Amazon’s Annual Shareholder Letter: Growth in Marketplace Business Comes from More Choices, Lower Prices, and Faster Speed

[Ebrun Original] Earlier, it was reported that Amazon CEO Andy Jassy released the 2023 annual shareholder letter. In the letter, he pointed out that not only did Amazon make meaningful progress in financial metrics in 2023—revenue, free cash flow, and other aspects all saw significant improvements, but also the overall business's customer experience continued to improve, including providing more product choices, more competitive prices, and faster delivery speed.

It is reported that in the marketplace business, Amazon already has hundreds of millions of products for consumers to choose from, with tens of millions added just last year, including some high-end brands (such as Coach, Victoria's Secret, Pit Viper, Martha Stewart, Clinique, Lanc?me, and others) that have begun selling on Amazon. With consumers being price-sensitive in the uncertain economic environment, in 2023 Amazon saved 24 billion USD for customers throughout the year through large promotional activities and vouchers, which increased nearly 70% year-on-year. As for delivery, the number of items delivered the same day or next day increased nearly 70% in 2023, exceeding 7 billion units.

Andy Jassy further explained that the record-breaking delivery speed was due to the combination of two factors: first, Amazon's regionalized FBA fulfillment network, which stores products closer to consumers; and second, the expansion of same-day delivery infrastructure.

Moreover, the construction of Amazon’s regionalized fulfillment network has also resulted in reduced service costs. In 2023, Amazon reduced the average service cost worldwide for the first time since 2018—solely in the US, the service cost per unit decreased by over 0.45 USD. This allows Amazon not only to invest in speed improvements but also to increase the selection of products and services at a lower average selling price.

"Having more choices and lower prices gives us more advantages in consumers' choices," Andy Jassy stated.

At the same time, he also indicated in the letter that in 2024 Amazon will continue to work on reducing service costs further. "We've identified several areas where we can further reduce costs while providing faster service to customers," he mentioned.

In terms of international business, Andy Jassy noted that Amazon had also made significant progress in emerging regions including India, Brazil, Australia, Mexico, the Middle East, and Africa. They will continue to expand their operations in these markets and work towards profitability. It is reported that in the fourth quarter of 2023, the Mexico site has become profitable.

Additionally, Andy Jassy emphasized Amazon's exploration in the field of AI and its next efforts. He stated, "If asked what the next growth pillar that Amazon is building is, today we would first mention Generative AI."

Here is the full text of Andy Jassy’s letter to shareholders:

Dear shareholders,

This time last year, I shared my passion and optimism for Amazon's future. Today, I have even more reasons to do so. There are many reasons, from the progress we’ve made in financial results and customer experience, to the continued innovation and the outstanding opportunities ahead.

In 2023, Amazon's total revenue increased by 12% year-on-year, from 514 billion USD to 575 billion USD. Broken down by segment, North American revenue grew by 12%, from 316 billion USD to 353 billion USD; International revenue grew by 11%, from 118 billion USD to 131 billion USD; and AWS revenue grew by 13%, from 80 billion USD to 91 billion USD.

Additionally, Amazon's operating income and free cash flow (FCF) saw significant improvement. In 2023, operating income improved by 201% year-on-year, from 12.2 billion USD (operating margin of 2.4%) to 36.9 billion USD (operating margin of 6.4%). Adjusted FCF for the past twelve months, excluding equipment finance lease, improved from a negative 12.8 billion USD in 2022 to 35.5 billion USD (an increase of 48.3 billion USD).

While our progress in financial metrics is noteworthy, we are most pleased to see the continued improvement in customer experience throughout our business.

In our marketplace business, the customer response to our constant focus on selection, price, and convenience has been very strong. We continue to maintain the widest retail selection, with hundreds of millions of products to choose from, increasing by tens of millions just last year, and several high-end brands starting to sell on Amazon (such as Coach, Victoria's Secret, Pit Viper, Martha Stewart, Clinique, Lanc?me, and Urban Decay).

In the uncertain economic environment, price sensitivity has always been crucial, and with that in mind, in the fourth quarter of 2023 we launched Prime Day, an exclusive event for Prime members, allowing them to start their holiday shopping early. This was followed by our extended Black Friday and Cyber Monday holiday shopping events, open to all customers, and became our largest ever revenue event. Over the entire year of 2023, customers saved nearly 24 billion USD through millions of transactions and vouchers, increasing by nearly 70% compared to the previous year.

We continued to improve delivery speeds, breaking numerous company records. In 2023, Amazon provided Prime members with the fastest delivery speed ever, delivering over 7 billion items the same day or the next day, including over 4 billion items in the US and over 2 billion items in Europe. In the US, this achievement is a combination of regional benefits—restructuring networks to store goods closer to customers— and the expansion of same-day delivery facilities. In 2023, the number of items delivered the same day or next day increased nearly 70%. Our rapid delivery of goods into customers' hands brings frequent use of Amazon to meet their shopping demands. This accomplishment is evident across the board, including how quickly our daily business is growing (increasing by more than 20% in the fourth quarter of 2023).

Our regionalization efforts have also reduced shipping distances, helping to lower service costs. In 2023, for the first time since 2018, we reduced the average service cost per unit worldwide. Solely in the US, the average service cost per unit decreased by over 0.45 USD. By lowering service costs, we can not only invest in speed improvements but also have more choices at lower prices. Having more choices and lower prices make us a more favorable option for consumer consideration.

Looking ahead to 2024 and beyond, we are not done with reducing service costs. We have challenged every belief tightly held in our fulfillment networks and re-evaluated every part of it. We have identified several areas where we can further reduce costs, all the while providing faster service to customers. Our Fulfillment by Amazon architecture and the resulting inventory placement are areas where we see further upward potential in 2024.

We like to see our progress in established countries, and meaningful progress in new regions (such as India, Brazil, Australia, Mexico, Middle East, Africa, etc.) as they continue to expand choices and functionalities and move towards profitability (in Q4 2023, Mexico became our latest international store region to achieve profitability). We are confident that these new regions will continue to grow and achieve profitability in the long term.

In addition to our retail business, Amazon's advertising business continues to be strong, growing from $38 billion in 2022 to $47 billion in 2023, a 24% year-over-year increase, mainly driven by our sponsored ads. We expanded this product with Sponsored TV, a self-service solution, where brands can create campaigns that appear on over 30 streaming TV services, including Amazon Freevee and Twitch, with no minimum spend. Recently, we extended our streaming TV ads by introducing ads in Prime Video, expanding our streaming TV ads where brands can reach over 200 million monthly viewers in our popular entertainment products, including blockbuster movies and TV shows, award-winning Amazon MGM originals, and live sports, such as Thursday Night Football. Streaming TV ads have grown rapidly and had a good start.

Turning to AWS, we saw significant cost optimization at the beginning of 2023, as most companies tried to save money in uncertain economic conditions. This optimization was largely due to AWS helping customers use the cloud more efficiently and taking advantage of more powerful, cost-effective AWS capabilities, such as Graviton chips (our general-purpose CPU chip, providing about 40% better price performance than other leading x86 processors), S3 Intelligent-Tiering (a storage class that uses AI to detect objects with lower access frequencies and stores them in cheaper storage tiers), and Savings Plans (offering customers a longer commitment at a lower price). This work reduced short-term revenue, but was best for customers, was well received, and should be beneficial to customers and AWS in the long term. By the end of 2023, we observed weakening of cost optimizations, accelerated new deals, longer-term commitments by customers, and growing migrations.

Last year was also an important year for AWS deliveries. We announced our next-generation general-purpose CPU chip (Graviton4), with a 30% increase in computational performance over the previous generation (Graviton3) and a 75% increase in memory bandwidth. We also announced the AWS Trainium2 chip, which will provide up to four times faster machine learning training speed for generative AI applications, with three times the memory capacity of Trainium1. We continue to expand our AWS infrastructure footprint, now offering 105 availability zones in 33 geographic regions worldwide, with six new regions on the way (Malaysia, Mexico, New Zealand, the Kingdom of Saudi Arabia, Thailand, and a second Germany region in Berlin).

In generative AI (GenAI), we added dozens of features to Amazon SageMaker, making it easier for developers to build new foundational models (FM). We invented and delivered a new service (Amazon Bedrock) that allows companies to leverage existing FMs to build GenAI applications. We introduced the most powerful coding assistant Amazon Q. Customers are excited about these capabilities, and we have seen significant progress in our GenAI offerings. ( There will be more information about how we handle GenAI and why we believe we will succeed in the letter later on.)

We have also made progress in many new business investments that could be important for customers and Amazon in the long run. Here are two examples:

We are increasingly confident that Prime Video can become an independent and profitable business. This confidence is supported by ongoing development of engaging exclusive content (such as Thursday Night Football, The Lord of the Rings, Reacher, The Boys, Citadel, Road House, etc.), Prime Video customer interaction with this content, our marketing plan growth (through our third-party channel program, as well as extensive selections of shows and movies for customer rental or purchase), and the addition of advertising support in Prime Video.

In October, we reached a significant milestone in the commercialization journey of Project Kuiper, when we launched two end-to-end prototype satellites into space and successfully verified all key systems and subsystems—the first time such a launch has been successful. Kuiper is our low Earth orbit satellite initiative, aimed at providing broadband access to 400 to 500 million households today without it (as well as serving governments and businesses seeking better connectivity and performance in more remote areas), a very large revenue opportunity for Amazon. We plan to launch our first production satellite in 2024. We still have a long way to go, but are encouraged by our progress.

Overall, 2023 was a strong year, thanks to our team's delivery to customers. These achievements represent a wealth of invention, collaboration, discipline, execution, and re-imagination at Amazon. However, I believe that every one of us at Amazon believes that we still have a long way to go in every business area before we run out of how to make customers' lives better and easier, and that every business area we invest in has considerable upside.

In the past three years' annual letters, I tried to provide shareholders with a deeper understanding of how we think about the company, the businesses we're pursuing, the opportunities ahead, and our motivation. We operate in multiple market segment areas, but what ties Amazon together is our common mission to make customers' lives better and easier every day. This applies to every group of customers we serve (consumers, sellers, brands, developers, enterprises, and creators). At our best, we are not just customer-focused, but also creative, thinking about things years in advance, learning quickly, and operating like the world's largest start-up.

We have spent a tremendous amount of energy thinking about empowering builders inside and outside the company. We describe builders as people who like to invent. They like to dissect the customer experience, evaluate what the problems are and reinvent it. Builders tend not to be satisfied until the customer experience is perfect. This doesn’t stop them from offering improvements along the way, but it drives them to knit and iterate constantly. While they are not afraid to invent from scratch, they also eagerly use high-quality, scalable, cost-effective components from others. For builders, it's about having the right tools to constantly improve the customer experience.

The best way we know to achieve this is by building primitive services. They can be thought of as discrete, foundational building blocks that builders can weave together in any combination they want. Here's how we described primitive services in an AWS vision document in 2003:

"Primitive services are essential parts of a software developer's arsenal, or the most basic building blocks. They are indivisible (if they could be functionally divided into two, they were) and they do one thing extremely well. They are meant to be used together, not as a solution in themselves. And, we are going to build them for the flexibility of the developer. We won't stuff a bunch of restrictions on primitive services to keep developers from harming themselves. Instead, we are going to optimize the developer's freedom and innovation."

Of course, the concept of primitive services can be applied not just to software development, but is especially relevant in the technology field. Primitive services have been at the core of our rapid innovations for the past 20 years.

One of the many advantages of building custom infrastructure from scratch is speed. Let me provide two counterexamples to illustrate this point. First, we had early on built a successful first-party inventory retail business on Amazon, where we purchased all products from publishers, manufacturers, and distributors, stored them in our warehouses, and handled the shipping ourselves. As time went on, we realized that by allowing third-party sellers to list their products on our high-traffic search and product detail pages, we could offer a broader selection and lower prices. We built several core retail services (such as payment, search, order processing, browsing, and product management) which made experimenting with different market concepts much simpler than it would have been without these components.

As it turned out, these core components were too intermingled and not properly delineated. We learned this lesson when working with companies like Target on our Merchant.com business. The concept was for Target.com to use Amazon's e-commerce components as the backbone of its website and customize them according to their requirements. To make this arrangement possible, we had to provide these components as detachable capabilities through application programming interfaces (APIs). This decoupling was much harder than anticipated, as we had built so many of these their services with so many interdependencies in Amazon's early growth years.

This coupling was further accentuated by a heavyweight mechanism we used called "NPI". Any new product requiring collaboration by multiple internal teams had to be reviewed by this NPI group, with each team communicating how many staff weeks their work necessitated. This bottleneck limited the accomplishments we could achieve, was quite frustrating, and motivated us to get rid of it by refactoring these e-commerce components into true original services with good documentation and stable APIs, making it possible for our builders to use each other's services without any coordination tax.

Midway through the challenges with Target and NPI, we were contemplating building a new set of infrastructure technology services that would enable Amazon to move faster and external developers to build anything they could imagine. This set of services became the widely known AWS. Based on this experience, we were convinced we should build a set of original services that anyone could combine in the way they saw fit. At the time, most tech products were very feature-rich and attempted to solve multiple jobs at once. As a result, they often didn’t do any one job well.

Our AWS original services were designed differently from the ground up. They provided critical, highly flexible, but focused functionality. For instance, our first major original service, Amazon Simple Storage Service ("S3"), launched in March 2006, aimed to provide highly secure object storage at an extremely high durability, availability, internet-scale, and very low cost, very reliably. In other words, it set the gold standard for object storage.

When we launched S3, developers were excited and a bit puzzled. It was a very useful original service, but they wondered why it was only object storage? When we launched Amazon Elastic Compute Cloud ("EC2") in August 2006 and Amazon SimpleDB in 2007, people realized that we were building a set of original infrastructure services that would allow them to build anything they could imagine faster and more economically, without managing or investing in capital for data centers or hardware in advance.

As AWS continued to unveil these building blocks with time (we now have over 240 of them for builders to use—more than any other provider), the entire industry rapidly emerged on top of AWS (e.g., Airbnb, Dropbox, Instagram, Pinterest, Stripe, etc.), reshaping the industry on AWS (e.g., streaming has Netflix, Disney+, Hulu, Max, Fox, Paramount, etc.), and even crucial government agencies turned to AWS (e.g., CIA and several other US intelligence agencies). However, a less-acknowledged beneficiary was our own consumer business. By leveraging AWS to build, they made dramatic innovations across retail, advertising, devices (e.g., Alexa and Fire TV), Prime Video and Music, Amazon Go, drones, and many other businesses. When done well, original services can rapidly accelerate builders' innovation capability.

So, how do you build the right set of original services?

Pursuing original services doesn't guarantee success. You can build many or even more combinations. However, a good compass is to select real customer problems you are trying to solve.

Our logistics original service is an instructive example. Early on at Amazon, we had built core capabilities around warehousing goods and then quickly and reliably picking, packing, and shipping them to customers. As we added third-party sellers to the marketplace, they often requested to use the same logistics capabilities. Because we had built this initial set of logistics original services, we were able to launch Fulfillment by Amazon ("FBA") in 2006, allowing sellers to use Amazon's fulfillment network to store goods, then have us pick, pack, and ship them to customers, with the added benefit that these products could be used for quick Prime delivery. This service saved sellers a great deal of time and money (usually about 70% cheaper than doing it themselves), and remains one of our most popular services.

As more and more businesses started operating their direct-to-consumer ("DTC") websites, many still desire to use our fulfillment capabilities, while also having access to our payment and identity original services to improve their own website's order conversion rates (as Prime members have shared these payment and identity information with Amazon). A few years ago, we launched Buy with Prime to meet this customer need. Prime members can quickly check out on DTC websites just like they do on Amazon and get fast Prime shipping on Buy with Prime items—resulting in an approximately 25% improved order conversion rate over their default experience.

As our store business grew significantly, our supply chain became more complex, and we had to develop a series of capabilities to provide unparalleled selection to customers in low prices and very fast delivery times. We became skilled at bringing products from other countries into the US, clearing them through customs, and then shipping them to storage facilities. Because our distribution centers didn’t have enough space to store the inventory we would need to maintain our expected inventory levels, we built a low-cost upstream warehouse devoted specifically to storage (without the complex end-customer, picking, packing, and shipping functions).

Owning these two pools of inventory led us to develop an algorithm predicting when we would run out of stock within our distribution centers and automatically replenish from these upstream warehouses. Also, over the past few years, our scale and available alternatives compelled us to build our own last-mile delivery capabilities (roughly comparable to UPS) to affordably serve the volume of consumers and sellers who wanted to use Amazon.

We have solved these customer needs through the construction of additional fulfillment original services that better serve both Amazon consumers and address the increasingly complex e-commerce activities of external sellers. For example:

For sellers needing assistance importing products, we offered a service called Global Mile, leveraging our expertise in that aspect;

To get your inventory to our fulfillment centers from across borders or anywhere in the country, we allow you to use our first-party Amazon Freight Services or our third-party freight partners through our Carrier Partner Program;

In order to store more inventory at lower cost and ensure higher inventory turnover and shorter delivery times, we have opened up our upstream Amazon Storage and Distribution facilities (and automatically replenish to our fulfillment centers, when needed) to sellers;

For those who want to manage their own transportation, we have started to allow customers to use our Last Mile Delivery network to have packages delivered directly to their end customers, a service called Amazon Shipping;

For those who want to leverage our fulfillment network as a central location to store inventory and fulfill orders to customers, no matter where they order from, we have Multi-Channel Fulfillment. These are the original services we've already opened up to sellers.

Expanding your capabilities through the original services makes sense. You can keep the original services for yourself and build attractive features and capabilities on top of them, benefiting your customers and business from rapid innovation. You can offer the original services as paid services to external customers (like we have done with AWS and our recent logistics products). Or you can combine these original services into paid external applications, like we have with FBA, Buy with Prime, or Amazon Supply Chain (a logistics service we recently launched that integrates several of our logistics original services). But it’s your choice. You are only limited by the original services you build and your imagination.

Consider our newest same-day fulfillment facilities in our retail business. They are located in America's largest metro areas (we currently have 58), hold our top 100,000 SKUs (but also encompass millions of other SKUs that can be injected from nearby fulfillment centers into these same-day facilities), and reduce the time it takes from picking customer orders to the start of the shipment to just 11 minutes. These facilities also have the lowest cost to serve in our network. This experience is highly positive for customers, and we plan to double the number of these facilities.

However, if we look at it as a core building block, can we use this capability in other ways? We have a very large and growing grocery business in organic food (through Whole Foods Market) and non-perishables (such as consumer goods, canned foods, health and beauty products, etc.). We have been working hard to build a large-scale physical store offering (Amazon Fresh) to provide a great perishables experience; but, what if we use our same-day facilities to make it easy for customers to add milk, eggs, or other perishables to any Amazon order and get them the same day? This could change how people think about their weekly grocery shopping and make perishables shopping as convenient as non-perishables shopping.

Or take something that is somewhat controversial but making substantial progress, something we believe is a highly valuable original service capability for the future—our drone delivery service (called Prime Air). Drones will ultimately allow us to get packages to customers in under an hour. It won't initially be applicable to all sizes of packages and all locations, but we believe it will become universal over time. Think about how ordering perishables would change with delivery in under an hour?

The same goes for Amazon Pharmacy. Need cough drops, Advil, antibiotics or other medications? Many of these items are already being delivered within hours by our same-day facilities, and with our broader rollout of Prime Air, this will only get shorter. Highly flexible building blocks can be combined across businesses and new combinations can unlock new possibilities for customers.

Consciously building original services takes patience. Rolling out the first suite of original services can sometimes feel random to customers (or to the public) until we reveal how these building blocks come together. I mentioned AWS and S3 as an example, but so is our grocery business.

Over the past 10 years, we have tried several health experiments across various teams—but they were not driven by our original services approach. This changed in 2022, when we applied our original services thinking to the massive global healthcare problem and opportunities. Now we have created several important building blocks to help change customers’ healthcare experiences: Acute care through Amazon Care, primary care through One Medical, and pharmacy services to buy any medication that patients may need. As we have seen increasing success, Amazon customers now ask us to help them solve a wide range of health and nutrition opportunities—opportunities that can be partially unlocked through our existing grocery building blocks, including Whole Foods Market or Amazon Fresh.

As a builder, it's hard to wait for these building blocks to be put in place rather than putting together a bunch of components to solve a specific problem. The latter can be faster, but almost always slows you down in the future.

We see this temptation in our robotics efforts in our fulfillment network. We are seeking to automate dozens of processes to enhance safety, productivity, and cost. Some of the biggest opportunities require invention in areas such as storage automation, manipulation, sorting, long-distance movement of large cages, and item recognition. Many teams would directly jump to complex solutions, integrating “enough of” these disciplines into a solution so that it works effectively but doesn't solve more problems, easily evolve with the emergence of new requirements, or be reused for many other initiatives that require many of the same components.

However, when you think about these processes in a original service way, like our robotics team does, you prioritize building blocks, choosing important initiatives that can benefit from each building block, but building a toolbox to freely (and more quickly) combine future complex requirements. Our robotics team has built original services in each of the above areas, which are critical to our next set of automation, including multi-level storage, trailer loading and unloading, large scale tray movement, and more flexible out-bound flow sorting processes (including vehicle-in). The team is also building a foundational set of AI models to better recognize products in complex environments, optimize the movement of our burgeoning fleet of robots, and better manage bottlenecks in our facilities.

Sometimes, people will ask us, "What is your next pillar? You have Marketplace, Prime, and AWS, what's next?" It's certainly a thought-provoking question. However, perhaps an even more interesting question that people don't ask is, what's the next set of original services you're building, that make breakthrough customer experiences possible? If you were to ask me today, I would first mention Generative AI.

The early public attention has mainly focused on GenAI applications with the launch of ChatGPT in 2022. However, for our “original service” mindset, there are three layers in the GenAI stack, each of which is massive, and we are deeply invested in.

The bottom layer is building the foundational models for developers and companies (FM). The main original services are training models and computation needed to generate inferences (or predictions), and software that makes building these models easier. Starting with computation, the key is the chips inside. So far, nearly all leading FMs have been trained on Nvidia chips, and we continue to offer a more extensive collection of Nvidia instances than any other provider.

Despite this, the supply has always been tight and the cost continues to be an issue, as customers expand their models and applications. Customers are demanding that we push the limits on the price performance of AI chips, just like we did with Graviton on general CPU chips. As a result, we've built custom AI training chips (named Trainium) and inference chips (named Infertia).

In 2023, we announced the second version of Trainium and Infertia chips, both of which are more price/performance efficient than the first version and other alternatives. Last fall, leading FM manufacturer Anthropic announced that they will use Trainium and Infertia to build, train, and deploy their future FM. We already have many customers using our AI chips, including Anthropic, Airbnb, Hugging Face, Qualtrics, Ricoh, and Snap.

Customers building their own FM must address several challenges to put models into production. Organizing data and fine-tuning, building scalable and efficient training infrastructure, and then deploying models at scale in low-latency, cost-effective ways is hard. That's why we built Amazon SageMaker, which is an end-to-end hosted service that's a game-changer for developers in preparing AI data, managing experiments, training models faster in SageMaker (e.g., Perplexity AI trained its models 40% faster in SageMaker), reducing inference latency (e.g., Workday reduced inference latency by 80% using SageMaker), and increasing developer productivity (e.g., NatWest shortened the time to realize the value of its AI from 12-18 months to just under seven months using SageMaker).

The middle layer is for customers who are seeking to leverage existing FMs, using their own datasets for customizations, and leveraging the security and capabilities of leading cloud providers to build GenAI applications—all as a hosted service. Amazon Bedrock invented this layer and provides customers with the easiest way to build and scale GenAI applications, with the widest selection of first-party and third-party FMs, as well as leading ease-of-use features, allowing GenAI builders to get higher quality model outputs faster.

Bedrock has gained tens of thousands of active customers in just a few months, making a very strong start. The team continues to rapidly iterate on Bedrock and recently rolled out Guardrails (to protect the questions the application will answer), Knowledge Bases (to augment generation— or RAG— and real-time querying to expand the model's knowledge base), Agents (to accomplish multi-step tasks), and Fine-Tuning (to continue teaching and improving models), all of which improve the quality of the customer's applications. We have also recently added new models from Anthropic (whose newly released Claude 3 is the best-performing large language model in the world), Meta (with Llama 2), Mistral, Stability AI, Cohere, and our own Amazon Titan series FMs.

Customers in the early stages of GenAI are learning that building production-quality GenAI applications requires meaningful iterations and comes with costs and latencies. Customers don't just need one model; they need access to a variety of models and model sizes to meet different types of applications. Customers want a service that makes this experimentation and iteration simple, which is what Bedrock does, and is also why customers are so excited. Customers using Bedrock now include ADP, Amdocs, Bridgewater Associates, Broadridge, Clariant, Dana-Farber Cancer Institute, Delta Air Lines, Druva, Genesys, Genomics England, GoDaddy, KT, Lonely Planet, LexisNexis, Netsmart, Perplexity AI, Pfizer, PGA TOUR, Ricoh, Rocket Companies, and Siemens.

At the top of this stack is the applications layer. We are building a large number of GenAI applications in every consumer business of Amazon. These range from our new AI shopping assistant Rufus, to a smarter, more powerful Alexa, to ad capabilities (via natural language prompt generation, making it easy to customize and edit high-quality images, ad copy, and video), to customer and seller productivity applications, and dozens of other applications.

We are also building several applications within AWS, including perhaps the most compelling early GenAI use case—a coding companion. We recently launched Amazon Q, which is an AWS expert, writing, debugging, testing, and deploying code, and also querying a variety of customer data repositories (e.g., internal networks, wikis, Salesforce, Amazon S3, ServiceNow, Slack, Atlassian, etc.), answering questions, summarizing data, having coherent conversations, and taking action. Q is the most capable assistant available today and is rapidly evolving.

While we are building many GenAI applications ourselves, most will ultimately be built by other companies. However, what we're building in AWS isn't just a striking application or foundational model. These AWS services, across all three layers of the stack, make up a set of primitive services that democratize the next epoch for AI and empower internal and external builders to transform almost every customer experience we know of (and invent new ones). We optimistically believe that most world-changing AI will be built on top of AWS.

(As a side note, don't underestimate the importance of security in GenAI. Customer AI models contain some of their most sensitive data. AWS and its partners offer the world's most powerful security capabilities and track records; hence, an increasing number of customers want to run their GenAI on AWS.)

Recently, someone asked me a challenging question—how does Amazon stay resilient? Though simple in wording, this question is profound because it touches at the heart of our success to date and beyond. The answer lies in our disciplined adherence to deeply-held principles: 1/ hire builders who are motivated to continuously improve and expand possibilities; 2/ address genuine customer challenges, rather than what we think might be interesting technology; 3/ building in primitive services so that we can innovate and experiment at the highest speed; 4/ not wasting time trying to fight gravity (spoiler alert: you always lose)—when we find a technology that brings a better customer experience, we embrace it; 5/ accepting and learning from failed experiments—thereby becoming more dynamic to try again and apply the new knowledge.

Today, we continue to operate in an unprecedented era of transformation, which brings extraordinary growth opportunities for the field we are engaged in. For example, although we have a nearly $500 billion consumer business, about 80% of the global retail market segment still remains in physical stores. Likewise, despite an operating revenue run rate of nearly $100 billion for our cloud computing business, over 85% of global IT spending is still done locally. These businesses will continue to shift online and to the cloud. In the media and advertising sector, content will continue to migrate from linear formats to streaming. Globally, hundreds of millions of people without adequate broadband access will gain this connectivity in the coming years.

Last but equally important, generative AI may be the biggest technological shift since the early days of cloud computing itself, perhaps since the advent of the internet. Unlike the large-scale modernization of local infrastructure to the cloud, this GenAI revolution would require work to be done from the start, built on the cloud. The number of social and business benefits that may be realized from these solutions will surprise all of us.

In Amazon's history, there has never been a moment when we feel there are so many opportunities to improve our customers' lives. We are incredibly excited about the possibilities, focusing on inventing the future, and looking forward to working together to achieve this goal.

Sincerely,

Andy Jassy

President and Chief Executive Officer of Amazon.com, Inc.