed has driven accelerated growth in Prime membership.

Prime members pay only $14.99 per month for unlimited free delivery on 300 million items, typically delivered same-day or next-day, exclusive shopping events like Prime Day, access to a vast array of premium shows and live sports events on Prime Video, ad-free listening to 100 million songs and podcasts through Amazon Music, unlimited generic prescriptions for just $5 a month, and unlimited grocery delivery on orders over $35 from Whole Foods Market and Amazon Fresh For $9.99 a month, you get free Grubhub+ membership with unlimited delivery, along with our latest benefit—enjoying a $0.10 fuel discount per gallon at BP, Ameco, and AMPM gas stations. When viewed as a whole and compared to many other membership services that offer only one benefit, such as video, Prime is a streaming package that will bring more benefits to Prime members in 2025.

We are also focused on the unit service cost of our logistics network, which has become an important driver of our operating profit growth. We have talked about the regionalization of the U.S. network. We recently launched a redesigned U.S. inbound network. Although still in the early stages, our inbound work has improved our inventory placement, bringing more products closer to end customers. Before Black Friday in November, we increased the percentage of available order units in ideal buildings by over 40%. We have also spent a significant amount of time optimizing the number of packages we send to customers, which reduces packaging, is more convenient for customers, and costs us less. As we build and optimize the last-mile network, our unit shipping costs continue to decline. Overall, we have reduced global unit service costs for the second consecutive year while improving speed, enhancing safety, and increasing the range of options. Looking ahead to 2025 and beyond, we see opportunities to further optimize inventory placement, expand same-day delivery networks, and accelerate robotics and automation throughout the network to reduce costs.

In terms of advertising, we are pleased with the strong growth on a massive base, achieving $17.3 billion in revenue this quarter, a year-over-year increase of 18%. This is a $69 billion annual revenue run rate, more than double the $29 billion from four years ago. Sponsored products are the largest portion of advertising revenue and are performing well, with more growth potential ahead. We also have some newer streaming services that are starting to become significant new revenue sources. In streaming video, we completed the first year of Prime Video advertising and are satisfied with the early progress, entering this year with momentum. We have made it easier for brands to work with us on omnichannel advertising. Omnichannel advertising starts with broad coverage ads at the top of the channel to drive brand awareness, moves to the mid-channel with sponsored brands allowing companies to specify keywords and audiences to attract people to their Amazon detail pages or brand stores. At the end of the channel, sponsored products help advertisers showcase relevant product ads to customers at the point of purchase. We enable brands to easily register and deploy these services in our growing advertising. We also have unique audience features that leverage billions of customer signals in Amazon Marketing Cloud's secure clean room to provide advertisers with analytics, generate core marketing metrics, and understand how their marketing performs across different channels, along with our new multi-touch attribution model that helps advertisers understand the effectiveness of their marketing. If advertisers use streaming TV, display ads, sponsored products, and other ad types in their campaigns, multi-touch attribution will show the relative contribution of each to sales Next is AWS. In the fourth quarter, AWS grew by 19% year-over-year, with a current annual revenue run rate of $115 billion. By most standards, AWS is already a substantial business. Although we expect growth to fluctuate in the coming years due to the impacts of enterprise adoption cycles, capacity considerations, and technological advancements, we remain optimistic about the future of AWS customers and business. I have spent a lot of time thinking about developments in the coming years. While some may find it hard to imagine a world where almost every application is integrated with generative artificial intelligence, where reasoning becomes a core building block like computing, storage, and databases, and most companies have their own agents to perform various tasks and interact with each other. We have been contemplating such a world, and we continue to believe that this world will primarily be built on the cloud, with most of it being built on AWS. To best help customers achieve this future, you need to have strong capabilities across all three layers of the stack. At the bottom layer, for those building models, unique and attractive chips are needed. Chips are the key elements driving training and inference computations. Most AI computations are powered by NVIDIA chips, and we clearly have a deep partnership with NVIDIA, which will remain so for a long time to come.

However, there are currently not many large-scale generative artificial intelligence applications. When you reach a scale like ours, applications such as Alexa and Rufus, costs can escalate quickly. Customers want better cost-performance ratios, which is why we are developing our own custom AI chips. Tranium 2 was just launched at the AWS re:Invent conference in December, and EC2 instances equipped with these chips typically offer 30% to 40% better cost-performance compared to other existing GPU-based instances. The advantages under this large-scale application are very significant. Some technically capable companies, such as Adobe, Databricks, Poolside, and Qualcomm, have achieved impressive results in early testing of Tranium 2. This is also why Anthropic chose to build their future cutting-edge models on Tranium 2. We are collaborating with Anthropic to develop Project Raynier, a cluster composed of Tranium 2 super servers containing hundreds of thousands of training chips. The scale of this cluster will be five times that of the cluster Anthropic uses to train its current leading cloud models. We are already developing Tranium 3 and plan to preview it by the end of 2025, followed by defining Tranium 4. Developing high-performance chips that offer leading cost-performance has become one of AWS's core advantages, starting with the Nitro and Graviton chips in our core business, now expanding into the AI domain with Tranium, which is AWS's uniqueness compared to other competing cloud providers. For model builders, another key component is the ability to simplify model building services. I won't elaborate too much on Amazon's SageMaker AI in these comments, as it has become the preferred service for AI model builders to manage AI data, build models, experiment, and deploy models. I just want to mention that SageMaker's Hyperpod feature can automatically allocate training workloads across multiple AI accelerators, preventing interruptions and saving up to 40% of training time by regularly saving checkpoints and automatically recovering failed instances from the last saved checkpoint. This feature continues to be our differentiated advantage, gaining several new compelling capabilities at re:Invent, including the ability to manage costs at the cluster level and prioritize workload allocation when budgets are reached. It is increasingly being adopted by model builders.

In the middle layer, for those looking to leverage cutting-edge models to build generative AI applications, Amazon Bedrock is our fully managed service that offers the broadest selection of high-performance foundational models, along with the most appealing feature set that makes building high-quality generative AI applications simple. We continue to iterate rapidly on Bedrock, announcing Luma AI, Poolside, and over 100 other popular emerging models joining Bedrock at re:Invent. We also quickly added the DeepSeek-R1 model to Bedrock and SageMaker. Additionally, we launched several compelling new Bedrock features at re:Invent, including prompt caching, intelligent prompt routing, and model distillation, all of which help customers achieve lower costs and latency in inference. Like SageMaker AI, Bedrock is growing rapidly and is well-received by customers.

Relatedly, we have also just launched Amazon's own cutting-edge model family, Nova, which is comparable in intelligence to the world's leading models but offers lower latency and lower prices, approximately 75% lower than other models in Bedrock, and integrates key Bedrock features such as fine-tuning, model distillation, retrieval-augmented generation (RAG), and agent capabilities. Thousands of AWS customers have already begun to leverage the capabilities and cost-effectiveness of Amazon Nova models, including Palantir, Deloitte, SAP, Dentsu, Fortinet, Trellix, and Robinhood. We are just getting started.

At the top of the stack, Amazon Q is the most powerful generative AI-driven assistant designed to support software development and leverage your own data. You may recall that in the last conference call, I shared a very practical use case where Q Transform helped the Amazon team save $260 million and 4,500 developers by migrating over 30,000 applications to the new Java JDK version. This is real value, and the company is asking for more, which we are meeting through the recently delivered Q Transform, capable of migrating Windows .NET applications to Linux, migrating from VMware to EC2, and accelerating mainframe migrations Early customer testing indicates that Q Transform will transform a large-scale mainframe migration from years of effort into a project spanning just a few quarters, reducing the time for mainframe migration by over 50%. This is a significant achievement, and these transformations are great examples of practical AI.

While AI continues to be an exciting new force in business, we have not lost focus on modernizing core company technology infrastructure from on-premises to the cloud. We have signed new AWS agreements with several companies, including Intuit, PayPal, Norwegian Cruise Line Holdings, Northrop Grumman, Prudential Financial, Reddit, Japan Airlines, Baker Hughes, Hertz, Redfin, Chime Financial, and Asana. The consistent customer feedback we received at the recent AWS re:Invent event is that we are still rapidly innovating in non-AI critical infrastructure areas such as storage, computing, databases, and analytics. Our leadership in capabilities continues to expand, with several key releases that have excited customers, including Amazon Aurora DSQL, our new serverless distributed SQL database that provides applications with the highest availability, strong consistency, PostgreSQL compatibility, and is four times faster than other popular distributed SQL databases; Amazon S3 Tables, making S3 the first cloud object storage with fully managed support for Apache Iceberg analytics capabilities; Amazon S3 Metadata, which automatically generates queryable metadata to simplify data discovery, business analytics, and real-time inference, helping customers unlock the value of data in S3; and the next generation of Amazon SageMaker, which integrates all data, analytics services, and AI services into a single interface for easier large-scale analytics and AI operations.

As we approach the end of 2024, I want to thank our team members and partners for their significant contributions over the past year. This has been a very successful year by any measure. We are far from done, and we look forward to delivering more results for our customers in 2025. Next, I will hand the meeting over to Brian for a financial update.

Brian T. Olsavsky, Chief Financial Officer: Thank you, Andy. First, let's take a look at our overall financial performance. Global revenue was $187.8 billion, an 11% year-over-year increase, excluding foreign exchange impacts. This equates to approximately $900 million of headwinds this quarter due to foreign exchange factors, about $700 million higher than we anticipated in our fourth-quarter guidance. Excluding this additional foreign exchange headwind, our revenue would have exceeded the high end of our revenue guidance range. Global operating income was $21.2 billion, our largest operating income quarter ever, and $1.2 billion above the high end of our guidance range. Across all business segments, we continue to innovate for customers while operating more efficiently. In North America, fourth-quarter revenue was $115.6 billion, a 10% year-over-year increase. International revenue was $43.4 billion, a 9% year-over-year increase, excluding foreign exchange impacts. Global paid units grew 11% year-over-year as we focus on low prices, broad selection, and fast delivery, which continue to resonate with customers Next, regarding profitability, the operating profit in North America was $9.3 billion, an increase of $2.8 billion year-on-year. The operating profit margin was 8%, an increase of 190 basis points year-on-year. In the international region, the operating profit was $1.3 billion, an increase of $1.7 billion year-on-year. The operating profit margin was 3%, an increase of 400 basis points year-on-year. This marks the eighth consecutive quarter of year-on-year growth in operating profit margins for both North America and international regions. 2024 also marks our second consecutive year of reducing global unit service costs.

In the fourth quarter, we achieved strong productivity in our transportation network, thanks to improvements in inventory placement, an increase in the number of units per package, and a reduction in transportation distances. We also saw productivity improvements in our fulfillment centers. Overall, our team executed exceptionally well throughout the quarter, especially during the peak season. I want to thank them for everything they have done for our customers. Looking ahead, we have multiple opportunities to continue reducing costs through better inventory placement, which also enables us to deliver goods to customers more quickly. In the U.S., we are adjusting our inbound network and continuing to expand our same-day delivery network. Globally, we are increasing automation and robotics across the entire network. While these efforts will take time to implement and progress may not be linear, we have a solid plan to continue driving improvements in our cost structure.

Advertising remains an important contributor to profitability in North America and international regions. This quarter, we achieved strong advertising revenue growth on a continually growing basis. We will also continue to invest in areas that are important to Amazon customers in the long term, such as Alexa, healthcare, and groceries, as well as Kuiper, including the production satellites we plan to launch in the coming months. As a reminder, we are currently expensing most of the costs associated with the development of our satellite network. Once the service achieves commercial viability, including sales to customers, we will capitalize certain costs. Next is our AWS business segment, with revenue of $28.8 billion, a year-on-year increase of 19%. AWS currently has an annual revenue run rate of $115 billion. In the fourth quarter, we continued to see growth in both generative AI and non-generative AI products, as companies began to focus on new projects, migrate more workloads to the cloud, restart or accelerate existing migrations from on-premises to the cloud, and leverage the power of generative AI. Customers recognize that to fully realize the benefits of generative AI, they must migrate to the cloud. AWS reported an operating profit of $10.6 billion, an increase of $3.5 billion year-on-year. This was due to strong growth, innovations in software and infrastructure driving efficiency improvements, and our ongoing focus on cost control across the business. As we have said in the past, we expect AWS's operating profit margin to fluctuate over time, partly due to the level of ongoing investments.

**Additionally, starting in 2024, we will extend the expected lifespan of servers, which contributed approximately 200 basis points to AWS's year-on-year profit margin growth in the fourth quarter. Now, let's look at our capital investments. As a reminder, we define it as the sum of cash capital expenditures plus equipment financing leases Capital investment for the fourth quarter was $26.3 billion, and we believe this run rate will reasonably represent our capital investment rate for 2025. Similar to 2024, most of the spending will be used to support the growing demand for technology infrastructure. This is primarily related to AWS, including supporting the demand for our AI services, as well as the technology infrastructure that supports our North American and international businesses. Additionally, we are continuing to invest in our logistics and transportation network capacity to support future growth. We are investing in building more same-day delivery facilities, optimizing our inbound logistics network, and increasing our investments in robotics and automation to improve delivery speed and reduce unit service costs. These capital investments will provide long-term support for future growth.

Next is the revenue guidance for the first quarter. We expect net sales to be between $151 billion and $155.5 billion. I want to highlight two factors that are impacting our first-quarter revenue guidance. First, the estimated impact of foreign exchange changes based on current exchange rates on year-over-year growth is expected to create a headwind of approximately $2.1 billion in the first quarter, equivalent to a year-over-year decline of 150 basis points. Please note that global currency exchange rates may fluctuate during the quarter, as we saw in the fourth quarter when the dollar strengthened against most other currencies. Secondly, as a reminder, we are comparing against last year's leap year impact. The extra day last year contributed approximately $1.5 billion in additional net sales for our business in the first quarter of 2024, equivalent to a decline in year-over-year growth rate of about 120 basis points, and this impact affects all business segments.

Operating profit for the first quarter is expected to be between $14 billion and $18 billion. This guidance includes the estimated impact of adjustments to the useful lives of certain fixed assets. I will elaborate on this later, but overall, we expect these adjustments to reduce operating profit for fiscal year 2025 by approximately $400 million on the balance sheet as of December 31, 2024.

First, in the fourth quarter, we completed a useful life study on servers and networking equipment and observed an acceleration in technological development, particularly in the fields of artificial intelligence and machine learning. Therefore, we decided to shorten the useful life of some servers and networking equipment from 6 years to 5 years starting January 2025. We expect this will reduce operating profit for fiscal year 2025 by approximately $700 million. Additionally, we also prematurely retired some servers and networking equipment, resulting in approximately $920 million in accelerated depreciation and related expenses recorded in the fourth quarter of 2024, and we expect this will also reduce operating profit for fiscal year 2025 by approximately $600 million. These two adjustments to the useful lives of servers and networking equipment primarily impact our AWS business segment.

Finally, we also completed a useful life study on certain heavy equipment used in our fulfillment centers and decided to extend the useful life of this equipment from 10 years to 13 years starting January 2025. We expect this will increase operating profit for fiscal year 2025 by approximately $900 million.

As we move toward 2025, we are excited about the achievements of our team. We remain focused on creating a better experience for our customers, and we firmly believe that being customer-centric is the only reliable way to create lasting value for our shareholders Now, let's move into the Q&A session.

Q&A Session

Host: Now, we will begin the question-and-answer session. Please follow the prompts. The first question we received is from Mr. Mark Mahaney of Evercore ISI. Please go ahead.

Mark Mahaney, Analyst: Thank you. I have two questions. First, Brian, should we be considering around $100 billion for capital expenditures in 2025? Second, Andy, do you think AWS's growth is currently impacted by supply chain constraints? Do you believe these constraints are widespread across the industry, or do they have a substantial impact on AWS? Thank you very much.

Andy Jassy, President and CEO: Let me address both of those questions. I'm Andy. Regarding capital expenditures, as Brian mentioned earlier, our capital expenditures in the fourth quarter were $26.3 billion. I think this can reasonably represent the annual capital expenditure rate for 2025. Most of the capital expenditures are directed towards the AI space in AWS. The way AWS operates and the cycle of cash flow is that the faster we grow, the higher the capital expenditures because we need to procure data centers, hardware, chips, and networking equipment in advance, which need to be invested before we can monetize them.

We do not procure without seeing significant demand signals. So, when AWS expands its capital expenditures, especially in a once-in-a-lifetime business opportunity like AI, I think this is a pretty good signal for the medium to long-term development of the AWS business. In fact, I believe that investing this capital to pursue this opportunity is wise because we believe that almost all applications as we know them today will be reinvented with AI integrated, and reasoning will become a core building block just like computing, storage, and databases.

If you believe that, combined with the brand new experiences that we could only dream of before, now made possible through AI. AI is undoubtedly the biggest opportunity since cloud computing and may be the biggest technological transformation and opportunity in business since the internet. Therefore, I believe that whether it is our business, our customers, or our shareholders, they will be satisfied with our pursuit of capital opportunities and business opportunities in the AI space from the medium to long term.

We are also investing capital in our store business, primarily to continue improving delivery speed and reducing unit service costs. So, you will see us increase the number of same-day delivery facilities from current levels. You will also see us increase the number of delivery stations in rural areas so that we can deliver goods to people living in rural areas more quickly. Additionally, we will be making significant investments in robotics and automation to reduce our unit service costs and continue to improve our productivity.

So, that’s the part about capital expenditures. I think the second question you asked, Mark, is whether AWS's growth is impacted by supply chain constraints? When you have a business like ours with a multi-billion dollar annual revenue run rate in AI and it is growing at triple-digit percentages year-over-year, it’s hard to complain. However, indeed, without some capacity constraints, we might be growing even faster These limitations are mainly reflected in the following aspects: First, the speed of chip supply from our third-party partners has slowed down somewhat compared to before, and the changes in the intermediate links require some time for the hardware to truly reach the level of healthy and high-quality servers that we expect. Second, this is also reflected in the launch of our own hardware and the Trainium 2 chip. We just introduced it to the general market at re:Invent, but most of the shipments will really come in the next few quarters and the coming months. In addition, there are power constraints; I believe the world is still limited in terms of power supply, and I think if we were not constrained, we could provide more services to our customers. There are also some components in the supply chain, such as motherboards, where supply is somewhat tight for various types of servers. Therefore, I think the team has done a great job in this regard, piecing together and providing capacity to allow our customers to continue to grow. As I mentioned earlier, our current growth rate is already quite good, but I believe we could grow faster without these limitations. I expect these constraints to gradually ease in the second half of 2025. As I said, although our current growth is already quite good, I think we could grow faster.

Host: The next question comes from Mr. Eric Sheridan of Goldman Sachs. Please go ahead.

Eric Sheridan, Analyst: Thank you very much for taking my question. I will continue to ask about the question Mark raised earlier. Andy, considering some of the news from China over the past few weeks, and the long-term perspective of reducing unit cost curves through AI, I understand the comments about capital expenditures in 2025. But from your position in the industry, how do you view the trends of open source, custom chips, and other factors in terms of reducing unit cost curves through AI, or accelerating or amplifying market deployment timelines, or potentially improving capital returns? Thank you very much.

Brian T. Olsavsky, Chief Financial Officer: Yes, I want to make a few points because there are several questions involved here. First, like many others, we are impressed with what DeepSeek is doing. I think part of the reason is their training techniques, particularly the reversal of the sequence of reinforcement training, where reinforcement learning is done earlier and without human intervention. We find this interesting before supervised fine-tuning. We also think the work they are doing in inference optimization is quite interesting.

For those of us building cutting-edge models, we are all doing similar things and learning from each other. I think you will see, and we will continue to see, mutual surpassing among us, with many innovations to come in the future.

If you run a business like AWS and believe, as we do, that almost all large generative AI applications will use multiple types of models, with different customers using different models to handle different types of workloads, then you will provide as many leading-edge models as possible for customers to choose from. That’s what we are doing through our Amazon Bedrock service. That’s also why we acted quickly to ensure that DeepSeek is available in Bedrock and SageMaker faster than others, and we already have customers starting to experiment with it I think an interesting phenomenon over the past few weeks is that sometimes people assume that if you can reduce the cost of a certain technological component, in this case, we are mainly discussing inference, it will somehow lead to a reduction in overall technology spending. We have never seen this happen.

We did the same thing in the cloud computing space. When we launched AWS in 2006, we offered S3 object storage at $0.15 per GB and compute services at $0.10 per hour, and of course, these prices have significantly decreased over the years. At that time, people thought that companies would spend less on infrastructure technology. But the reality is that companies spend less per unit on infrastructure, which is very useful for their business, but then they get excited about projects they previously thought were too expensive and ultimately end up spending more on technology.

I believe the same will happen in the AI space. The cost of inference will decrease significantly. What everyone has heard about DeepSeek in the past few weeks is just part of it, but everyone is working hard. I believe the cost of inference will drop significantly. This will make it easier for companies to integrate inference and generative AI into all their applications.

I think if you run a business like ours, we want to make it as easy as possible for customers to build customer experiences on top of our various infrastructure services, then the decrease in inference costs will be very beneficial for both customers and our business.

Host: The next question comes from Mr. Doug Anmuth of JP Morgan. Please go ahead.

Doug Anmuth, Analyst: Thank you for taking my question. I'll start with AWS. Brian, can you elaborate on the profitability issues with AWS? Over the past two years, profitability has fluctuated from over 20% to over 30%. Given your significant investment in generative AI, how should we think about a more normalized level of profitability? Additionally, in terms of the store business, can you discuss the impact of reduced shipping volumes through UPS (United Parcel Service) in the future? Can you manage the additional shipping volume required? Thank you.

Brian T. Olsavsky, Chief Financial Officer: Sure, Doug. Thank you for your question. First, regarding AWS, yes, we have indeed seen significant fluctuations in AWS's operating profit margins. We have previously said that these margins will be volatile over time. Currently, AI is still in its early stages. It initially does come with lower margins and a heavy investment burden, as we have discussed before. In the short term, this may put some pressure on margins.

But in the long term, we believe margins will be comparable to non-AI businesses. Therefore, we are very pleased with the strong growth of the AWS team, their focus on improving efficiency across all data centers, saving power, reusing power in new generative AI applications, and overall cost reduction. So, we are very satisfied with the performance of the AWS team and look forward to a strong performance in 2025 Andy Jassy, President and CEO: I'll take the question about UPS. UPS has been our partner for many years, and we expect to continue that partnership for many more years to come. As you know, over the past few years, especially during the pandemic, we have been shipping more and more packages through our own logistics network, and that trend has accelerated. Part of the reason we did this is that we needed to scale quickly during the pandemic when everything was shut down, and we needed to serve a larger share of the retail market during that time, and we needed to do it with a low-cost structure because our customers expect low prices, which is the nature of our business.

I think UPS may believe that the margins for serving Amazon are lower. Therefore, they chose to walk away from some of the volume they could have had in the partnership. We are able to handle that volume through our own logistics capabilities, and we will see how that partnership continues to evolve.

Host: The next question comes from Mr. Brian Nowak of Morgan Stanley. Please go ahead.

Brian Nowak, Analyst: Thank you for taking my question. Andy, could you elaborate on the accelerated application of robotics technology that you mentioned? Can you share any new data or lessons learned from Shreveport? How should we think about the scalability of robotics technology, and when can we expect to see it have a material impact on profitability?

Then, from a more macro perspective, regarding generative AI and GPU-driven changes, could you provide more examples of how you see the Amazon retail shopping experience changing by 2025 through better utilization of generative AI or GPU-driven machine learning?

Andy Jassy, President and CEO: Sure. Regarding robotics technology, I want to say that we have been integrating robotics technology at scale into our fulfillment network for many years now, and we have seen cost savings, productivity improvements, and safety enhancements. So, we have already gained significant value from robotics technology.

Recently, I think when you mentioned Shreveport, what we are seeing is that the next batch of robotics technology projects has started to go live, and we have integrated them all into the experience at the Shreveport facility for the first time. We are very encouraged by the speed improvements, productivity enhancements, and reductions in unit service costs that we are seeing there. It is still in the early stages, and these technologies are currently only applied in Shreveport, but we plan to start expanding and rolling them out to other facilities in the network, some of which will be our new facilities, while others will retrofit existing facilities to use these same robotics technologies. I also want to tell you that this group of about six new robotics technology projects is far from reaching what we believe is the potential for improving productivity, unit service costs, and safety in the fulfillment network through robotics technology. We are working on the next wave of innovations, but I think this will be a multi-year effort as we continue to adjust different parts of the fulfillment network to leverage robotics technology. In fact, we don't think there are many things we can't improve the experience with through robotics technology Regarding your other question, I think it actually relates to how we can better leverage AI in other areas of the business, especially in AWS. Perhaps you are asking about our retail business. I think it can be thought of this way: currently, we and other companies using AWS are primarily deriving value from AI in two macro ways.

The first macro category revolves around productivity and cost savings. In many ways, this is the low-hanging fruit in AI, and you can see it everywhere in our retail business. For example, if you look at customer service and the chatbots we’ve built, we have completely restructured it using generative AI. It has a fairly high customer satisfaction rate, and the new generative AI chatbot has improved customer satisfaction by 500 basis points. If you look at our millions of third-party seller partners, one of their biggest pain points is that because we place a high value on organizing our marketplace to make it easy to find products, when you create a new product detail page, you need to fill out many different fields. However, we have built a generative AI application for them where they can just fill in a few lines of text, or take a picture, or point to a URL, and the generative AI application will automatically fill in most of the other information, speeding up the listing of products on the site and making the sellers' jobs easier.

If you look at how we manage inventory, trying to understand what inventory we need to place at what time and in what facilities, the generative AI application we’ve built has improved our forecasting accuracy by 10%, and regional forecasting accuracy by 20%. In our robotics, many of the "brains" of the robots we just talked about are integrated with generative AI, which can tell the robotic arms what is in the boxes, what to pick up, how to move it, and where to place it in another box. Therefore, generative AI is actually the "brain" of many of our robots.

So, we have many very important productivity and cost-saving initiatives in our retail business that are all using generative AI, and this is just part of what we are doing. I think another large macro category is entirely new experiences. Similarly, you can see many examples of this in our retail business, starting with Rufus (our AI shopping assistant), which continues to grow significantly, to features like Amazon Lens, where you can take a picture of a product in front of you—in the app, you can find it in the small box at the top. You can take a picture of the item in front of you, and it uses computer vision and generative AI to find the exact same product in the search results. As for sizing, we have essentially collected the catalogs of all these different clothing manufacturers and compared them with each other. So, we know which brands run larger or smaller relative to others. Therefore, when you come to buy shoes, it can recommend the size you need, and even what we did during the Thursday night football game, where we used generative AI to create some truly creative features, such as defensive alerts, predicting which player will pressure the quarterback, or defensive gaps, where we can show the audience which area of the field has a gap Therefore, we are actually using it across our entire retail business and all the businesses we are involved in. We currently have about 1,000 different generative AI applications, either already built or in the process of being built.

Host: The next question comes from Mr. John Blackledge of TD Cowen. Please go ahead.

John Blackledge, Analyst: Okay, thank you. Can you talk about the current delivery speed and how much room there is for improvement? How does it drive the growth of the essentials business? Additionally, related to this, can you elaborate on the improvements you expect to see this year in inbound logistics network efficiency, as you try to continue to lower unit service costs? Thank you.

Andy Jassy, President and CEO: Yes, what I would say is that we are measuring delivery speed very carefully. We measure the conversion rates of people who see faster delivery promises on product detail pages compared to those who see slower delivery promises, as well as the downstream situation after customers make purchases and what they ultimately buy over the course of a year. So far, we have not seen diminishing returns from continuing to improve delivery speed. That does not mean there won't be situations where some people are willing to receive their goods later. We have a program where if people want to consolidate multiple packages and have them delivered on a certain day of the week for more sustainable and environmentally friendly delivery, they can choose to do so. We have many customers opting for this option, but time and again, we see that when we can deliver goods to their homes or wherever they are more quickly, people purchase from us more frequently, and this actually allows us to cater more to their everyday purchases.

I think if you look at the Prime Air project we are working on, the promise is that for many items, we will be able to deliver products to customers within an hour. I think when you order essentials, if you need to get items faster, that is a big deal. You can see that this has a significant impact on our essentials business. It has also had a big impact on our pharmacy business in many cities in the U.S., where people can now receive items the same day, and they are using our services more frequently than before.

Regarding inbound logistics network efficiency, I would say that we have been undergoing significant restructuring of our inbound logistics network for most of the past year and just launched it a few months ago. Similarly, we find that when we make such significant structural changes, you typically gain some easy-to-achieve efficiency improvements early on, but once it is actually running in a truly massive network, you need to make various adjustments and optimizations. We have various methods here, and I think we are still in the early stages, and I believe we will gain additional efficiency improvements throughout the year. But I expect that we will have the opportunity to continue to lower unit service costs this year, which will be an important part of that.

Host: Our final question comes from Mr. Michael Morton of MoffettNathanson. Please go ahead Michael Morton, Analyst: Hello, thank you very much for taking my question. I would like to follow up on Andy's comments about using AI in the Amazon e-commerce experience. However, I would like to discuss the other side of the coin, which is the actual discovery process that leads people to Amazon's e-commerce. Many companies are launching agent programs and assistants, and I am very interested to hear Amazon's plans regarding the potential disruption of this channel, as well as the trends in e-commerce channels over the next few years. You mentioned Rufus; could you highlight it a bit more? Thank you.

Andy Jassy, President and CEO: I think that retailers themselves, as well as many other retailers, will have their own views on how to interact with agent programs. This is an emerging field. Given our investments in fulfillment networks, websites, and the vast selection and organization of products we have built, most retailers will have their own ways to interact with agent programs, and we are no exception. I think Rufus stands out in this regard. If you observe its impact on customer experience and actually use it every month, you will find that it is getting better and better. If you are purchasing something and you are on the product detail page, our product detail page provides so much information that sometimes it can be difficult to scroll through and find that one piece of information quickly. Therefore, we now have many customers using Rufus to help them quickly find facts about products. They also use Rufus to summarize customer reviews, so they don’t have to read 100 customer reviews to understand what people really think about the product. If you look at the personalization features, especially now, you can go into Rufus and ask about the status of your order, or what I just ordered, or can you pull up this item I ordered two months ago? The personalization features are getting better and better. Therefore, we expect that by 2025, the situations where you are unsure of what to buy and need Rufus's help will continue to increase, and it will be increasingly helpful to customers.**

Host: Thank you for joining our call today and for your questions. The recording of the meeting will be available on the investor relations website for at least three months. We appreciate your interest in Amazon and look forward to speaking with you again next quarter.

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