Amazon: How much impact does AI have, and can profit margins continue to improve?

The following is the summary of Amazon's Q4 2024 earnings call. For the earnings report interpretation, please refer to Amazon: Rapid Investment in Cloud, Is the Profit Release Period About to End?

1. Core Information Review of the Earnings Report:

2. Detailed Content of the Earnings Call

2.1 Key Information from Executives:

I) Retail Business

Price Advantage and Customer Savings: Amazon's low-price strategy saved consumers over $15 billion in the fourth quarter. Portfolio's annual price research shows that it has had the lowest online prices for eight consecutive years during the holiday shopping season, averaging 14% lower than other major U.S. retailers.

Delivery Service Optimization: In 2024, Prime member delivery speeds increased, with the number of same-day delivery sites growing by over 60%, covering more than 140 metropolitan areas, achieving over 9 billion same-day or next-day deliveries globally. The rich benefits for Prime members have accelerated membership growth, with more benefits to be launched in 2025.

Cost Control Measures: The cost of delivery network services has decreased, promoting regionalization of the U.S. network and launching U.S. inbound network services. This model is still in its early stages but has already improved inventory allocation, with the on-time arrival rate of ordered products increasing by over 40% year-on-year before Black Friday in November. We have also spent considerable time optimizing the number of items launched in a single package to reduce packaging and lower the transportation cost per item, further reducing the globally calculated service cost per unit.

(II) Advertising Business

This quarter, advertising revenue reached $17.3 billion, a year-on-year increase of 18%, with an annualized revenue of $69 billion, doubling compared to four years ago. Sponsored products are the main source of revenue and are growing well, while new streaming services have become an important new revenue source. Prime Video advertising is progressing smoothly, with a full-funnel advertising model making brand registration and ad deployment more convenient, and unique audience characteristics and analytical tools helping advertisers assess marketing effectiveness.

(III) AWS Business

Business Growth and Future Outlook: In the fourth quarter, AWS revenue grew by 19% year-on-year, with an annual revenue run rate of $115 billion. Although future growth is expected to be volatile, there is optimism about its future development, believing that much of the future world will be built on the cloud, primarily based on AWS.

Chip R&D and Advantages: Deep collaboration with NVIDIA while developing custom artificial intelligence chips. Trainium2 will be launched in December 2023, with EC2 instances equipped with this chip being 30%-40% more cost-effective, and many companies have tested it successfully, with Anthropic choosing to build cutting-edge models on Trainium2. Trainium3 and Trainium4 are under development, making chip R&D a core advantage for AWSModel Building Services: Amazon SageMaker AI is the preferred service for model builders, with its Hyperpod feature saving training time. New features were launched at the re:Invent conference and are being adopted by an increasing number of model builders.

Future Outlook: Amazon is actively advancing the research and development of the Trainium series of chips. Currently, Trainium 3 is under development, with plans for a preview by the end of 2025, followed by the development of Trainium 4. With continuous technological advancements and iterations, the Trainium series of chips is expected to achieve greater breakthroughs in performance, further enhancing cost-effectiveness. In the future, Trainium chips will not only continue to optimize training and inference computing, providing stronger support for artificial intelligence model building, but may also expand application areas, helping more industries leverage artificial intelligence technology for innovative development. As more customers adopt services based on Trainium chips, Amazon's competitiveness in the artificial intelligence chip market will continue to strengthen, consolidating AWS's leading position in the cloud computing field and promoting the development of the entire cloud computing and artificial intelligence industry.

Generative Artificial Intelligence Applications: Amazon Bedrock offers a wide range of high-performance foundational model options, continuously iterating, with multiple new models added and new features launched at the conference. The Amazon Nova model family has been introduced, boasting advantages in intelligence, latency, and pricing, with thousands of AWS customers already using it.

Software Development Assistant: Amazon Q is a powerful generative artificial intelligence-driven software development assistant that can utilize user data. QTransform helps Amazon teams save costs and workload, while also accelerating the migration of various applications. Early customer tests show that it can significantly shorten migration times.

(4) Capital Expenditure

Capital expenditure in the fourth quarter reached $26.3 billion, which can represent the expected capital input rate for 2025. Most capital expenditures are aimed at meeting the growth demands of technological infrastructure, primarily related to AWS business, including supporting AI service infrastructure and technology demands in North American and international business markets, as well as investments in delivery and transportation network capacity, such as investments in same-day delivery facilities and inbound networks, utilizing robotics and automation technology to enhance delivery speed and reduce costs. These capital investments are expected to continue to grow in the coming years.

(5) Q1 2025 Guidance

Net sales for Q1 are expected to reach between $151 billion and $155.5 billion. Two key factors will impact first-quarter revenue: first, currency fluctuations are expected to cause an impact of about $2.1 billion (approximately 150 basis points) year-on-year, with global currencies potentially fluctuating this quarter, and the dollar showing a strengthening trend; second, the extra day from last year's leap year will bring about a year-on-year impact of approximately $1.5 billion in overseas sales, with these impacts varying across different market segments. Q1 operating profit is expected to be between $14 billion and $18 billion.

Impact of Server and Product Lifespan Adjustments: The fourth quarter completed a study on the lifespan of server network equipment, 1) reducing the lifespan of servers from six years to five years, starting from January 2025, which is expected to impact operating profit for the entire year of 2025 by approximately $700 million2) At the same time, the early retirement of some servers and network equipment, along with accelerated depreciation and other expenses, is expected to result in a decrease of $600 million in operating profit for the entire year of 2025, primarily affecting the AWS business. 3) The useful life of products has increased from ten years to thirteen years, and starting from January 2025, it is expected to have an impact of approximately $900 million on operating profit for the entire year of 2025.

2.2 Q&A Analyst Q&A

Q: How should we understand the $100 billion capital expenditure for 2025? Is AWS's growth suppressed by supply constraints, is this a common phenomenon in the industry, or does it only affect AWS?

A: The capital expenditure of $26.3 billion in the fourth quarter can represent the annual level for 2025, with most of it allocated to the AWS artificial intelligence sector. AWS is growing rapidly, and high capital expenditure is due to the need to procure equipment in advance. The increase in capital expenditure is a positive signal for AWS business in the medium to long term, and artificial intelligence is a significant opportunity. AWS's growth is affected by supply chain constraints, such as slow third-party chip supply, hardware launch and production issues, power limitations, and tight motherboard supply, but the team is doing its best to ensure customer capacity, and it is expected that constraints will ease in the second half of 2025.

Q: Considering the news from China, the rise of open source, and custom chip elements, how do you view the cost curve reduction through artificial intelligence, and its impact on time to market and capital returns?

A: I am impressed with DeepSeek's work; their training technology and inference optimization are noteworthy, and there is continuous mutual learning and innovation within the industry. AWS meets customer needs by providing various cutting-edge models, such as integrating new models in Amazon Bedrock. Although inference costs are decreasing, this will not reduce total technology expenditure; rather, the reduction in unit costs will encourage the company to increase technology investment, benefiting both customers and the business.

Q: How should we view the fluctuations in AWS's profit margin over the past two years, especially in the context of significant investments in generative artificial intelligence? In the retail business, what impact will the future reduction in UPS shipping volume have, and can you manage the additional transportation?

A: The fluctuations in AWS's operating profit margin are normal; early investments in artificial intelligence are large and have low profit margins, which poses some short-term resistance to profit margins, but in the long term, it will be comparable to non-artificial intelligence businesses, and we are satisfied with AWS's growth and team performance.

Q: How does the retail business respond to the impact of reduced shipping volume from transportation partner UPS, and how do you manage changes in shipping volume?

A: UPS has been a partner for many years, and we will continue to cooperate in the future. Over the past few years, some trends have accelerated due to the pandemic, and many of the company's goods are transported through its own logistics network, including last-mile delivery. During the pandemic, the business scale expanded rapidly, and many facilities were closed, requiring low-cost services for the retail segment. UPS considered that the profit margin for providing services to Amazon was low and abandoned some business, and the company supplemented this through its own logistics capabilities, focusing on development trends in the future.

Q: How is the acceleration of robotics technology and business development, what are the cost-saving situations in the later stages, and when can we see further impacts on the company's profits?

A: Amazon has been vigorously integrating robotics technology into its delivery network and has gained significant value from it, such as cost savings, productivity improvements, and enhanced operational safetyRecently, next-generation robotic technology has been put into production and integrated into factories, improving overall operational efficiency and speed, while further optimizing production costs, although it is still in the early stages. The company plans to incorporate improvements in robotic technology and efficiency enhancements into other distribution network facilities to accelerate productivity gains. In the fulfillment network, there is still room for optimization in service costs and safety, and the company is preparing for the next step in technology, which will enhance collaboration across different parts of the distribution network, further improving efficiency and productivity.

Q: How do you view the potential changes in Amazon's retail business by 2025 due to the increased use of generative artificial intelligence and machine learning capabilities?

A: The growth of retail business powered by artificial intelligence is mainly reflected in two aspects. First, productivity improvements and cost savings, such as customer service chatbots being restructured based on generative artificial intelligence, resulting in a customer satisfaction increase of over 500 basis points compared to before; third-party sales partners can use generative artificial intelligence applications when creating product details, as well as for inventory management and adjustments, enhancing the forecasting capabilities of both buyers and sellers; many robotic technologies also utilize generative artificial intelligence to boost productivity and save costs. Second, new experiences and trials, such as Alexa growing as an AI shopping assistant, and Amazon Lens helping to search for products and determine sizes through photos, applicable across most categories; generative artificial intelligence will also be used during Thursday Night Football; over 1,000 different generative artificial intelligence applications have been used in retail and other businesses.

Q: What is the current situation regarding delivery speed, how much room for improvement is there, and what impact does it have on the essentials business? What are the expectations for inbound network efficiency in terms of cost reduction services this year?

A: Currently, improving delivery speed still has positive effects, with no diminishing returns observed, such as PrimeAir achieving one-hour delivery for certain products, significantly boosting the essentials and pharmacy businesses. The transformation of the inbound network architecture has just been launched, with some efficiency improvements already achieved in the early stages, and it is expected that service costs can continue to decrease throughout the year.

Q: How do you view the impact of agents and assistants on Amazon during the e-commerce discovery process, and what changes do you foresee in future e-commerce channels?

A: Retailers have their own considerations when interacting with agents, and Amazon is no exception. The Rufus shopping assistant can help customers quickly obtain product information, summarize reviews, and achieve personalized queries, and it is expected that its usage scenarios will continue to increase by 2025, providing greater assistance to customers.

Risk disclosure and statement of this article: Dolphin Investment Research Disclaimer and General Disclosure