Bezos’s 2010 Shareholder Letter: What a Technology Company Looks Like

#BezosShareholderLetter2010 It’s All About The Long Term. Amazon shows off its technological arsenal. Its earlier letters revolved around the customer experience; in 2010, it also demonstrated the investment and positioning in technology that a technology giant needs. To our shareholders: Random forests, naive Bayes classifiers, RESTful services, gossip protocols, eventual consistency, database sharding, anti-entropy, the Byzantine Generals Problem, erasure coding, vector clocks… These terms come up in Amazon meetings, where you might think you had wandered into a computer science class… Open a contemporary book on software design, and you will find scarcely a pattern that we do not use at Amazon. We use high-performance transaction systems; complex rendering and object caching; workflow and queuing systems; business intelligence and data analysis; machine learning and pattern recognition; neural networks and probabilistic decision-making; and a host of other technologies. Although many of our systems draw on the latest research, that still is not enough. Our architects and engineers have to explore areas that academic research has not yet reached. Many of our problems have no standard solution, so we *happily* investigate new approaches. We implement our technology almost entirely as *services*: encapsulated modules that package the data they handle and expose only external interfaces for calling their functions. It is like building with LEGO bricks. Services do not interfere with one another, allowing each to develop independently without disrupting the overall architecture. Service-oriented architecture, or SOA, is the fundamental abstraction underlying Amazon’s technology. Thanks to a thoughtful, forward-looking group of engineers and architects, we had been working this way long before SOA became a buzzword. Amazon’s e-commerce platform combines hundreds of services that work together on everything from personalized recommendations and order processing to inventory tracking. For example, the product detail page a customer sees calls on 200 to 300 services to create a highly customized page. State management is critical to any system that will become very large. Years ago, Amazon’s requirements reached a tipping point: our systems had become too big for any technology provider at the time to handle. Our core data services stored multiple petabytes of information and had to process millions of requests per second. To meet these extraordinary demands, we developed several purpose-built alternatives for persistent storage, including our own key-value store (Chinese translator’s note: a technical database term) and single-table store. We relied heavily on fundamental principles from distributed-systems and database research and began innovating from there. Our industry-leading storage systems provide exceptional scalability (Chinese translator’s note: the capacity to store vast amounts of data) while keeping performance, reliability, and cost under tight control. To achieve that extreme scalability, we adopted a new approach to managing data updates. By relaxing synchronization requirements for updates that must be replicated widely, these systems can operate under stringent performance and availability requirements. They are designed around eventual consistency. These data-management technologies, developed by Amazon engineers, gave rise to the cloud storage and data-management services offered by AWS. Simple Storage Service, Elastic Block Store, and SimpleDB, for example, all grew out of Amazon’s technology. Other Amazon businesses face similarly complex data-processing and decision problems, including ingesting and classifying product data, forecasting demand, allocating inventory, and detecting fraud. “Rule-based systems” can work, but they are difficult to maintain and become brittle over time. In many situations, advanced machine-learning techniques classify more accurately and adapt to changing conditions. Our search engine, for example, uses data-mining and machine-learning algorithms running in the background to build models around keywords. We then use information-extraction algorithms to identify features and extract relevant content from unstructured product descriptions, helping customers find products that meet their needs precisely. To predict customer interest accurately and optimize search-result rankings, we consider many dimensions of search relevance. Amazon’s product diversity means that we must use modern regression techniques, such as trained random forests of decision trees, to flexibly combine thousands of products when ranking results. What does all this work behind the scenes produce? Accurate search results that help you find the product you want! Our technology investments might not have produced results this powerful if we had simply tucked technology into an R&D department. Instead, technology reaches every team, every process, every decision, and every business innovation. It is embedded deeply in everything we do. Take Whispersync, a Kindle service that ensures you can access your library and all your highlights, notes, and bookmarks wherever you are, on any device. The technical challenge is doing this simultaneously for millions of Kindle users, across hundreds of millions of books, hundreds of device types, and more than 100 countries—with 24/7 reliability. At Whispersync’s core is an eventually consistent replicated data store, with application-defined conflict resolution that can handle devices remaining out of sync for weeks or longer. Of course, we hide all those technical processes from Kindle users. All you need to know is that when you open your Kindle, your reading progress is already synchronized! To borrow an idea from Arthur C. Clarke, any sufficiently mature technology is like magic. Now, if our shareholders have dutifully read this far and their eyelids are growing heavy, that is all right. Here is the one thing I want you to know: these technologies were not chosen at random. They lead directly to *free cash flow*. 💰 💰 💰 💰 💰 💰 💰 💰 💰 💰 We live in an age when available bandwidth, disk space, and computing performance are growing rapidly while becoming steadily cheaper. Our teams are mastering some of the world’s most complex technologies to tackle problems at the frontier of what can now be solved. As we have discussed repeatedly, we firmly believe that shareholders’ long-term interests are closely aligned with customers’ interests. Innovation is in our DNA. Technology is the fundamental tool we use to improve the customer experience. We still have much to learn, and I hope we will keep enjoying that learning. I am proud to be part of this team. As usual, the 1997 letter follows. Our original commitment is unchanged. It is still Day 1. Jeffrey P. Bezos (Translated into Chinese by Vinchent, with blood, sweat, and tears) ——Verification note—— Source: Amazon’s 2010 shareholder letter, an official public document. The Chinese translation was reposted from Dedao App user Vinchent, as credited in the original post. Verification: Compared with the official original on SEC EDGAR; selected major figures and facts were checked and were consistent. The entire text was not compared word for word.