Amazon: New demand for cloud computing will rebound
Here is the summary of Amazon's 3Q23 conference call. For a review of the financial report, please refer to the article "Google, Meta, and others falter while Amazon becomes the backbone of the US stock market" (link: here).
I. Management Remarks
Improvement in fulfillment efficiency and regionalization progress: Earlier this year, the retail business transitioned from a single nationwide distribution network in the US to eight different regions. By optimizing the links between fulfillment centers and delivery stations, we have reduced the distance and frequency of contact in delivering goods to customers. Shorter travel distances and fewer touchpoints result in lower service costs. We will continue to strive to provide the fastest delivery speed in our 29-year history for Prime customers.
The year-over-year growth rate in the third quarter continues to remain stable. While we still see ongoing cost optimization compared to a year ago, the impact of cost optimization is diminishing as more companies shift towards deploying net-new workloads. In the uncertain economic situation of 2023, the speed at which companies fulfill orders may slow down. However, we have observed an increase in the speed and quantity of completed orders, and the strong performance of new orders signed in the past few months is encouraging. We signed several new agreements in September, which took effect in October. Although these will not be reflected in the 3Q performance, they contribute to the overall volume of third-quarter orders reported.
Generative Artificial Intelligence (AI): At the foundational level, we train large-scale language models and generate inference and predictions. Recently, Anthropic, a leading manufacturer of LLMs, selected AWS as its primary cloud provider and will use training and inference to build, train, and deploy future LLMs. As part of this collaboration, AWS and Anthropic will work together on the future development of training and inference technologies.
At the intermediate level, we offer LLM as a service. We recently launched Amazon Bedrock, which allows customers to access LLMs from third-party providers (such as Anthropic Stable AI, CoherentAI21) as well as Amazon's own LLM called Titan. Customers can use their own models to customize data for these models without leaking the data back to the LLM. We will be adding Meta's Llama2 to Bedrock, making it available through a fully managed service for the first time.
Bedrock introduces some new exciting features, including the ability to create agents that can be programmed to answer questions or automate workflows. Bedrock enables customers to achieve this agility, allowing them to quickly experiment with different types and scales of models and choose the right tools. The customer response to Bedrock has been very positive and has further improved overall availability. Bedrock is the easiest way to build and scale enterprise-level general artificial intelligence applications, and it is a true game-changer for developers and companies looking to extract value from this new technology.
At the top level, we have LLM applications. Amazon CodeWhisperer has garnered a lot of early attention and has become even more powerful with the introduction of new custom features.
Gen AI: The final discussion on AWS is our work on Gen AI. Most of the data is stored in AWS, which is a clear market leader in the cloud infrastructure space. There is a large and rapidly growing number of companies building generative AI applications in AWS, including Adidas, Booking.com, Bridgewater, Clariant, LexisNexis, Merck, Royal Philips, and United Airlines. We also see success from general AI startups like Perplexity.ai, who choose to fully utilize AWS, including running future models and training and inferring on them. The AWS team has many new features to share with customers at the upcoming AWS Reinvent conference.
In addition to AWS, we have built an early version of a broader LLM preview in Alexa. In addition to being a smarter version, Alexa also has new conversational AI capabilities, including the ability to make multiple requests at once and the ability to make more natural conversational requests without using specific phrases.
Prime Video: Prime Video remains an integral part of the Prime value and is often one of the two main drivers for customers to sign up for Prime. We also believe that Prime Video itself can be a large and profitable business, and we continue to invest in providing compelling exclusive content for Prime members, as well as offering the best choice for high-quality streaming video content anywhere through our marketplace products, including channels. Customers can subscribe to channels such as Max, Paramount plus, BT plus, and MGM plus. If customers prefer an ad-free option, we plan to charge an additional $2.99 per month for US customers.
Buy with Prime: Our e-commerce mission is to continuously improve the experience and profitability of emerging international customers, as well as our growth trajectory. Both consumers and sellers are excited about "Buy with Prime," which allows third-party sellers directly targeting consumer websites to offer the same fast payment and delivery options as on Amazon.com to Amazon Prime members. We recently announced the ability for sellers to integrate Buy with Prime with their Shopify accounts, making it easier for Shopify merchants to manage their business. Including inventory pricing and promotional activities will be automatically synchronized in one place. Through a fully automated supply chain service, Amazon can extract inventory from manufacturing factories around the world, cross-border transportation, handle customs clearance, ground transportation, store inventory, and batch manage replenishment for Amazon and other sales channels, and deliver directly to customers.
6. Others: Our healthcare team continues to make it easier for people to access healthcare. We have established x Pass for customers, which provides unlimited supply of eligible medications for only $5 per month, significantly reducing the cost for customers to obtain insulin and diabetes products. We have also partnered with California Blue Shield to launch the first of its kind model, providing more affordable pharmacy services to its 4.8 million members, offering fast and free prescription drug delivery and 24/7 access to pharmacists. Our Low Earth Orbit satellite program Kuiper aims to provide fast and affordable broadband to underserved communities around the world. In the past few weeks, we have successfully launched two prototype satellites, taking a meaningful step forward. We will use this months-long mission to test our satellite network and collect data for the satellite production plan later this year.
7. Prime Day: This quarter, we held the largest Prime Day ever, with Prime members purchasing over 375 million items globally, saving over $2.5 billion and millions of transactions in the Amazon store. In the third quarter, our delivery speed was the fastest in US history. Excluding the impact of foreign exchange, 3P sales grew by 18% YoY, primarily driven by the expansion of selection and the optional services we provide to sellers, including Amazon fulfillment and account management.
During this quarter, we hosted the Amazon Annual Seller Conference, where we introduced many new innovations and product developments for sellers, including Amazon's supply chain. We also continued to see continued growth in our advertising business, with a 25% YoY growth excluding foreign exchange, primarily driven by sponsored products as we rely on machine learning to improve the relevance of the ads we show to customers and enhance our ability to measure advertisers.
8. Improvement in operating profit: Our profitability has shown strong improvement. Operating profit margin was 4.9%, up 100 basis points MoM. Regionalization allows us to simplify the network by reducing the number of trunk transportation channels, increasing the volume of existing trunk transportation channels, and connecting more direct delivery centers with distribution stations. We are also committed to optimizing inventory layout in new regional networks.
In this quarter, we saw the benefits of a decrease in inflation, mainly in long-haul ocean and rail transportation rates, which were partially offset by rising fuel prices. In terms of international business, our operating loss this quarter was $95 million, approaching breakeven, narrowing by $2.4 billion compared to the same period last year. This improvement was mainly due to cost reduction, alleviation of inflationary pressures, and improvement in leverage in both mature and emerging international countries, as well as our continued focus on customer investment and operational efficiency improvement. 9. AWS: Customers continue to focus on driving innovation and introducing new workloads into the cloud, similar to what we shared last quarter. While optimization remains a challenge, we have seen a slowdown in the pace of new cost optimizations in AWS. This quarter, our operating profit margin was 30.3%, an improvement of approximately 600 basis points compared to the previous quarter, primarily due to the leverage improvement in our workforce costs.
Cash Flow: Over the past 12 months, our free cash flow, adjusted for finance lease, was $20.2 billion, an increase of $41.7 billion compared to the same period last year. The key driver of the improvement in free cash flow was the increase in our operating revenue, which we saw across all three segments. This improvement was driven by lower service costs, continued growth in advertising, and increased leverage of fixed costs. We also saw improvements in working capital, particularly as we improved our inventory layout and increased inventory efficiency.
Capital Investment: For the 12-month period ending September 30, these investments amounted to $50 billion, lower than the $60 billion in the same period last year. For the full year 2023, we expect capital investments to be approximately $50 billion, compared to $59 billion in 2022. We expect fulfillment and transportation capital expenditures to decrease YoY, partially offset by increased infrastructure capital expenditures to drive growth in our AWS business, including additional investments related to AI and large-scale language model work.
Q&A
Q1: Will there be further cost optimizations for AWS in the future, or will it be similar to Q3?
A1: Customer Optimization: Not all customers decide to shut down workloads. A significant part of optimization is our customers' ability to leverage the cost-effectiveness of AWS. We have seen growth in EC2 customers, with a much higher proportion using the Graviton foundation compared to before. Customers like Graviton because it offers a 40% performance improvement over other x86 processors. Many customers are transitioning from on-demand hourly rate workloads to 1-3 year commitments, which we call savings plans.
There are significant opportunities in generative AI that will bring billions of dollars in revenue to AWS in the coming years. We have completed a significant number of transactions in the past few months. Key things include the ecosystem, operational performance, security, and customer satisfaction with the products you build in generative AI. AWS will continue to see a lot of growth in the future, and I am very optimistic about it.
Q2: Are you just talking more about how the regional fulfillment network has exceeded your expectations? How does this support the confidence that the North American business will exceed profitability levels above mid-single digits? Generative AI has a lot of innovation, and you mentioned many customers running early-stage workloads. How should we consider the timing for monetization? A2: Regionalization is a huge change for our network. I believe that through our algorithms, we can obtain more localized inventory levels. We have made many changes, from fulfillment centers to sorting centers, and distribution stations can be more directly connected to fulfillment centers, allowing us to achieve shorter transportation distances than expected and deliver items to people faster.
The timing of generative AI and monetization: We have been prototyping various applications around generative AI. Bedrock is working hard to address the issues related to training, application development, and operation. One useful aspect of Bedrock is that customers are now trying so many different use cases that it is convenient to have a service that not only utilizes a large number of third-party and Amazon's large models but also has different sizes and can easily transfer workloads between them.
Q3: Regarding AWS, please share 1-2 generative AI products where you saw the earliest demand. Are there any pain points in your products that have not been fully addressed when interacting with people? What solutions and innovations do you hope to have in the AI field in the next few years? Where does warehouse robot technology stand? How do you consider its impact on driving profitability?
A3: There are many products, starting with Bedrock. Bedrock makes it easier to access applications. Machine learning and AI have been exciting for people for 25 years. Considering the affordability of capabilities, computing, and storage, as well as the tools we have built, such as SageMaker, it is much easier for developers to interact with AI on a daily basis. However, Bedrock has taken a meaningful step forward in generative AI by eliminating many difficulties in decision-making. Anthropic and Bedrock have exclusive early models and customized tools, as well as fine-tuning capabilities, giving customers more control. In the future, there will be many expansions and use cases based on Bedrock, and we will continue to improve chips and training.
Robots: Looking ahead to the next few months, the new products launched by the invention team in the field of robotics are a significant investment. For us, there have been significant changes in terms of cost productivity and safety, allowing for safer working partners. We have made substantial investments in additional robot plans, many of which will be realized in 2024 and 2025, and we believe this will further impact our performance in terms of cost, productivity, and safety.
Q4: 1) The theme of cloud optimization began in the second half of 22, and the theme of AI has only truly emerged in the past 8-9 months. In 2024, the new IT budget cycle may push optimization teams to the background, while AI teams will take the forefront with a more clear budget. What are your thoughts on this?
The advertising business has performed well. What excites you the most? Continuing to expand on Amazon's properties and potentially becoming a broader digital advertising company? A4: Cloud: Many relatively lower optimization goals will occur in 2023, with optimization decreasing over the course of several months and continuing this trend. In recent years, we have seen many companies not only focusing on new generative AI workloads, but also undergoing a significant shift from on-premises to the cloud as a result of a large number of new customer transformations. Many companies have experienced stagnation in 2023 due to more conservative spending and a cautious attitude towards uncertain economic conditions. I believe you will increasingly see both companies returning to their ongoing transformations, with many system integration partners, including ourselves, starting to see them prototype designs and begin deploying and producing general AI applications on a large scale.
Advertising: While we have seen companies being more cautious in advertising, we have performed quite well in video, partly because we have many self-operated properties and a large number of advertising and brand orders.
We are currently in the second season of football. The start has been good, with viewership six weeks ago being 25% higher than a year ago. Furthermore, we have done better in advertising compared to the first year. On the other hand, most of our resources in advertising have been spent on this, and machine learning experts are discussing these issues to ensure that sponsored results are obtained when people search for relevant content. I believe both of these are our real businesses in the advertising field, where we can intelligently integrate ads into video, audio, and groceries, and have several sponsored products for third-party websites. We have done everything for Pinterest, BuzzFeed, and Hearst Newspapers. This field is still in its early stages, but it is growing well. We are very focused on continuing to create excellent customer experiences.
Q5: International retail business: When can it reach sustainable profitability? Have enough efficiency been exhausted without sufficient scale? Are there any sufficiently large, scalable, and profitable opportunities in mature markets to offset the impact of emerging countries?
A5: We will continue to focus on price selection, convenience, and retail infrastructure, increase sellers and suppliers, and improve the cost structure of our operational network. We are working on dealing with productivity factors while supplying globally. In terms of international business, we have launched in 10 new countries in the past six years, and it took us 10 years to become profitable in the United States. We will continue to invest in high-quality projects and expand premium benefits for Ashley Company.
We will continue to build fulfillment and transportation capabilities for our customers, and I cannot say it is permanent. We have reached the breakeven point of profitability, and I believe the trend line is clear and accelerating the turnaround process in all countries.